diff options
33 files changed, 4574 insertions, 158 deletions
@@ -58,6 +58,9 @@ tests/getclient tests/getstats tests/getversion tests/lock +tests/gem_basic +tests/gem_mmap +tests/gem_readwrite tests/openclose tests/setversion tests/updatedraw diff --git a/linux-core/Makefile b/linux-core/Makefile index 3af6f370..fc32676f 100644 --- a/linux-core/Makefile +++ b/linux-core/Makefile @@ -30,6 +30,7 @@ # # make DRM_MODULES="r128 radeon" # +DRM_MODULES=i915 SHELL=/bin/sh diff --git a/linux-core/Makefile.kernel b/linux-core/Makefile.kernel index 7a477a10..82d200b3 100644 --- a/linux-core/Makefile.kernel +++ b/linux-core/Makefile.kernel @@ -12,15 +12,15 @@ drm-objs := drm_auth.o drm_bufs.o drm_context.o drm_dma.o drm_drawable.o \ drm_lock.o drm_memory.o drm_proc.o drm_stub.o drm_vm.o \ drm_sysfs.o drm_pci.o drm_agpsupport.o drm_scatter.o \ drm_memory_debug.o ati_pcigart.o drm_sman.o \ - drm_hashtab.o drm_mm.o drm_object.o drm_compat.o \ + drm_hashtab.o drm_memrange.o drm_object.o drm_compat.o \ drm_fence.o drm_ttm.o drm_bo.o drm_bo_move.o drm_bo_lock.o \ - drm_regman.o drm_vm_nopage_compat.o + drm_regman.o drm_vm_nopage_compat.o drm_gem.o tdfx-objs := tdfx_drv.o r128-objs := r128_drv.o r128_cce.o r128_state.o r128_irq.o mga-objs := mga_drv.o mga_dma.o mga_state.o mga_warp.o mga_irq.o i810-objs := i810_drv.o i810_dma.o i915-objs := i915_drv.o i915_dma.o i915_irq.o i915_mem.o i915_fence.o \ - i915_buffer.o i915_compat.o i915_execbuf.o + i915_buffer.o i915_compat.o i915_execbuf.o i915_gem.o nouveau-objs := nouveau_drv.o nouveau_state.o nouveau_fifo.o nouveau_mem.o \ nouveau_object.o nouveau_irq.o nouveau_notifier.o nouveau_swmthd.o \ nouveau_sgdma.o nouveau_dma.o nouveau_bo.o nouveau_fence.o \ diff --git a/linux-core/drm-gem.txt b/linux-core/drm-gem.txt new file mode 100644 index 00000000..5cda87f8 --- /dev/null +++ b/linux-core/drm-gem.txt @@ -0,0 +1,805 @@ + The Graphics Execution Manager + Part of the Direct Rendering Manager + ============================== + + Keith Packard <keithp@keithp.com> + Eric Anholt <eric@anholt.net> + 2008-5-9 + +Contents: + + 1. GEM Overview + 2. API overview and conventions + 3. Object Creation/Destruction + 4. Reading/writing contents + 5. Mapping objects to userspace + 6. Memory Domains + 7. Execution (Intel specific) + 8. Other misc Intel-specific functions + +1. Graphics Execution Manager Overview + +Gem is designed to manage graphics memory, control access to the graphics +device execution context and handle the essentially NUMA environment unique +to modern graphics hardware. Gem allows multiple applications to share +graphics device resources without the need to constantly reload the entire +graphics card. Data may be shared between multiple applications with gem +ensuring that the correct memory synchronization occurs. + +Graphics data can consume arbitrary amounts of memory, with 3D applications +constructing ever larger sets of textures and vertices. With graphics cards +memory space growing larger every year, and graphics APIs growing more +complex, we can no longer insist that each application save a complete copy +of their graphics state so that the card can be re-initialized from user +space at each context switch. Ensuring that graphics data remains persistent +across context switches allows applications significant new functionality +while also improving performance for existing APIs. + +Modern linux desktops include significant 3D rendering as a fundemental +component of the desktop image construction process. 2D and 3D applications +paint their content to offscreen storage and the central 'compositing +manager' constructs the final screen image from those window contents. This +means that pixel image data from these applications must move within reach +of the compositing manager and used as source operands for screen image +rendering operations. + +Gem provides simple mechanisms to manage graphics data and control execution +flow within the linux operating system. Using many existing kernel +subsystems, it does this with a modest amount of code. + +2. API Overview and Conventions + +All APIs here are defined in terms of ioctls appplied to the DRM file +descriptor. To create and manipulate objects, an application must be +'authorized' using the DRI or DRI2 protocols with the X server. To relax +that, we will need to implement some better access control mechanisms within +the hardware portion of the driver to prevent inappropriate +cross-application data access. + +Any DRM driver which does not support GEM will return -ENODEV for all of +these ioctls. Invalid object handles return -EINVAL. Invalid object names +return -ENOENT. Other errors are as documented in the specific API below. + +To avoid the need to translate ioctl contents on mixed-size systems (with +32-bit user space running on a 64-bit kernel), the ioctl data structures +contain explicitly sized objects, using 64-bits for all size and pointer +data and 32-bits for identifiers. In addition, the 64-bit objects are all +carefully aligned on 64-bit boundaries. Because of this, all pointers in the +ioctl data structures are passed as uint64_t values. Suitable casts will +be necessary. + +One significant operation which is explicitly left out of this API is object +locking. Applications are expected to perform locking of shared objects +outside of the GEM api. This kind of locking is not necessary to safely +manipulate the graphics engine, and with multiple objects interacting in +unknown ways, per-object locking would likely introduce all kinds of +lock-order issues. Punting this to the application seems like the only +sensible plan. Given that DRM already offers a global lock on the hardware, +this doesn't change the current situation. + +3. Object Creation and Destruction + +Gem provides explicit memory management primitives. System pages are +allocated when the object is created, either as the fundemental storage for +hardware where system memory is used by the graphics processor directly, or +as backing store for graphics-processor resident memory. + +Objects are referenced from user space using handles. These are, for all +intents and purposes, equivalent to file descriptors. We could simply use +file descriptors were it not for the small limit (1024) of file descriptors +available to applications, and for the fact that the X server (a rather +significant user of this API) uses 'select' and has a limited maximum file +descriptor for that operation. Given the ability to allocate more file +descriptors, and given the ability to place these 'higher' in the file +descriptor space, we'd love to simply use file descriptors. + +Objects may be published with a name so that other applications can access +them. The name remains valid as long as the object exists. Right now, our +DRI APIs use 32-bit integer names, so that's what we expose here + + A. Creation + + struct drm_gem_create { + /** + * Requested size for the object. + * + * The (page-aligned) allocated size for the object + * will be returned. + */ + uint64_t size; + /** + * Returned handle for the object. + * + * Object handles are nonzero. + */ + uint32_t handle; + uint32_t pad; + }; + + /* usage */ + create.size = 16384; + ret = ioctl (fd, DRM_IOCTL_GEM_CREATE, &create); + if (ret == 0) + return create.handle; + + Note that the size is rounded up to a page boundary, and that + the rounded-up size is returned in 'size'. No name is assigned to + this object, making it local to this process. + + If insufficient memory is availabe, -ENOMEM will be returned. + + B. Closing + + struct drm_gem_close { + /** Handle of the object to be closed. */ + uint32_t handle; + uint32_t pad; + }; + + + /* usage */ + close.handle = <handle>; + ret = ioctl (fd, DRM_IOCTL_GEM_CLOSE, &close); + + This call makes the specified handle invalid, and if no other + applications are using the object, any necessary graphics hardware + synchronization is performed and the resources used by the object + released. + + C. Naming + + struct drm_gem_flink { + /** Handle for the object being named */ + uint32_t handle; + + /** Returned global name */ + uint32_t name; + }; + + /* usage */ + flink.handle = <handle>; + ret = ioctl (fd, DRM_IOCTL_GEM_FLINK, &flink); + if (ret == 0) + return flink.name; + + Flink creates a name for the object and returns it to the + application. This name can be used by other applications to gain + access to the same object. + + D. Opening by name + + struct drm_gem_open { + /** Name of object being opened */ + uint32_t name; + + /** Returned handle for the object */ + uint32_t handle; + + /** Returned size of the object */ + uint64_t size; + }; + + /* usage */ + open.name = <name>; + ret = ioctl (fd, DRM_IOCTL_GEM_OPEN, &open); + if (ret == 0) { + *sizep = open.size; + return open.handle; + } + + Open accesses an existing object and returns a handle for it. If the + object doesn't exist, -ENOENT is returned. The size of the object is + also returned. This handle has all the same capabilities as the + handle used to create the object. In particular, the object is not + destroyed until all handles are closed. + +4. Basic read/write operations + +By default, gem objects are not mapped to the applications address space, +getting data in and out of them is done with I/O operations instead. This +allows the data to reside in otherwise unmapped pages, including pages in +video memory on an attached discrete graphics card. In addition, using +explicit I/O operations allows better control over cache contents, as +graphics devices are generally not cache coherent with the CPU, mapping +pages used for graphics into an application address space requires the use +of expensive cache flushing operations. Providing direct control over +graphics data access ensures that data are handled in the most efficient +possible fashion. + + A. Reading + + struct drm_gem_pread { + /** Handle for the object being read. */ + uint32_t handle; + uint32_t pad; + /** Offset into the object to read from */ + uint64_t offset; + /** Length of data to read */ + uint64_t size; + /** Pointer to write the data into. */ + uint64_t data_ptr; /* void * */ + }; + + This copies data into the specified object at the specified + position. Any necessary graphics device synchronization and + flushing will be done automatically. + + struct drm_gem_pwrite { + /** Handle for the object being written to. */ + uint32_t handle; + uint32_t pad; + /** Offset into the object to write to */ + uint64_t offset; + /** Length of data to write */ + uint64_t size; + /** Pointer to read the data from. */ + uint64_t data_ptr; /* void * */ + }; + + This copies data out of the specified object into the + waiting user memory. Again, device synchronization will + be handled by the kernel to ensure user space sees a + consistent view of the graphics device. + +5. Mapping objects to user space + +For most objects, reading/writing is the preferred interaction mode. +However, when the CPU is involved in rendering to cover deficiencies in +hardware support for particular operations, the CPU will want to directly +access the relevant objects. + +Because mmap is fairly heavyweight, we allow applications to retain maps to +objects persistently and then update how they're using the memory through a +separate interface. Applications which fail to use this separate interface +may exhibit unpredictable behaviour as memory consistency will not be +preserved. + + A. Mapping + + struct drm_gem_mmap { + /** Handle for the object being mapped. */ + uint32_t handle; + uint32_t pad; + /** Offset in the object to map. */ + uint64_t offset; + /** + * Length of data to map. + * + * The value will be page-aligned. + */ + uint64_t size; + /** Returned pointer the data was mapped at */ + uint64_t addr_ptr; /* void * */ + }; + + /* usage */ + mmap.handle = <handle>; + mmap.offset = <offset>; + mmap.size = <size>; + ret = ioctl (fd, DRM_IOCTL_GEM_MMAP, &mmap); + if (ret == 0) + return (void *) (uintptr_t) mmap.addr_ptr; + + + B. Unmapping + + munmap (addr, length); + + Nothing strange here, just use the normal munmap syscall. + +6. Memory Domains + +Graphics devices remain a strong bastion of non cache-coherent memory. As a +result, accessing data through one functional unit will end up loading that +cache with data which then needs to be manually synchronized when that data +is used with another functional unit. + +Tracking where data are resident is done by identifying how functional units +deal with caches. Each cache is labeled as a separate memory domain. Then, +each sequence of operations is expected to load data into various read +domains and leave data in at most one write domain. Gem tracks the read and +write memory domains of each object and performs the necessary +synchronization operations when objects move from one domain set to another. + +For example, if operation 'A' constructs an image that is immediately used +by operation 'B', then when the read domain for 'B' is not the same as the +write domain for 'A', then the write domain must be flushed, and the read +domain invalidated. If these two operations are both executed in the same +command queue, then the flush operation can go inbetween them in the same +queue, avoiding any kind of CPU-based synchronization and leaving the GPU to +do the work itself. + +6.1 Memory Domains (GPU-independent) + + * DRM_GEM_DOMAIN_CPU. + + Objects in this domain are using caches which are connected to the CPU. + Moving objects from non-CPU domains into the CPU domain can involve waiting + for the GPU to finish with operations using this object. Moving objects + from this domain to a GPU domain can involve flushing CPU caches and chipset + buffers. + +6.1 GPU-independent memory domain ioctl + +This ioctl is independent of the GPU in use. So far, no use other than +synchronizing objects to the CPU domain have been found; if that turns out +to be generally true, this ioctl may be simplified further. + + A. Explicit domain control + + struct drm_gem_set_domain { + /** Handle for the object */ + uint32_t handle; + + /** New read domains */ + uint32_t read_domains; + + /** New write domain */ + uint32_t write_domain; + }; + + /* usage */ + set_domain.handle = <handle>; + set_domain.read_domains = <read_domains>; + set_domain.write_domain = <write_domain>; + ret = ioctl (fd, DRM_IOCTL_GEM_SET_DOMAIN, &set_domain); + + When the application wants to explicitly manage memory domains for + an object, it can use this function. Usually, this is only used + when the application wants to synchronize object contents between + the GPU and CPU-based application rendering. In that case, + the <read_domains> would be set to DRM_GEM_DOMAIN_CPU, and if the + application were going to write to the object, the <write_domain> + would also be set to DRM_GEM_DOMAIN_CPU. After the call, gem + guarantees that all previous rendering operations involving this + object are complete. The application is then free to access the + object through the address returned by the mmap call. Afterwards, + when the application again uses the object through the GPU, any + necessary CPU flushing will occur and the object will be correctly + synchronized with the GPU. + + Note that this synchronization is not required for any accesses + going through the driver itself. The pread, pwrite and execbuffer + ioctls all perform the necessary domain management internally. + Explicit synchronization is only necessary when accessing the object + through the mmap'd address. + +7. Execution (Intel specific) + +Managing the command buffers is inherently chip-specific, so the core of gem +doesn't have any intrinsic functions. Rather, execution is left to the +device-specific portions of the driver. + +The Intel DRM_I915_GEM_EXECBUFFER ioctl takes a list of gem objects, all of +which are mapped to the graphics device. The last object in the list is the +command buffer. + +7.1. Relocations + +Command buffers often refer to other objects, and to allow the kernel driver +to move objects around, a sequence of relocations is associated with each +object. Device-specific relocation operations are used to place the +target-object relative value into the object. + +The Intel driver has a single relocation type: + + struct drm_i915_gem_relocation_entry { + /** + * Handle of the buffer being pointed to by this + * relocation entry. + * + * It's appealing to make this be an index into the + * mm_validate_entry list to refer to the buffer, + * but this allows the driver to create a relocation + * list for state buffers and not re-write it per + * exec using the buffer. + */ + uint32_t target_handle; + + /** + * Value to be added to the offset of the target + * buffer to make up the relocation entry. + */ + uint32_t delta; + + /** + * Offset in the buffer the relocation entry will be + * written into + */ + uint64_t offset; + + /** + * Offset value of the target buffer that the + * relocation entry was last written as. + * + * If the buffer has the same offset as last time, we + * can skip syncing and writing the relocation. This + * value is written back out by the execbuffer ioctl + * when the relocation is written. + */ + uint64_t presumed_offset; + + /** + * Target memory domains read by this operation. + */ + uint32_t read_domains; + + /* + * Target memory domains written by this operation. + * + * Note that only one domain may be written by the + * whole execbuffer operation, so that where there are + * conflicts, the application will get -EINVAL back. + */ + uint32_t write_domain; + }; + + 'target_handle', the handle to the target object. This object must + be one of the objects listed in the execbuffer request or + bad things will happen. The kernel doesn't check for this. + + 'offset' is where, in the source object, the relocation data + are written. Each relocation value is a 32-bit value consisting + of the location of the target object in the GPU memory space plus + the 'delta' value included in the relocation. + + 'presumed_offset' is where user-space believes the target object + lies in GPU memory space. If this value matches where the object + actually is, then no relocation data are written, the kernel + assumes that user space has set up data in the source object + using this presumption. This offers a fairly important optimization + as writing relocation data requires mapping of the source object + into the kernel memory space. + + 'read_domains' and 'write_domains' list the usage by the source + object of the target object. The kernel unions all of the domain + information from all relocations in the execbuffer request. No more + than one write_domain is allowed, otherwise an EINVAL error is + returned. read_domains must contain write_domain. This domain + information is used to synchronize buffer contents as described + above in the section on domains. + +7.1.1 Memory Domains (Intel specific) + +The Intel GPU has several internal caches which are not coherent and hence +require explicit synchronization. Memory domains provide the necessary data +to synchronize what is needed while leaving other cache contents intact. + + * DRM_GEM_DOMAIN_I915_RENDER. + The GPU 3D and 2D rendering operations use a unified rendering cache, so + operations doing 3D painting and 2D blts will use this domain + + * DRM_GEM_DOMAIN_I915_SAMPLER + Textures are loaded by the sampler through a separate cache, so + any texture reading will use this domain. Note that the sampler + and renderer use different caches, so moving an object from render target + to texture source will require a domain transfer. + + * DRM_GEM_DOMAIN_I915_COMMAND + The command buffer doesn't have an explicit cache (although it does + read ahead quite a bit), so this domain just indicates that the object + needs to be flushed to the GPU. + + * DRM_GEM_DOMAIN_I915_INSTRUCTION + All of the programs on Gen4 and later chips use an instruction cache to + speed program execution. It must be explicitly flushed when new programs + are written to memory by the CPU. + + * DRM_GEM_DOMAIN_I915_VERTEX + Vertex data uses two different vertex caches, but they're + both flushed with the same instruction. + +7.2 Execution object list (Intel specific) + + struct drm_i915_gem_exec_object { + /** + * User's handle for a buffer to be bound into the GTT + * for this operation. + */ + uint32_t handle; + + /** + * List of relocations to be performed on this buffer + */ + uint32_t relocation_count; + /* struct drm_i915_gem_relocation_entry *relocs */ + uint64_t relocs_ptr; + + /** + * Required alignment in graphics aperture + */ + uint64_t alignment; + + /** + * Returned value of the updated offset of the object, + * for future presumed_offset writes. + */ + uint64_t offset; + }; + + Each object involved in a particular execution operation must be + listed using one of these structures. + + 'handle' references the object. + + 'relocs_ptr' is a user-mode pointer to a array of 'relocation_count' + drm_i915_gem_relocation_entry structs (see above) that + define the relocations necessary in this buffer. Note that all + relocations must reference other exec_object structures in the same + execbuffer ioctl and that those other buffers must come earlier in + the exec_object array. In other words, the dependencies mapped by the + exec_object relocations must form a directed acyclic graph. + + 'alignment' is the byte alignment necessary for this buffer. Each + object has specific alignment requirements, as the kernel doesn't + know what each object is being used for, those requirements must be + provided by user mode. If an object is used in two different ways, + it's quite possible that the alignment requirements will differ. + + 'offset' is a return value, receiving the location of the object + during this execbuffer operation. The application should use this + as the presumed offset in future operations; if the object does not + move, then kernel need not write relocation data. + +7.3 Execbuffer ioctl (Intel specific) + + struct drm_i915_gem_execbuffer { + /** + * List of buffers to be validated with their + * relocations to be performend on them. + * + * These buffers must be listed in an order such that + * all relocations a buffer is performing refer to + * buffers that have already appeared in the validate + * list. + */ + /* struct drm_i915_gem_validate_entry *buffers */ + uint64_t buffers_ptr; + uint32_t buffer_count; + + /** + * Offset in the batchbuffer to start execution from. + */ + uint32_t batch_start_offset; + + /** + * Bytes used in batchbuffer from batch_start_offset + */ + uint32_t batch_len; + uint32_t DR1; + uint32_t DR4; + uint32_t num_cliprects; + uint64_t cliprects_ptr; /* struct drm_clip_rect *cliprects */ + }; + + + 'buffers_ptr' is a user-mode pointer to an array of 'buffer_count' + drm_i915_gem_exec_object structures which contains the complete set + of objects required for this execbuffer operation. The last entry in + this array, the 'batch buffer', is the buffer of commands which will + be linked to the ring and executed. + + 'batch_start_offset' is the byte offset within the batch buffer which + contains the first command to execute. So far, we haven't found a + reason to use anything other than '0' here, but the thought was that + some space might be allocated for additional initialization which + could be skipped in some cases. This must be a multiple of 4. + + 'batch_len' is the length, in bytes, of the data to be executed + (i.e., the amount of data after batch_start_offset). This must + be a multiple of 4. + + 'num_cliprects' and 'cliprects_ptr' reference an array of + drm_clip_rect structures that is num_cliprects long. The entire + batch buffer will be executed multiple times, once for each + rectangle in this list. If num_cliprects is 0, then no clipping + rectangle will be set. + + 'DR1' and 'DR4' are portions of the 3DSTATE_DRAWING_RECTANGLE + command which will be queued when this operation is clipped + (num_cliprects != 0). + + DR1 bit definition + 31 Fast Scissor Clip Disable (debug only). + Disables a hardware optimization that + improves performance. This should have + no visible effect, other than reducing + performance + + 30 Depth Buffer Coordinate Offset Disable. + This disables the addition of the + depth buffer offset bits which are used + to change the location of the depth buffer + relative to the front buffer. + + 27:26 X Dither Offset. Specifies the X pixel + offset to use when accessing the dither table + + 25:24 Y Dither Offset. Specifies the Y pixel + offset to use when accessing the dither + table. + + DR4 bit definition + 31:16 Drawing Rectangle Origin Y. Specifies the Y + origin of coordinates relative to the + draw buffer. + + 15:0 Drawing Rectangle Origin X. Specifies the X + origin of coordinates relative to the + draw buffer. + + As you can see, these two fields are necessary for correctly + offsetting drawing within a buffer which contains multiple surfaces. + Note that DR1 is only used on Gen3 and earlier hardware and that + newer hardware sticks the dither offset elsewhere. + +7.3.1 Detailed Execution Description + + Execution of a single batch buffer requires several preparatory + steps to make the objects visible to the graphics engine and resolve + relocations to account for their current addresses. + + A. Mapping and Relocation + + Each exec_object structure in the array is examined in turn. + + If the object is not already bound to the GTT, it is assigned a + location in the graphics address space. If no space is available in + the GTT, some other object will be evicted. This may require waiting + for previous execbuffer requests to complete before that object can + be unmapped. With the location assigned, the pages for the object + are pinned in memory using find_or_create_page and the GTT entries + updated to point at the relevant pages using drm_agp_bind_pages. + + Then the array of relocations is traversed. Each relocation record + looks up the target object and, if the presumed offset does not + match the current offset (remember that this buffer has already been + assigned an address as it must have been mapped earlier), the + relocation value is computed using the current offset. If the + object is currently in use by the graphics engine, writing the data + out must be preceeded by a delay while the object is still busy. + Once it is idle, then the page containing the relocation is mapped + by the CPU and the updated relocation data written out. + + The read_domains and write_domain entries in each relocation are + used to compute the new read_domains and write_domain values for the + target buffers. The actual execution of the domain changes must wait + until all of the exec_object entries have been evaluated as the + complete set of domain information will not be available until then. + + B. Memory Domain Resolution + + After all of the new memory domain data has been pulled out of the + relocations and computed for each object, the list of objects is + again traversed and the new memory domains compared against the + current memory domains. There are two basic operations involved here: + + * Flushing the current write domain. If the new read domains + are not equal to the current write domain, then the current + write domain must be flushed. Otherwise, reads will not see data + present in the write domain cache. In addition, any new read domains + other than the current write domain must be invalidated to ensure + that the flushed data are re-read into their caches. + + * Invaliding new read domains. Any domains which were not currently + used for this object must be invalidated as old objects which + were mapped at the same location may have stale data in the new + domain caches. + + If the CPU cache is being invalidated and some GPU cache is being + flushed, then we'll have to wait for rendering to complete so that + any pending GPU writes will be complete before we flush the GPU + cache. + + If the CPU cache is being flushed, then we use 'clflush' to get data + written from the CPU. + + Because the GPU caches cannot be partially flushed or invalidated, + we don't actually flush them during this traversal stage. Rather, we + gather the invalidate and flush bits up in the device structure. + + Once all of the object domain changes have been evaluated, then the + gathered invalidate and flush bits are examined. For any GPU flush + operations, we emit a single MI_FLUSH command that performs all of + the necessary flushes. We then look to see if the CPU cache was + flushed. If so, we use the chipset flush magic (writing to a special + page) to get the data out of the chipset and into memory. + + C. Queuing Batch Buffer to the Ring + + With all of the objects resident in graphics memory space, and all + of the caches prepared with appropriate data, the batch buffer + object can be queued to the ring. If there are clip rectangles, then + the buffer is queued once per rectangle, with suitable clipping + inserted into the ring just before the batch buffer. + + D. Creating an IRQ Cookie + + Right after the batch buffer is placed in the ring, a request to + generate an IRQ is added to the ring along with a command to write a + marker into memory. When the IRQ fires, the driver can look at the + memory location to see where in the ring the GPU has passed. This + magic cookie value is stored in each object used in this execbuffer + command; it is used whereever you saw 'wait for rendering' above in + this document. + + E. Writing back the new object offsets + + So that the application has a better idea what to use for + 'presumed_offset' values later, the current object offsets are + written back to the exec_object structures. + + +8. Other misc Intel-specific functions. + +To complete the driver, a few other functions were necessary. + +8.1 Initialization from the X server + +As the X server is currently responsible for apportioning memory between 2D +and 3D, it must tell the kernel which region of the GTT aperture is +available for 3D objects to be mapped into. + + struct drm_i915_gem_init { + /** + * Beginning offset in the GTT to be managed by the + * DRM memory manager. + */ + uint64_t gtt_start; + /** + * Ending offset in the GTT to be managed by the DRM + * memory manager. + */ + uint64_t gtt_end; + }; + /* usage */ + init.gtt_start = <gtt_start>; + init.gtt_end = <gtt_end>; + ret = ioctl (fd, DRM_IOCTL_I915_GEM_INIT, &init); + + The GTT aperture between gtt_start and gtt_end will be used to map + objects. This also tells the kernel that the ring can be used, + pulling the ring addresses from the device registers. + +8.2 Pinning objects in the GTT + +For scan-out buffers and the current shared depth and back buffers, we need +to have them always available in the GTT, at least for now. Pinning means to +lock their pages in memory along with keeping them at a fixed offset in the +graphics aperture. These operations are available only to root. + + struct drm_i915_gem_pin { + /** Handle of the buffer to be pinned. */ + uint32_t handle; + uint32_t pad; + + /** alignment required within the aperture */ + uint64_t alignment; + + /** Returned GTT offset of the buffer. */ + uint64_t offset; + }; + + /* usage */ + pin.handle = <handle>; + pin.alignment = <alignment>; + ret = ioctl (fd, DRM_IOCTL_I915_GEM_PIN, &pin); + if (ret == 0) + return pin.offset; + + Pinning an object ensures that it will not be evicted from the GTT + or moved. It will stay resident until destroyed or unpinned. + + struct drm_i915_gem_unpin { + /** Handle of the buffer to be unpinned. */ + uint32_t handle; + uint32_t pad; + }; + + /* usage */ + unpin.handle = <handle>; + ret = ioctl (fd, DRM_IOCTL_I915_GEM_UNPIN, &unpin); + + Unpinning an object makes it possible to evict this object from the + GTT. It doesn't ensure that it will be evicted, just that it may. + diff --git a/linux-core/drmP.h b/linux-core/drmP.h index 69d31e14..8246f44a 100644 --- a/linux-core/drmP.h +++ b/linux-core/drmP.h @@ -54,6 +54,7 @@ #include <linux/smp_lock.h> /* For (un)lock_kernel */ #include <linux/dma-mapping.h> #include <linux/mm.h> +#include <linux/kref.h> #include <linux/pagemap.h> #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,16) #include <linux/mutex.h> @@ -89,6 +90,10 @@ struct drm_device; struct drm_file; +#if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,24) +typedef unsigned long uintptr_t; +#endif + /* If you want the memory alloc debug functionality, change define below */ /* #define DEBUG_MEMORY */ @@ -107,7 +112,7 @@ struct drm_file; #define DRIVER_IRQ_SHARED 0x80 #define DRIVER_DMA_QUEUE 0x100 #define DRIVER_FB_DMA 0x200 - +#define DRIVER_GEM 0x400 /*@}*/ @@ -427,6 +432,11 @@ struct drm_file { struct list_head refd_objects; + /** Mapping of mm object handles to object pointers. */ + struct idr object_idr; + /** Lock for synchronization of access to object_idr. */ + spinlock_t table_lock; + struct drm_open_hash refd_object_hash[_DRM_NO_REF_TYPES]; struct file *filp; void *driver_priv; @@ -464,6 +474,11 @@ struct drm_lock_data { uint32_t kernel_waiters; uint32_t user_waiters; int idle_has_lock; + /** + * Boolean signaling that the lock is held on behalf of the + * file_priv client by the kernel in an ioctl handler. + */ + int kernel_held; }; /** @@ -539,17 +554,17 @@ struct drm_sigdata { * Generic memory manager structs */ -struct drm_mm_node { +struct drm_memrange_node { struct list_head fl_entry; struct list_head ml_entry; int free; unsigned long start; unsigned long size; - struct drm_mm *mm; + struct drm_memrange *mm; void *private; }; -struct drm_mm { +struct drm_memrange { struct list_head fl_entry; struct list_head ml_entry; }; @@ -563,7 +578,7 @@ struct drm_map_list { struct drm_hash_item hash; struct drm_map *map; /**< mapping */ uint64_t user_token; - struct drm_mm_node *file_offset_node; + struct drm_memrange_node *file_offset_node; }; typedef struct drm_map drm_local_map_t; @@ -604,6 +619,56 @@ struct drm_ati_pcigart_info { int table_size; }; +/** + * This structure defines the drm_mm memory object, which will be used by the + * DRM for its buffer objects. + */ +struct drm_gem_object { + /** Reference count of this object */ + struct kref refcount; + + /** Handle count of this object. Each handle also holds a reference */ + struct kref handlecount; + + /** Related drm device */ + struct drm_device *dev; + + /** File representing the shmem storage */ + struct file *filp; + + /** + * Size of the object, in bytes. Immutable over the object's + * lifetime. + */ + size_t size; + + /** + * Global name for this object, starts at 1. 0 means unnamed. + * Access is covered by the object_name_lock in the related drm_device + */ + int name; + + /** + * Memory domains. These monitor which caches contain read/write data + * related to the object. When transitioning from one set of domains + * to another, the driver is called to ensure that caches are suitably + * flushed and invalidated + */ + uint32_t read_domains; + uint32_t write_domain; + + /** + * While validating an exec operation, the + * new read/write domain values are computed here. + * They will be transferred to the above values + * at the point that any cache flushing occurs + */ + uint32_t pending_read_domains; + uint32_t pending_write_domain; + + void *driver_private; +}; + #include "drm_objects.h" /** @@ -705,6 +770,30 @@ struct drm_driver { void (*set_version) (struct drm_device *dev, struct drm_set_version *sv); + /** + * Driver-specific constructor for drm_gem_objects, to set up + * obj->driver_private. + * + * Returns 0 on success. + */ + int (*gem_init_object) (struct drm_gem_object *obj); + void (*gem_free_object) (struct drm_gem_object *obj); + + /** + * Driver-specific callback to set memory domains from userspace + */ + int (*gem_set_domain) (struct drm_gem_object *obj, + struct drm_file *file_priv, + uint32_t read_domains, + uint32_t write_domain); + + /** + * Driver-specific callback to flush pwrite through chipset + */ + int (*gem_flush_pwrite) (struct drm_gem_object *obj, + uint64_t offset, + uint64_t size); + struct drm_fence_driver *fence_driver; struct drm_bo_driver *bo_driver; @@ -787,7 +876,7 @@ struct drm_device { struct list_head maplist; /**< Linked list of regions */ int map_count; /**< Number of mappable regions */ struct drm_open_hash map_hash; /**< User token hash table for maps */ - struct drm_mm offset_manager; /**< User token manager */ + struct drm_memrange offset_manager; /**< User token manager */ struct drm_open_hash object_hash; /**< User token hash table for objects */ struct address_space *dev_mapping; /**< For unmap_mapping_range() */ struct page *ttm_dummy_page; @@ -885,6 +974,15 @@ struct drm_device { spinlock_t drw_lock; struct idr drw_idr; /*@} */ + + /** \name GEM information */ + /*@{ */ + spinlock_t object_name_lock; + struct idr object_name_idr; + atomic_t object_count; + uint32_t invalidate_domains; /* domains pending invalidation */ + uint32_t flush_domains; /* domains pending flush */ + /*@} */ }; #if __OS_HAS_AGP @@ -1000,6 +1098,10 @@ extern void drm_free_pages(unsigned long address, int order, int area); extern DRM_AGP_MEM *drm_alloc_agp(struct drm_device *dev, int pages, u32 type); extern int drm_free_agp(DRM_AGP_MEM * handle, int pages); extern int drm_bind_agp(DRM_AGP_MEM * handle, unsigned int start); +extern DRM_AGP_MEM *drm_agp_bind_pages(struct drm_device *dev, + struct page **pages, + unsigned long num_pages, + uint32_t gtt_offset); extern int drm_unbind_agp(DRM_AGP_MEM * handle); extern void drm_free_memctl(size_t size); @@ -1082,6 +1184,9 @@ extern int drm_lock_take(struct drm_lock_data *lock_data, unsigned int context); extern int drm_lock_free(struct drm_lock_data *lock_data, unsigned int context); extern void drm_idlelock_take(struct drm_lock_data *lock_data); extern void drm_idlelock_release(struct drm_lock_data *lock_data); +extern int drm_client_lock_take(struct drm_device *dev, + struct drm_file *file_priv); +extern void drm_client_lock_release(struct drm_device *dev); /* * These are exported to drivers so that they can implement fencing using @@ -1234,26 +1339,108 @@ extern int drm_sysfs_device_add(struct drm_minor *minor); extern void drm_sysfs_device_remove(struct drm_minor *minor); /* - * Basic memory manager support (drm_mm.c) + * Basic memory manager support (drm_memrange.c) */ -extern struct drm_mm_node * drm_mm_get_block(struct drm_mm_node * parent, unsigned long size, - unsigned alignment); -extern void drm_mm_put_block(struct drm_mm_node *cur); -extern struct drm_mm_node *drm_mm_search_free(const struct drm_mm *mm, unsigned long size, - unsigned alignment, int best_match); -extern int drm_mm_init(struct drm_mm *mm, unsigned long start, unsigned long size); -extern void drm_mm_takedown(struct drm_mm *mm); -extern int drm_mm_clean(struct drm_mm *mm); -extern unsigned long drm_mm_tail_space(struct drm_mm *mm); -extern int drm_mm_remove_space_from_tail(struct drm_mm *mm, unsigned long size); -extern int drm_mm_add_space_to_tail(struct drm_mm *mm, unsigned long size); - -static inline struct drm_mm *drm_get_mm(struct drm_mm_node *block) +extern struct drm_memrange_node *drm_memrange_get_block(struct drm_memrange_node * parent, + unsigned long size, + unsigned alignment); +extern void drm_memrange_put_block(struct drm_memrange_node *cur); +extern struct drm_memrange_node *drm_memrange_search_free(const struct drm_memrange *mm, + unsigned long size, + unsigned alignment, int best_match); +extern int drm_memrange_init(struct drm_memrange *mm, + unsigned long start, unsigned long size); +extern void drm_memrange_takedown(struct drm_memrange *mm); +extern int drm_memrange_clean(struct drm_memrange *mm); +extern unsigned long drm_memrange_tail_space(struct drm_memrange *mm); +extern int drm_memrange_remove_space_from_tail(struct drm_memrange *mm, + unsigned long size); +extern int drm_memrange_add_space_to_tail(struct drm_memrange *mm, + unsigned long size); + +static inline struct drm_memrange *drm_get_mm(struct drm_memrange_node *block) { return block->mm; } +/* Graphics Execution Manager library functions (drm_gem.c) */ +int +drm_gem_init (struct drm_device *dev); + +void +drm_gem_object_free (struct kref *kref); + +void +drm_gem_object_handle_free (struct kref *kref); + +static inline void drm_gem_object_reference(struct drm_gem_object *obj) +{ + kref_get(&obj->refcount); +} + +static inline void drm_gem_object_unreference(struct drm_gem_object *obj) +{ + if (obj == NULL) + return; + + kref_put (&obj->refcount, drm_gem_object_free); +} + +static inline void drm_gem_object_handle_reference (struct drm_gem_object *obj) +{ + drm_gem_object_reference (obj); + kref_get(&obj->handlecount); +} + +static inline void drm_gem_object_handle_unreference (struct drm_gem_object *obj) +{ + if (obj == NULL) + return; + + /* + * Must bump handle count first as this may be the last + * ref, in which case the object would disappear before we + * checked for a name + */ + kref_put (&obj->handlecount, drm_gem_object_handle_free); + drm_gem_object_unreference (obj); +} + +struct drm_gem_object * +drm_gem_object_lookup(struct drm_device *dev, struct drm_file *filp, + int handle); +int drm_gem_create_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int drm_gem_close_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int drm_gem_pread_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int drm_gem_pwrite_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int drm_gem_mmap_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int drm_gem_flink_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int drm_gem_open_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int drm_gem_set_domain_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); + +void drm_gem_open(struct drm_device *dev, struct drm_file *file_private); +void drm_gem_release(struct drm_device *dev, struct drm_file *file_private); + + +/* + * Given the new read/write domains for an object, + * compute the invalidate/flush domains for the whole device. + * + */ +int drm_gem_object_set_domain (struct drm_gem_object *object, + uint32_t read_domains, + uint32_t write_domains); + + extern void drm_core_ioremap(struct drm_map *map, struct drm_device *dev); extern void drm_core_ioremapfree(struct drm_map *map, struct drm_device *dev); diff --git a/linux-core/drm_agpsupport.c b/linux-core/drm_agpsupport.c index 0aa94a75..d6594b87 100644 --- a/linux-core/drm_agpsupport.c +++ b/linux-core/drm_agpsupport.c @@ -484,7 +484,50 @@ int drm_agp_unbind_memory(DRM_AGP_MEM * handle) return agp_unbind_memory(handle); } +/** + * Binds a collection of pages into AGP memory at the given offset, returning + * the AGP memory structure containing them. + * + * No reference is held on the pages during this time -- it is up to the + * caller to handle that. + */ +DRM_AGP_MEM * +drm_agp_bind_pages(struct drm_device *dev, + struct page **pages, + unsigned long num_pages, + uint32_t gtt_offset) +{ + DRM_AGP_MEM *mem; + int ret, i; + + DRM_DEBUG("drm_agp_populate_ttm\n"); +#if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,11) + mem = drm_agp_allocate_memory(num_pages, AGP_USER_MEMORY); +#else + mem = drm_agp_allocate_memory(dev->agp->bridge, num_pages, + AGP_USER_MEMORY); +#endif + if (mem == NULL) { + DRM_ERROR("Failed to allocate memory for %ld pages\n", + num_pages); + return NULL; + } + + for (i = 0; i < num_pages; i++) + mem->memory[i] = phys_to_gart(page_to_phys(pages[i])); + mem->page_count = num_pages; + + mem->is_flushed = TRUE; + ret = drm_agp_bind_memory(mem, gtt_offset / PAGE_SIZE); + if (ret != 0) { + DRM_ERROR("Failed to bind AGP memory: %d\n", ret); + agp_free_memory(mem); + return NULL; + } + return mem; +} +EXPORT_SYMBOL(drm_agp_bind_pages); /* * AGP ttm backend interface. diff --git a/linux-core/drm_bo.c b/linux-core/drm_bo.c index 88b2ee66..3abbb8c4 100644 --- a/linux-core/drm_bo.c +++ b/linux-core/drm_bo.c @@ -418,14 +418,14 @@ static void drm_bo_cleanup_refs(struct drm_buffer_object *bo, int remove_all) if (!bo->fence) { list_del_init(&bo->lru); if (bo->mem.mm_node) { - drm_mm_put_block(bo->mem.mm_node); + drm_memrange_put_block(bo->mem.mm_node); if (bo->pinned_node == bo->mem.mm_node) bo->pinned_node = NULL; bo->mem.mm_node = NULL; } list_del_init(&bo->pinned_lru); if (bo->pinned_node) { - drm_mm_put_block(bo->pinned_node); + drm_memrange_put_block(bo->pinned_node); bo->pinned_node = NULL; } list_del_init(&bo->ddestroy); @@ -791,7 +791,7 @@ out: mutex_lock(&dev->struct_mutex); if (evict_mem.mm_node) { if (evict_mem.mm_node != bo->pinned_node) - drm_mm_put_block(evict_mem.mm_node); + drm_memrange_put_block(evict_mem.mm_node); evict_mem.mm_node = NULL; } drm_bo_add_to_lru(bo); @@ -810,7 +810,7 @@ static int drm_bo_mem_force_space(struct drm_device *dev, struct drm_bo_mem_reg *mem, uint32_t mem_type, int no_wait) { - struct drm_mm_node *node; + struct drm_memrange_node *node; struct drm_buffer_manager *bm = &dev->bm; struct drm_buffer_object *entry; struct drm_mem_type_manager *man = &bm->man[mem_type]; @@ -820,7 +820,7 @@ static int drm_bo_mem_force_space(struct drm_device *dev, mutex_lock(&dev->struct_mutex); do { - node = drm_mm_search_free(&man->manager, num_pages, + node = drm_memrange_search_free(&man->manager, num_pages, mem->page_alignment, 1); if (node) break; @@ -846,7 +846,7 @@ static int drm_bo_mem_force_space(struct drm_device *dev, return -ENOMEM; } - node = drm_mm_get_block(node, num_pages, mem->page_alignment); + node = drm_memrange_get_block(node, num_pages, mem->page_alignment); if (unlikely(!node)) { mutex_unlock(&dev->struct_mutex); return -ENOMEM; @@ -924,7 +924,7 @@ int drm_bo_mem_space(struct drm_buffer_object *bo, int type_found = 0; int type_ok = 0; int has_eagain = 0; - struct drm_mm_node *node = NULL; + struct drm_memrange_node *node = NULL; int ret; mem->mm_node = NULL; @@ -952,10 +952,10 @@ int drm_bo_mem_space(struct drm_buffer_object *bo, mutex_lock(&dev->struct_mutex); if (man->has_type && man->use_type) { type_found = 1; - node = drm_mm_search_free(&man->manager, mem->num_pages, + node = drm_memrange_search_free(&man->manager, mem->num_pages, mem->page_alignment, 1); if (node) - node = drm_mm_get_block(node, mem->num_pages, + node = drm_memrange_get_block(node, mem->num_pages, mem->page_alignment); } mutex_unlock(&dev->struct_mutex); @@ -1340,7 +1340,7 @@ out_unlock: if (ret || !move_unfenced) { if (mem.mm_node) { if (mem.mm_node != bo->pinned_node) - drm_mm_put_block(mem.mm_node); + drm_memrange_put_block(mem.mm_node); mem.mm_node = NULL; } drm_bo_add_to_lru(bo); @@ -1432,7 +1432,7 @@ static int drm_buffer_object_validate(struct drm_buffer_object *bo, if (bo->pinned_node != bo->mem.mm_node) { if (bo->pinned_node != NULL) - drm_mm_put_block(bo->pinned_node); + drm_memrange_put_block(bo->pinned_node); bo->pinned_node = bo->mem.mm_node; } @@ -1443,7 +1443,7 @@ static int drm_buffer_object_validate(struct drm_buffer_object *bo, mutex_lock(&dev->struct_mutex); if (bo->pinned_node != bo->mem.mm_node) - drm_mm_put_block(bo->pinned_node); + drm_memrange_put_block(bo->pinned_node); list_del_init(&bo->pinned_lru); bo->pinned_node = NULL; @@ -2082,7 +2082,7 @@ static int drm_bo_leave_list(struct drm_buffer_object *bo, if (bo->pinned_node == bo->mem.mm_node) bo->pinned_node = NULL; if (bo->pinned_node != NULL) { - drm_mm_put_block(bo->pinned_node); + drm_memrange_put_block(bo->pinned_node); bo->pinned_node = NULL; } mutex_unlock(&dev->struct_mutex); @@ -2223,8 +2223,8 @@ int drm_bo_clean_mm(struct drm_device *dev, unsigned mem_type, int kern_clean) drm_bo_force_list_clean(dev, &man->lru, mem_type, 1, 0, 0); drm_bo_force_list_clean(dev, &man->pinned, mem_type, 1, 0, 1); - if (drm_mm_clean(&man->manager)) { - drm_mm_takedown(&man->manager); + if (drm_memrange_clean(&man->manager)) { + drm_memrange_takedown(&man->manager); } else { ret = -EBUSY; } @@ -2295,7 +2295,7 @@ int drm_bo_init_mm(struct drm_device *dev, unsigned type, DRM_ERROR("Zero size memory manager type %d\n", type); return ret; } - ret = drm_mm_init(&man->manager, p_offset, p_size); + ret = drm_memrange_init(&man->manager, p_offset, p_size); if (ret) return ret; } @@ -2713,7 +2713,7 @@ static void drm_bo_takedown_vm_locked(struct drm_buffer_object *bo) list->user_token = 0; } if (list->file_offset_node) { - drm_mm_put_block(list->file_offset_node); + drm_memrange_put_block(list->file_offset_node); list->file_offset_node = NULL; } @@ -2756,7 +2756,7 @@ static int drm_bo_setup_vm_locked(struct drm_buffer_object *bo) atomic_inc(&bo->usage); map->handle = (void *)bo; - list->file_offset_node = drm_mm_search_free(&dev->offset_manager, + list->file_offset_node = drm_memrange_search_free(&dev->offset_manager, bo->mem.num_pages, 0, 0); if (unlikely(!list->file_offset_node)) { @@ -2764,7 +2764,7 @@ static int drm_bo_setup_vm_locked(struct drm_buffer_object *bo) return -ENOMEM; } - list->file_offset_node = drm_mm_get_block(list->file_offset_node, + list->file_offset_node = drm_memrange_get_block(list->file_offset_node, bo->mem.num_pages, 0); if (unlikely(!list->file_offset_node)) { diff --git a/linux-core/drm_bo_move.c b/linux-core/drm_bo_move.c index bf0e1b74..850be5a3 100644 --- a/linux-core/drm_bo_move.c +++ b/linux-core/drm_bo_move.c @@ -41,7 +41,7 @@ static void drm_bo_free_old_node(struct drm_buffer_object *bo) if (old_mem->mm_node && (old_mem->mm_node != bo->pinned_node)) { mutex_lock(&bo->dev->struct_mutex); - drm_mm_put_block(old_mem->mm_node); + drm_memrange_put_block(old_mem->mm_node); mutex_unlock(&bo->dev->struct_mutex); } old_mem->mm_node = NULL; diff --git a/linux-core/drm_drv.c b/linux-core/drm_drv.c index b8b8333e..edc1f057 100644 --- a/linux-core/drm_drv.c +++ b/linux-core/drm_drv.c @@ -150,6 +150,15 @@ static struct drm_ioctl_desc drm_ioctls[] = { DRM_IOCTL_DEF(DRM_IOCTL_BO_VERSION, drm_bo_version_ioctl, 0), DRM_IOCTL_DEF(DRM_IOCTL_MM_INFO, drm_mm_info_ioctl, 0), + + DRM_IOCTL_DEF(DRM_IOCTL_GEM_CREATE, drm_gem_create_ioctl, 0), + DRM_IOCTL_DEF(DRM_IOCTL_GEM_CLOSE, drm_gem_close_ioctl, 0), + DRM_IOCTL_DEF(DRM_IOCTL_GEM_PREAD, drm_gem_pread_ioctl, 0), + DRM_IOCTL_DEF(DRM_IOCTL_GEM_PWRITE, drm_gem_pwrite_ioctl, 0), + DRM_IOCTL_DEF(DRM_IOCTL_GEM_MMAP, drm_gem_mmap_ioctl, 0), + DRM_IOCTL_DEF(DRM_IOCTL_GEM_FLINK, drm_gem_flink_ioctl, DRM_AUTH), + DRM_IOCTL_DEF(DRM_IOCTL_GEM_OPEN, drm_gem_open_ioctl, DRM_AUTH), + DRM_IOCTL_DEF(DRM_IOCTL_GEM_SET_DOMAIN, drm_gem_set_domain_ioctl, DRM_AUTH), }; #define DRM_CORE_IOCTL_COUNT ARRAY_SIZE( drm_ioctls ) @@ -415,7 +424,7 @@ static void drm_cleanup(struct drm_device * dev) drm_ctxbitmap_cleanup(dev); drm_ht_remove(&dev->map_hash); - drm_mm_takedown(&dev->offset_manager); + drm_memrange_takedown(&dev->offset_manager); drm_ht_remove(&dev->object_hash); drm_put_minor(&dev->primary); diff --git a/linux-core/drm_fops.c b/linux-core/drm_fops.c index 3bc25f24..ec521101 100644 --- a/linux-core/drm_fops.c +++ b/linux-core/drm_fops.c @@ -274,6 +274,9 @@ static int drm_open_helper(struct inode *inode, struct file *filp, goto out_free; } + if (dev->driver->driver_features & DRIVER_GEM) + drm_gem_open(dev, priv); + if (dev->driver->open) { ret = dev->driver->open(dev, priv); if (ret < 0) @@ -444,6 +447,9 @@ int drm_release(struct inode *inode, struct file *filp) dev->driver->reclaim_buffers(dev, file_priv); } + if (dev->driver->driver_features & DRIVER_GEM) + drm_gem_release(dev, file_priv); + drm_fasync(-1, filp, 0); mutex_lock(&dev->ctxlist_mutex); diff --git a/linux-core/drm_gem.c b/linux-core/drm_gem.c new file mode 100644 index 00000000..b726e598 --- /dev/null +++ b/linux-core/drm_gem.c @@ -0,0 +1,639 @@ +/* + * Copyright © 2008 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt <eric@anholt.net> + * + */ + +#include <linux/types.h> +#include <linux/slab.h> +#include <linux/mm.h> +#include <linux/uaccess.h> +#include <linux/fs.h> +#include <linux/file.h> +#include <linux/module.h> +#include <linux/mman.h> +#include <linux/pagemap.h> +#include "drmP.h" + +/** @file drm_gem.c + * + * This file provides some of the base ioctls and library routines for + * the graphics memory manager implemented by each device driver. + * + * Because various devices have different requirements in terms of + * synchronization and migration strategies, implementing that is left up to + * the driver, and all that the general API provides should be generic -- + * allocating objects, reading/writing data with the cpu, freeing objects. + * Even there, platform-dependent optimizations for reading/writing data with + * the CPU mean we'll likely hook those out to driver-specific calls. However, + * the DRI2 implementation wants to have at least allocate/mmap be generic. + * + * The goal was to have swap-backed object allocation managed through + * struct file. However, file descriptors as handles to a struct file have + * two major failings: + * - Process limits prevent more than 1024 or so being used at a time by + * default. + * - Inability to allocate high fds will aggravate the X Server's select() + * handling, and likely that of many GL client applications as well. + * + * This led to a plan of using our own integer IDs (called handles, following + * DRM terminology) to mimic fds, and implement the fd syscalls we need as + * ioctls. The objects themselves will still include the struct file so + * that we can transition to fds if the required kernel infrastructure shows + * up at a later date, and as our interface with shmfs for memory allocation. + */ + +/** + * Initialize the GEM device fields + */ + +int +drm_gem_init(struct drm_device *dev) +{ + spin_lock_init(&dev->object_name_lock); + idr_init(&dev->object_name_idr); + atomic_set(&dev->object_count, 0); + return 0; +} + +/** + * Allocate a GEM object of the specified size with shmfs backing store + */ +static struct drm_gem_object * +drm_gem_object_alloc(struct drm_device *dev, size_t size) +{ + struct drm_gem_object *obj; + + BUG_ON((size & (PAGE_SIZE - 1)) != 0); + + obj = kcalloc(1, sizeof(*obj), GFP_KERNEL); + + obj->dev = dev; + obj->filp = shmem_file_setup("drm mm object", size, 0); + if (IS_ERR(obj->filp)) { + kfree(obj); + return NULL; + } + + kref_init(&obj->refcount); + kref_init(&obj->handlecount); + obj->size = size; + + /* + * We've just allocated pages from the kernel, + * so they've just been written by the CPU with + * zeros. They'll need to be clflushed before we + * use them with the GPU. + */ + obj->write_domain = DRM_GEM_DOMAIN_CPU; + obj->read_domains = DRM_GEM_DOMAIN_CPU; + if (dev->driver->gem_init_object != NULL && + dev->driver->gem_init_object(obj) != 0) { + fput(obj->filp); + kfree(obj); + return NULL; + } + atomic_inc(&dev->object_count); + return obj; +} + +/** + * Removes the mapping from handle to filp for this object. + */ +static int +drm_gem_handle_delete(struct drm_file *filp, int handle) +{ + struct drm_device *dev; + struct drm_gem_object *obj; + + /* This is gross. The idr system doesn't let us try a delete and + * return an error code. It just spews if you fail at deleting. + * So, we have to grab a lock around finding the object and then + * doing the delete on it and dropping the refcount, or the user + * could race us to double-decrement the refcount and cause a + * use-after-free later. Given the frequency of our handle lookups, + * we may want to use ida for number allocation and a hash table + * for the pointers, anyway. + */ + spin_lock(&filp->table_lock); + + /* Check if we currently have a reference on the object */ + obj = idr_find(&filp->object_idr, handle); + if (obj == NULL) { + spin_unlock(&filp->table_lock); + return -EINVAL; + } + dev = obj->dev; + + /* Release reference and decrement refcount. */ + idr_remove(&filp->object_idr, handle); + spin_unlock(&filp->table_lock); + + mutex_lock(&dev->struct_mutex); + drm_gem_object_handle_unreference(obj); + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +/** + * Create a handle for this object. This adds a handle reference + * to the object, which includes a regular reference count. Callers + * will likely want to dereference the object afterwards. + */ +static int +drm_gem_handle_create(struct drm_file *file_priv, + struct drm_gem_object *obj, + int *handlep) +{ + int ret; + + /* + * Get the user-visible handle using idr. + */ +again: + /* ensure there is space available to allocate a handle */ + if (idr_pre_get(&file_priv->object_idr, GFP_KERNEL) == 0) + return -ENOMEM; + + /* do the allocation under our spinlock */ + spin_lock(&file_priv->table_lock); + ret = idr_get_new_above(&file_priv->object_idr, obj, 1, handlep); + spin_unlock(&file_priv->table_lock); + if (ret == -EAGAIN) + goto again; + + if (ret != 0) + return ret; + + drm_gem_object_handle_reference(obj); + return 0; +} + +/** Returns a reference to the object named by the handle. */ +struct drm_gem_object * +drm_gem_object_lookup(struct drm_device *dev, struct drm_file *filp, + int handle) +{ + struct drm_gem_object *obj; + + spin_lock(&filp->table_lock); + + /* Check if we currently have a reference on the object */ + obj = idr_find(&filp->object_idr, handle); + if (obj == NULL) { + spin_unlock(&filp->table_lock); + return NULL; + } + + drm_gem_object_reference(obj); + + spin_unlock(&filp->table_lock); + + return obj; +} +EXPORT_SYMBOL(drm_gem_object_lookup); + +/** + * Creates a new mm object and returns a handle to it. + */ +int +drm_gem_create_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_create *args = data; + struct drm_gem_object *obj; + int handle, ret; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + args->size = roundup(args->size, PAGE_SIZE); + + /* Allocate the new object */ + obj = drm_gem_object_alloc(dev, args->size); + if (obj == NULL) + return -ENOMEM; + + ret = drm_gem_handle_create(file_priv, obj, &handle); + mutex_lock(&dev->struct_mutex); + drm_gem_object_handle_unreference(obj); + mutex_unlock(&dev->struct_mutex); + + if (ret) + return ret; + + args->handle = handle; + + return 0; +} + +/** + * Releases the handle to an mm object. + */ +int +drm_gem_close_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_close *args = data; + int ret; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + ret = drm_gem_handle_delete(file_priv, args->handle); + + return ret; +} + +/** + * Reads data from the object referenced by handle. + * + * On error, the contents of *data are undefined. + */ +int +drm_gem_pread_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_pread *args = data; + struct drm_gem_object *obj; + ssize_t read; + loff_t offset; + int ret; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EINVAL; + + mutex_lock(&dev->struct_mutex); + if (dev->driver->gem_set_domain) { + ret = dev->driver->gem_set_domain(obj, file_priv, + DRM_GEM_DOMAIN_CPU, + 0); + if (ret) { + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return ret; + } + } + offset = args->offset; + + read = vfs_read(obj->filp, (char __user *)(uintptr_t)args->data_ptr, + args->size, &offset); + if (read != args->size) { + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + if (read < 0) + return read; + else + return -EINVAL; + } + + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +/** + * Maps the contents of an object, returning the address it is mapped + * into. + * + * While the mapping holds a reference on the contents of the object, it doesn't + * imply a ref on the object itself. + */ +int +drm_gem_mmap_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_mmap *args = data; + struct drm_gem_object *obj; + loff_t offset; + unsigned long addr; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EINVAL; + + offset = args->offset; + + down_write(¤t->mm->mmap_sem); + addr = do_mmap(obj->filp, 0, args->size, + PROT_READ | PROT_WRITE, MAP_SHARED, + args->offset); + up_write(¤t->mm->mmap_sem); + mutex_lock(&dev->struct_mutex); + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + if (IS_ERR((void *)addr)) + return addr; + + args->addr_ptr = (uint64_t) addr; + + return 0; +} + +/** + * Writes data to the object referenced by handle. + * + * On error, the contents of the buffer that were to be modified are undefined. + */ +int +drm_gem_pwrite_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_pwrite *args = data; + struct drm_gem_object *obj; + ssize_t written; + loff_t offset; + int ret; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EINVAL; + + mutex_lock(&dev->struct_mutex); + if (dev->driver->gem_set_domain) { + ret = dev->driver->gem_set_domain(obj, file_priv, + DRM_GEM_DOMAIN_CPU, + DRM_GEM_DOMAIN_CPU); + if (ret) { + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return ret; + } + } + offset = args->offset; + + written = vfs_write(obj->filp, + (char __user *)(uintptr_t) args->data_ptr, + args->size, &offset); + + if (written != args->size) { + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + if (written < 0) + return written; + else + return -EINVAL; + } + + if (dev->driver->gem_flush_pwrite) + dev->driver->gem_flush_pwrite(obj, + args->offset, + args->size); + + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +/** + * Create a global name for an object, returning the name. + * + * Note that the name does not hold a reference; when the object + * is freed, the name goes away. + */ +int +drm_gem_flink_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_flink *args = data; + struct drm_gem_object *obj; + int ret; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EINVAL; + +again: + if (idr_pre_get(&dev->object_name_idr, GFP_KERNEL) == 0) + return -ENOMEM; + + spin_lock(&dev->object_name_lock); + if (obj->name) { + spin_unlock(&dev->object_name_lock); + return -EEXIST; + } + ret = idr_get_new_above(&dev->object_name_idr, obj, 1, + &obj->name); + spin_unlock(&dev->object_name_lock); + if (ret == -EAGAIN) + goto again; + + if (ret != 0) { + mutex_lock(&dev->struct_mutex); + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return ret; + } + + /* + * Leave the reference from the lookup around as the + * name table now holds one + */ + args->name = (uint64_t) obj->name; + + return 0; +} + +/** + * Open an object using the global name, returning a handle and the size. + * + * This handle (of course) holds a reference to the object, so the object + * will not go away until the handle is deleted. + */ +int +drm_gem_open_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_open *args = data; + struct drm_gem_object *obj; + int ret; + int handle; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + spin_lock(&dev->object_name_lock); + obj = idr_find(&dev->object_name_idr, (int) args->name); + if (obj) + drm_gem_object_reference(obj); + spin_unlock(&dev->object_name_lock); + if (!obj) + return -ENOENT; + + ret = drm_gem_handle_create(file_priv, obj, &handle); + mutex_lock(&dev->struct_mutex); + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + if (ret) + return ret; + + args->handle = handle; + args->size = obj->size; + + return 0; +} + +/** + * Called when user space prepares to use an object + */ +int +drm_gem_set_domain_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_gem_set_domain *args = data; + struct drm_gem_object *obj; + int ret; + + if (!(dev->driver->driver_features & DRIVER_GEM)) + return -ENODEV; + + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) + return -EINVAL; + + mutex_lock(&dev->struct_mutex); + if (dev->driver->gem_set_domain) { + ret = dev->driver->gem_set_domain(obj, file_priv, + args->read_domains, + args->write_domain); + } else { + obj->read_domains = args->read_domains; + obj->write_domain = args->write_domain; + ret = 0; + } + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return ret; +} + +/** + * Called at device open time, sets up the structure for handling refcounting + * of mm objects. + */ +void +drm_gem_open(struct drm_device *dev, struct drm_file *file_private) +{ + idr_init(&file_private->object_idr); + spin_lock_init(&file_private->table_lock); +} + +/** + * Called at device close to release the file's + * handle references on objects. + */ +static int +drm_gem_object_release_handle(int id, void *ptr, void *data) +{ + struct drm_gem_object *obj = ptr; + + drm_gem_object_handle_unreference(obj); + + return 0; +} + +/** + * Called at close time when the filp is going away. + * + * Releases any remaining references on objects by this filp. + */ +void +drm_gem_release(struct drm_device *dev, struct drm_file *file_private) +{ + mutex_lock(&dev->struct_mutex); + idr_for_each(&file_private->object_idr, + &drm_gem_object_release_handle, NULL); + + idr_destroy(&file_private->object_idr); + mutex_unlock(&dev->struct_mutex); +} + +/** + * Called after the last reference to the object has been lost. + * + * Frees the object + */ +void +drm_gem_object_free(struct kref *kref) +{ + struct drm_gem_object *obj = (struct drm_gem_object *) kref; + struct drm_device *dev = obj->dev; + + BUG_ON(!mutex_is_locked(&dev->struct_mutex)); + + if (dev->driver->gem_free_object != NULL) + dev->driver->gem_free_object(obj); + + fput(obj->filp); + atomic_dec(&dev->object_count); + kfree(obj); +} +EXPORT_SYMBOL(drm_gem_object_free); + +/** + * Called after the last handle to the object has been closed + * + * Removes any name for the object. Note that this must be + * called before drm_gem_object_free or we'll be touching + * freed memory + */ +void +drm_gem_object_handle_free(struct kref *kref) +{ + struct drm_gem_object *obj = container_of(kref, + struct drm_gem_object, + handlecount); + struct drm_device *dev = obj->dev; + + /* Remove any name for this object */ + spin_lock(&dev->object_name_lock); + if (obj->name) { + idr_remove(&dev->object_name_idr, obj->name); + spin_unlock(&dev->object_name_lock); + /* + * The object name held a reference to this object, drop + * that now. + */ + drm_gem_object_unreference(obj); + } else + spin_unlock(&dev->object_name_lock); + +} +EXPORT_SYMBOL(drm_gem_object_handle_free); + diff --git a/linux-core/drm_lock.c b/linux-core/drm_lock.c index d4c2da0c..58c5f08d 100644 --- a/linux-core/drm_lock.c +++ b/linux-core/drm_lock.c @@ -213,22 +213,16 @@ int drm_lock_take(struct drm_lock_data *lock_data, } while (prev != old); spin_unlock_bh(&lock_data->spinlock); - if (_DRM_LOCKING_CONTEXT(old) == context) { - if (old & _DRM_LOCK_HELD) { - if (context != DRM_KERNEL_CONTEXT) { - DRM_ERROR("%d holds heavyweight lock\n", - context); - } - return 0; + /* Warn on recursive locking of user contexts. */ + if (_DRM_LOCKING_CONTEXT(old) == context && _DRM_LOCK_IS_HELD(old)) { + if (context != DRM_KERNEL_CONTEXT) { + DRM_ERROR("%d holds heavyweight lock\n", + context); } + return 0; } - if ((_DRM_LOCKING_CONTEXT(new)) == context && (new & _DRM_LOCK_HELD)) { - /* Have lock */ - - return 1; - } - return 0; + return !_DRM_LOCK_IS_HELD(old); } /** @@ -381,6 +375,60 @@ void drm_idlelock_release(struct drm_lock_data *lock_data) } EXPORT_SYMBOL(drm_idlelock_release); +/** + * Takes the lock on behalf of the client if needed, using the kernel context. + * + * This allows us to hide the hardware lock when it's required for protection + * of data structures (such as command ringbuffer) shared with the X Server, and + * a way for us to transition to lockless for those requests when the X Server + * stops accessing the ringbuffer directly, without having to update the + * other userland clients. + */ +int drm_client_lock_take(struct drm_device *dev, struct drm_file *file_priv) +{ + int ret; + unsigned long irqflags; + + /* If the client has the lock, we're already done. */ + if (drm_i_have_hw_lock(dev, file_priv)) + return 0; + + /* Client doesn't hold the lock. Block taking the lock with the kernel + * context on behalf of the client, and return whether we were + * successful. + */ + spin_lock_irqsave(&dev->lock.spinlock, irqflags); + dev->lock.user_waiters++; + spin_unlock_irqrestore(&dev->lock.spinlock, irqflags); + ret = wait_event_interruptible(dev->lock.lock_queue, + drm_lock_take(&dev->lock, + DRM_KERNEL_CONTEXT)); + spin_lock_irqsave(&dev->lock.spinlock, irqflags); + dev->lock.user_waiters--; + if (ret != 0) { + spin_unlock_irqrestore(&dev->lock.spinlock, irqflags); + return ret; + } else { + dev->lock.file_priv = file_priv; + dev->lock.lock_time = jiffies; + dev->lock.kernel_held = 1; + file_priv->lock_count++; + spin_unlock_irqrestore(&dev->lock.spinlock, irqflags); + return 0; + } +} +EXPORT_SYMBOL(drm_client_lock_take); + +void drm_client_lock_release(struct drm_device *dev) +{ + if (dev->lock.kernel_held) { + dev->lock.kernel_held = 0; + dev->lock.file_priv = NULL; + drm_lock_free(&dev->lock, DRM_KERNEL_CONTEXT); + } +} +EXPORT_SYMBOL(drm_client_lock_release); + int drm_i_have_hw_lock(struct drm_device *dev, struct drm_file *file_priv) { diff --git a/linux-core/drm_memory.c b/linux-core/drm_memory.c index 75f5b521..4b494f9c 100644 --- a/linux-core/drm_memory.c +++ b/linux-core/drm_memory.c @@ -310,6 +310,7 @@ int drm_free_agp(DRM_AGP_MEM * handle, int pages) { return drm_agp_free_memory(handle) ? 0 : -EINVAL; } +EXPORT_SYMBOL(drm_free_agp); /** Wrapper around agp_bind_memory() */ int drm_bind_agp(DRM_AGP_MEM * handle, unsigned int start) @@ -322,6 +323,7 @@ int drm_unbind_agp(DRM_AGP_MEM * handle) { return drm_agp_unbind_memory(handle); } +EXPORT_SYMBOL(drm_unbind_agp); #else /* __OS_HAS_AGP*/ static void *agp_remap(unsigned long offset, unsigned long size, diff --git a/linux-core/drm_mm.c b/linux-core/drm_memrange.c index 59110293..7014c4e2 100644 --- a/linux-core/drm_mm.c +++ b/linux-core/drm_memrange.c @@ -44,26 +44,26 @@ #include "drmP.h" #include <linux/slab.h> -unsigned long drm_mm_tail_space(struct drm_mm *mm) +unsigned long drm_memrange_tail_space(struct drm_memrange *mm) { struct list_head *tail_node; - struct drm_mm_node *entry; + struct drm_memrange_node *entry; tail_node = mm->ml_entry.prev; - entry = list_entry(tail_node, struct drm_mm_node, ml_entry); + entry = list_entry(tail_node, struct drm_memrange_node, ml_entry); if (!entry->free) return 0; return entry->size; } -int drm_mm_remove_space_from_tail(struct drm_mm *mm, unsigned long size) +int drm_memrange_remove_space_from_tail(struct drm_memrange *mm, unsigned long size) { struct list_head *tail_node; - struct drm_mm_node *entry; + struct drm_memrange_node *entry; tail_node = mm->ml_entry.prev; - entry = list_entry(tail_node, struct drm_mm_node, ml_entry); + entry = list_entry(tail_node, struct drm_memrange_node, ml_entry); if (!entry->free) return -ENOMEM; @@ -75,13 +75,13 @@ int drm_mm_remove_space_from_tail(struct drm_mm *mm, unsigned long size) } -static int drm_mm_create_tail_node(struct drm_mm *mm, +static int drm_memrange_create_tail_node(struct drm_memrange *mm, unsigned long start, unsigned long size) { - struct drm_mm_node *child; + struct drm_memrange_node *child; - child = (struct drm_mm_node *) + child = (struct drm_memrange_node *) drm_ctl_alloc(sizeof(*child), DRM_MEM_MM); if (!child) return -ENOMEM; @@ -98,26 +98,26 @@ static int drm_mm_create_tail_node(struct drm_mm *mm, } -int drm_mm_add_space_to_tail(struct drm_mm *mm, unsigned long size) +int drm_memrange_add_space_to_tail(struct drm_memrange *mm, unsigned long size) { struct list_head *tail_node; - struct drm_mm_node *entry; + struct drm_memrange_node *entry; tail_node = mm->ml_entry.prev; - entry = list_entry(tail_node, struct drm_mm_node, ml_entry); + entry = list_entry(tail_node, struct drm_memrange_node, ml_entry); if (!entry->free) { - return drm_mm_create_tail_node(mm, entry->start + entry->size, size); + return drm_memrange_create_tail_node(mm, entry->start + entry->size, size); } entry->size += size; return 0; } -static struct drm_mm_node *drm_mm_split_at_start(struct drm_mm_node *parent, +static struct drm_memrange_node *drm_memrange_split_at_start(struct drm_memrange_node *parent, unsigned long size) { - struct drm_mm_node *child; + struct drm_memrange_node *child; - child = (struct drm_mm_node *) + child = (struct drm_memrange_node *) drm_ctl_alloc(sizeof(*child), DRM_MEM_MM); if (!child) return NULL; @@ -137,19 +137,19 @@ static struct drm_mm_node *drm_mm_split_at_start(struct drm_mm_node *parent, return child; } -struct drm_mm_node *drm_mm_get_block(struct drm_mm_node * parent, +struct drm_memrange_node *drm_memrange_get_block(struct drm_memrange_node * parent, unsigned long size, unsigned alignment) { - struct drm_mm_node *align_splitoff = NULL; - struct drm_mm_node *child; + struct drm_memrange_node *align_splitoff = NULL; + struct drm_memrange_node *child; unsigned tmp = 0; if (alignment) tmp = parent->start % alignment; if (tmp) { - align_splitoff = drm_mm_split_at_start(parent, alignment - tmp); + align_splitoff = drm_memrange_split_at_start(parent, alignment - tmp); if (!align_splitoff) return NULL; } @@ -159,40 +159,41 @@ struct drm_mm_node *drm_mm_get_block(struct drm_mm_node * parent, parent->free = 0; return parent; } else { - child = drm_mm_split_at_start(parent, size); + child = drm_memrange_split_at_start(parent, size); } if (align_splitoff) - drm_mm_put_block(align_splitoff); + drm_memrange_put_block(align_splitoff); return child; } +EXPORT_SYMBOL(drm_memrange_get_block); /* * Put a block. Merge with the previous and / or next block if they are free. * Otherwise add to the free stack. */ -void drm_mm_put_block(struct drm_mm_node * cur) +void drm_memrange_put_block(struct drm_memrange_node * cur) { - struct drm_mm *mm = cur->mm; + struct drm_memrange *mm = cur->mm; struct list_head *cur_head = &cur->ml_entry; struct list_head *root_head = &mm->ml_entry; - struct drm_mm_node *prev_node = NULL; - struct drm_mm_node *next_node; + struct drm_memrange_node *prev_node = NULL; + struct drm_memrange_node *next_node; int merged = 0; if (cur_head->prev != root_head) { - prev_node = list_entry(cur_head->prev, struct drm_mm_node, ml_entry); + prev_node = list_entry(cur_head->prev, struct drm_memrange_node, ml_entry); if (prev_node->free) { prev_node->size += cur->size; merged = 1; } } if (cur_head->next != root_head) { - next_node = list_entry(cur_head->next, struct drm_mm_node, ml_entry); + next_node = list_entry(cur_head->next, struct drm_memrange_node, ml_entry); if (next_node->free) { if (merged) { prev_node->size += next_node->size; @@ -215,16 +216,16 @@ void drm_mm_put_block(struct drm_mm_node * cur) drm_ctl_free(cur, sizeof(*cur), DRM_MEM_MM); } } -EXPORT_SYMBOL(drm_mm_put_block); +EXPORT_SYMBOL(drm_memrange_put_block); -struct drm_mm_node *drm_mm_search_free(const struct drm_mm * mm, +struct drm_memrange_node *drm_memrange_search_free(const struct drm_memrange * mm, unsigned long size, unsigned alignment, int best_match) { struct list_head *list; const struct list_head *free_stack = &mm->fl_entry; - struct drm_mm_node *entry; - struct drm_mm_node *best; + struct drm_memrange_node *entry; + struct drm_memrange_node *best; unsigned long best_size; unsigned wasted; @@ -232,7 +233,7 @@ struct drm_mm_node *drm_mm_search_free(const struct drm_mm * mm, best_size = ~0UL; list_for_each(list, free_stack) { - entry = list_entry(list, struct drm_mm_node, fl_entry); + entry = list_entry(list, struct drm_memrange_node, fl_entry); wasted = 0; if (entry->size < size) @@ -257,30 +258,31 @@ struct drm_mm_node *drm_mm_search_free(const struct drm_mm * mm, return best; } +EXPORT_SYMBOL(drm_memrange_search_free); -int drm_mm_clean(struct drm_mm * mm) +int drm_memrange_clean(struct drm_memrange * mm) { struct list_head *head = &mm->ml_entry; return (head->next->next == head); } -int drm_mm_init(struct drm_mm * mm, unsigned long start, unsigned long size) +int drm_memrange_init(struct drm_memrange * mm, unsigned long start, unsigned long size) { INIT_LIST_HEAD(&mm->ml_entry); INIT_LIST_HEAD(&mm->fl_entry); - return drm_mm_create_tail_node(mm, start, size); + return drm_memrange_create_tail_node(mm, start, size); } -EXPORT_SYMBOL(drm_mm_init); +EXPORT_SYMBOL(drm_memrange_init); -void drm_mm_takedown(struct drm_mm * mm) +void drm_memrange_takedown(struct drm_memrange * mm) { struct list_head *bnode = mm->fl_entry.next; - struct drm_mm_node *entry; + struct drm_memrange_node *entry; - entry = list_entry(bnode, struct drm_mm_node, fl_entry); + entry = list_entry(bnode, struct drm_memrange_node, fl_entry); if (entry->ml_entry.next != &mm->ml_entry || entry->fl_entry.next != &mm->fl_entry) { @@ -293,4 +295,4 @@ void drm_mm_takedown(struct drm_mm * mm) drm_ctl_free(entry, sizeof(*entry), DRM_MEM_MM); } -EXPORT_SYMBOL(drm_mm_takedown); +EXPORT_SYMBOL(drm_memrange_takedown); diff --git a/linux-core/drm_objects.h b/linux-core/drm_objects.h index 770fbc56..6ec09ef8 100644 --- a/linux-core/drm_objects.h +++ b/linux-core/drm_objects.h @@ -301,7 +301,12 @@ struct drm_ttm_backend_func { void (*destroy) (struct drm_ttm_backend *backend); }; - +/** + * This structure associates a set of flags and methods with a drm_ttm + * object, and will also be subclassed by the particular backend. + * + * \sa #drm_agp_ttm_backend + */ struct drm_ttm_backend { struct drm_device *dev; uint32_t flags; @@ -413,7 +418,7 @@ extern int drm_ttm_destroy(struct drm_ttm *ttm); */ struct drm_bo_mem_reg { - struct drm_mm_node *mm_node; + struct drm_memrange_node *mm_node; unsigned long size; unsigned long num_pages; uint32_t page_alignment; @@ -494,7 +499,7 @@ struct drm_buffer_object { unsigned long num_pages; /* For pinned buffers */ - struct drm_mm_node *pinned_node; + struct drm_memrange_node *pinned_node; uint32_t pinned_mem_type; struct list_head pinned_lru; @@ -529,7 +534,7 @@ struct drm_mem_type_manager { int has_type; int use_type; int kern_init_type; - struct drm_mm manager; + struct drm_memrange manager; struct list_head lru; struct list_head pinned; uint32_t flags; diff --git a/linux-core/drm_proc.c b/linux-core/drm_proc.c index 42da5c69..b1b976b2 100644 --- a/linux-core/drm_proc.c +++ b/linux-core/drm_proc.c @@ -51,6 +51,10 @@ static int drm_bufs_info(char *buf, char **start, off_t offset, int request, int *eof, void *data); static int drm_objects_info(char *buf, char **start, off_t offset, int request, int *eof, void *data); +static int drm_gem_name_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data); +static int drm_gem_object_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data); #if DRM_DEBUG_CODE static int drm_vma_info(char *buf, char **start, off_t offset, int request, int *eof, void *data); @@ -70,6 +74,8 @@ static struct drm_proc_list { {"queues", drm_queues_info}, {"bufs", drm_bufs_info}, {"objects", drm_objects_info}, + {"gem_names", drm_gem_name_info}, + {"gem_objects", drm_gem_object_info}, #if DRM_DEBUG_CODE {"vma", drm_vma_info}, #endif @@ -582,6 +588,79 @@ static int drm_clients_info(char *buf, char **start, off_t offset, return ret; } +struct drm_gem_name_info_data { + int len; + char *buf; + int eof; +}; + +static int drm_gem_one_name_info (int id, void *ptr, void *data) +{ + struct drm_gem_object *obj = ptr; + struct drm_gem_name_info_data *nid = data; + + DRM_INFO ("name %d size %d\n", obj->name, obj->size); + if (nid->eof) + return 0; + + nid->len += sprintf (&nid->buf[nid->len], + "%6d%9d%8d%9d\n", + obj->name, obj->size, + atomic_read(&obj->handlecount.refcount), + atomic_read(&obj->refcount.refcount)); + if (nid->len > DRM_PROC_LIMIT) { + nid->eof = 1; + return 0; + } + return 0; +} + +static int drm_gem_name_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data) +{ + struct drm_minor *minor = (struct drm_minor *) data; + struct drm_device *dev = minor->dev; + struct drm_gem_name_info_data nid; + + if (offset > DRM_PROC_LIMIT) { + *eof = 1; + return 0; + } + + nid.len = sprintf (buf, " name size handles refcount\n"); + nid.buf = buf; + nid.eof = 0; + idr_for_each (&dev->object_name_idr, drm_gem_one_name_info, &nid); + + *start = &buf[offset]; + *eof = 0; + if (nid.len > request + offset) + return request; + *eof = 1; + return nid.len - offset; +} + +static int drm_gem_object_info(char *buf, char **start, off_t offset, + int request, int *eof, void *data) +{ + struct drm_minor *minor = (struct drm_minor *) data; + struct drm_device *dev = minor->dev; + int len = 0; + + if (offset > DRM_PROC_LIMIT) { + *eof = 1; + return 0; + } + + *start = &buf[offset]; + *eof = 0; + DRM_PROC_PRINT ("%d objects\n", atomic_read (&dev->object_count)); + if (len > request + offset) + return request; + *eof = 1; + return len - offset; +} + #if DRM_DEBUG_CODE static int drm__vma_info(char *buf, char **start, off_t offset, int request, diff --git a/linux-core/drm_sman.c b/linux-core/drm_sman.c index 8421a939..7c16f685 100644 --- a/linux-core/drm_sman.c +++ b/linux-core/drm_sman.c @@ -88,34 +88,34 @@ EXPORT_SYMBOL(drm_sman_init); static void *drm_sman_mm_allocate(void *private, unsigned long size, unsigned alignment) { - struct drm_mm *mm = (struct drm_mm *) private; - struct drm_mm_node *tmp; + struct drm_memrange *mm = (struct drm_memrange *) private; + struct drm_memrange_node *tmp; - tmp = drm_mm_search_free(mm, size, alignment, 1); + tmp = drm_memrange_search_free(mm, size, alignment, 1); if (!tmp) { return NULL; } - tmp = drm_mm_get_block(tmp, size, alignment); + tmp = drm_memrange_get_block(tmp, size, alignment); return tmp; } static void drm_sman_mm_free(void *private, void *ref) { - struct drm_mm_node *node = (struct drm_mm_node *) ref; + struct drm_memrange_node *node = (struct drm_memrange_node *) ref; - drm_mm_put_block(node); + drm_memrange_put_block(node); } static void drm_sman_mm_destroy(void *private) { - struct drm_mm *mm = (struct drm_mm *) private; - drm_mm_takedown(mm); + struct drm_memrange *mm = (struct drm_memrange *) private; + drm_memrange_takedown(mm); drm_free(mm, sizeof(*mm), DRM_MEM_MM); } static unsigned long drm_sman_mm_offset(void *private, void *ref) { - struct drm_mm_node *node = (struct drm_mm_node *) ref; + struct drm_memrange_node *node = (struct drm_memrange_node *) ref; return node->start; } @@ -124,7 +124,7 @@ drm_sman_set_range(struct drm_sman * sman, unsigned int manager, unsigned long start, unsigned long size) { struct drm_sman_mm *sman_mm; - struct drm_mm *mm; + struct drm_memrange *mm; int ret; BUG_ON(manager >= sman->num_managers); @@ -135,7 +135,7 @@ drm_sman_set_range(struct drm_sman * sman, unsigned int manager, return -ENOMEM; } sman_mm->private = mm; - ret = drm_mm_init(mm, start, size); + ret = drm_memrange_init(mm, start, size); if (ret) { drm_free(mm, sizeof(*mm), DRM_MEM_MM); diff --git a/linux-core/drm_sman.h b/linux-core/drm_sman.h index 39a39fef..0299776c 100644 --- a/linux-core/drm_sman.h +++ b/linux-core/drm_sman.h @@ -45,7 +45,7 @@ /* * A class that is an abstration of a simple memory allocator. * The sman implementation provides a default such allocator - * using the drm_mm.c implementation. But the user can replace it. + * using the drm_memrange.c implementation. But the user can replace it. * See the SiS implementation, which may use the SiS FB kernel module * for memory management. */ @@ -116,7 +116,7 @@ extern int drm_sman_init(struct drm_sman * sman, unsigned int num_managers, unsigned int user_order, unsigned int owner_order); /* - * Initialize a drm_mm.c allocator. Should be called only once for each + * Initialize a drm_memrange.c allocator. Should be called only once for each * manager unless a customized allogator is used. */ diff --git a/linux-core/drm_stub.c b/linux-core/drm_stub.c index c68adbaf..55841826 100644 --- a/linux-core/drm_stub.c +++ b/linux-core/drm_stub.c @@ -115,15 +115,15 @@ static int drm_fill_in_dev(struct drm_device * dev, struct pci_dev *pdev, if (drm_ht_create(&dev->map_hash, DRM_MAP_HASH_ORDER)) { return -ENOMEM; } - if (drm_mm_init(&dev->offset_manager, DRM_FILE_PAGE_OFFSET_START, - DRM_FILE_PAGE_OFFSET_SIZE)) { + if (drm_memrange_init(&dev->offset_manager, DRM_FILE_PAGE_OFFSET_START, + DRM_FILE_PAGE_OFFSET_SIZE)) { drm_ht_remove(&dev->map_hash); return -ENOMEM; } if (drm_ht_create(&dev->object_hash, DRM_OBJECT_HASH_ORDER)) { drm_ht_remove(&dev->map_hash); - drm_mm_takedown(&dev->offset_manager); + drm_memrange_takedown(&dev->offset_manager); return -ENOMEM; } @@ -163,7 +163,16 @@ static int drm_fill_in_dev(struct drm_device * dev, struct pci_dev *pdev, goto error_out_unreg; } + if (driver->driver_features & DRIVER_GEM) { + retcode = drm_gem_init (dev); + if (retcode) { + DRM_ERROR("Cannot initialize graphics execution manager (GEM)\n"); + goto error_out_unreg; + } + } + drm_fence_manager_init(dev); + return 0; error_out_unreg: diff --git a/linux-core/i915_drv.c b/linux-core/i915_drv.c index 91c2da23..3f246a08 100644 --- a/linux-core/i915_drv.c +++ b/linux-core/i915_drv.c @@ -566,7 +566,7 @@ static struct drm_driver driver = { */ .driver_features = DRIVER_USE_AGP | DRIVER_REQUIRE_AGP | /* DRIVER_USE_MTRR | */ - DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED, + DRIVER_HAVE_IRQ | DRIVER_IRQ_SHARED | DRIVER_GEM, .load = i915_driver_load, .unload = i915_driver_unload, .firstopen = i915_driver_firstopen, @@ -586,6 +586,10 @@ static struct drm_driver driver = { .get_map_ofs = drm_core_get_map_ofs, .get_reg_ofs = drm_core_get_reg_ofs, .ioctls = i915_ioctls, + .gem_init_object = i915_gem_init_object, + .gem_free_object = i915_gem_free_object, + .gem_set_domain = i915_gem_set_domain, + .gem_flush_pwrite = i915_gem_flush_pwrite, .fops = { .owner = THIS_MODULE, .open = drm_open, diff --git a/linux-core/i915_gem.c b/linux-core/i915_gem.c new file mode 100644 index 00000000..bcc15dd3 --- /dev/null +++ b/linux-core/i915_gem.c @@ -0,0 +1,1693 @@ +/* + * Copyright © 2008 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt <eric@anholt.net> + * + */ + +#include "drmP.h" +#include "drm.h" +#include "i915_drm.h" +#include "i915_drv.h" + +#define WATCH_COHERENCY 0 +#define WATCH_BUF 0 +#define WATCH_EXEC 0 +#define WATCH_LRU 0 +#define WATCH_RELOC 0 + +static void +i915_gem_object_set_domain(struct drm_gem_object *obj, + uint32_t read_domains, + uint32_t write_domain); + +int +i915_gem_init_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_init *args = data; + + mutex_lock(&dev->struct_mutex); + + if (args->gtt_start >= args->gtt_end || + (args->gtt_start & (PAGE_SIZE - 1)) != 0 || + (args->gtt_end & (PAGE_SIZE - 1)) != 0) { + mutex_unlock(&dev->struct_mutex); + return -EINVAL; + } + + drm_memrange_init(&dev_priv->mm.gtt_space, args->gtt_start, + args->gtt_end - args->gtt_start); + + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +static void +i915_gem_object_free_page_list(struct drm_gem_object *obj) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int page_count = obj->size / PAGE_SIZE; + int i; + + if (obj_priv->page_list == NULL) + return; + + + for (i = 0; i < page_count; i++) + if (obj_priv->page_list[i] != NULL) + page_cache_release(obj_priv->page_list[i]); + + drm_free(obj_priv->page_list, + page_count * sizeof(struct page *), + DRM_MEM_DRIVER); + obj_priv->page_list = NULL; +} + +static void +i915_gem_object_move_to_active(struct drm_gem_object *obj) +{ + struct drm_device *dev = obj->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + /* Add a reference if we're newly entering the active list. */ + if (!obj_priv->active) { + drm_gem_object_reference(obj); + obj_priv->active = 1; + } + /* Move from whatever list we were on to the tail of execution. */ + list_move_tail(&obj_priv->list, + &dev_priv->mm.active_list); +} + +static void +i915_gem_object_move_to_inactive(struct drm_gem_object *obj) +{ + struct drm_device *dev = obj->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + if (obj_priv->pin_count != 0) + list_del_init(&obj_priv->list); + else + list_move_tail(&obj_priv->list, &dev_priv->mm.inactive_list); + + if (obj_priv->active) { + obj_priv->active = 0; + drm_gem_object_unreference(obj); + } +} + +/** + * Creates a new sequence number, emitting a write of it to the status page + * plus an interrupt, which will trigger i915_user_interrupt_handler. + * + * Must be called with struct_lock held. + * + * Returned sequence numbers are nonzero on success. + */ +static uint32_t +i915_add_request(struct drm_device *dev, uint32_t flush_domains) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_request *request; + uint32_t seqno; + RING_LOCALS; + + request = drm_calloc(1, sizeof(*request), DRM_MEM_DRIVER); + if (request == NULL) + return 0; + + /* Grab the seqno we're going to make this request be, and bump the + * next (skipping 0 so it can be the reserved no-seqno value). + */ + seqno = dev_priv->mm.next_gem_seqno; + dev_priv->mm.next_gem_seqno++; + if (dev_priv->mm.next_gem_seqno == 0) + dev_priv->mm.next_gem_seqno++; + + BEGIN_LP_RING(4); + OUT_RING(CMD_STORE_DWORD_IDX); + OUT_RING(I915_GEM_HWS_INDEX << STORE_DWORD_INDEX_SHIFT); + OUT_RING(seqno); + + OUT_RING(GFX_OP_USER_INTERRUPT); + ADVANCE_LP_RING(); + + DRM_DEBUG("%d\n", seqno); + + request->seqno = seqno; + request->emitted_jiffies = jiffies; + request->flush_domains = flush_domains; + list_add_tail(&request->list, &dev_priv->mm.request_list); + + return seqno; +} + +/** + * Command execution barrier + * + * Ensures that all commands in the ring are finished + * before signalling the CPU + */ + +uint32_t +i915_retire_commands(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + uint32_t cmd = CMD_MI_FLUSH | MI_NO_WRITE_FLUSH; + uint32_t flush_domains = 0; + RING_LOCALS; + + /* The sampler always gets flushed on i965 (sigh) */ + if (IS_I965G(dev)) + flush_domains |= DRM_GEM_DOMAIN_I915_SAMPLER; + BEGIN_LP_RING(2); + OUT_RING(cmd); + OUT_RING(0); /* noop */ + ADVANCE_LP_RING(); + return flush_domains; +} + +/** + * Moves buffers associated only with the given active seqno from the active + * to inactive list, potentially freeing them. + */ +static void +i915_gem_retire_request(struct drm_device *dev, + struct drm_i915_gem_request *request) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + + if (request->flush_domains != 0) { + struct drm_i915_gem_object *obj_priv, *next; + + /* First clear any buffers that were only waiting for a flush + * matching the one just retired. + */ + + list_for_each_entry_safe(obj_priv, next, + &dev_priv->mm.flushing_list, list) { + struct drm_gem_object *obj = obj_priv->obj; + + if (obj->write_domain & request->flush_domains) { + obj->write_domain = 0; + i915_gem_object_move_to_inactive(obj); + } + } + + } + + /* Move any buffers on the active list that are no longer referenced + * by the ringbuffer to the flushing/inactive lists as appropriate. + */ + while (!list_empty(&dev_priv->mm.active_list)) { + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + + obj_priv = list_first_entry(&dev_priv->mm.active_list, + struct drm_i915_gem_object, + list); + obj = obj_priv->obj; + + /* If the seqno being retired doesn't match the oldest in the + * list, then the oldest in the list must still be newer than + * this seqno. + */ + if (obj_priv->last_rendering_seqno != request->seqno) + return; +#if WATCH_LRU + DRM_INFO("%s: retire %d moves to inactive list %p\n", + __func__, request->seqno, obj); +#endif + + if (obj->write_domain != 0) { + list_move_tail(&obj_priv->list, + &dev_priv->mm.flushing_list); + } else { + i915_gem_object_move_to_inactive(obj); + } + } +} + +/** + * Returns true if seq1 is later than seq2. + */ +static int +i915_seqno_passed(uint32_t seq1, uint32_t seq2) +{ + return (int32_t)(seq1 - seq2) >= 0; +} + +static uint32_t +i915_get_gem_seqno(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + + return READ_HWSP(dev_priv, I915_GEM_HWS_INDEX); +} + +/** + * This function clears the request list as sequence numbers are passed. + */ +void +i915_gem_retire_requests(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + uint32_t seqno; + + seqno = i915_get_gem_seqno(dev); + + while (!list_empty(&dev_priv->mm.request_list)) { + struct drm_i915_gem_request *request; + uint32_t retiring_seqno; + + request = list_first_entry(&dev_priv->mm.request_list, + struct drm_i915_gem_request, + list); + retiring_seqno = request->seqno; + + if (i915_seqno_passed(seqno, retiring_seqno)) { + i915_gem_retire_request(dev, request); + + list_del(&request->list); + drm_free(request, sizeof(*request), DRM_MEM_DRIVER); + } else + break; + } +} + +/** + * Waits for a sequence number to be signaled, and cleans up the + * request and object lists appropriately for that event. + */ +int +i915_wait_request(struct drm_device *dev, uint32_t seqno) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + int ret = 0; + + BUG_ON(seqno == 0); + + if (!i915_seqno_passed(i915_get_gem_seqno(dev), seqno)) { + i915_user_irq_on(dev_priv); + ret = wait_event_interruptible(dev_priv->irq_queue, + i915_seqno_passed(i915_get_gem_seqno(dev), + seqno)); + i915_user_irq_off(dev_priv); + } + + /* Directly dispatch request retiring. While we have the work queue + * to handle this, the waiter on a request often wants an associated + * buffer to have made it to the inactive list, and we would need + * a separate wait queue to handle that. + */ + if (ret == 0) + i915_gem_retire_requests(dev); + + return ret; +} + +static void +i915_gem_flush(struct drm_device *dev, + uint32_t invalidate_domains, + uint32_t flush_domains) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + uint32_t cmd; + RING_LOCALS; + +#if WATCH_EXEC + DRM_INFO("%s: invalidate %08x flush %08x\n", __func__, + invalidate_domains, flush_domains); +#endif + + if (flush_domains & DRM_GEM_DOMAIN_CPU) + drm_agp_chipset_flush(dev); + + if ((invalidate_domains|flush_domains) & ~DRM_GEM_DOMAIN_CPU) { + /* + * read/write caches: + * + * DRM_GEM_DOMAIN_I915_RENDER is always invalidated, but is + * only flushed if MI_NO_WRITE_FLUSH is unset. On 965, it is + * also flushed at 2d versus 3d pipeline switches. + * + * read-only caches: + * + * DRM_GEM_DOMAIN_I915_SAMPLER is flushed on pre-965 if + * MI_READ_FLUSH is set, and is always flushed on 965. + * + * DRM_GEM_DOMAIN_I915_COMMAND may not exist? + * + * DRM_GEM_DOMAIN_I915_INSTRUCTION, which exists on 965, is + * invalidated when MI_EXE_FLUSH is set. + * + * DRM_GEM_DOMAIN_I915_VERTEX, which exists on 965, is + * invalidated with every MI_FLUSH. + * + * TLBs: + * + * On 965, TLBs associated with DRM_GEM_DOMAIN_I915_COMMAND + * and DRM_GEM_DOMAIN_CPU in are invalidated at PTE write and + * DRM_GEM_DOMAIN_I915_RENDER and DRM_GEM_DOMAIN_I915_SAMPLER + * are flushed at any MI_FLUSH. + */ + + cmd = CMD_MI_FLUSH | MI_NO_WRITE_FLUSH; + if ((invalidate_domains|flush_domains) & + DRM_GEM_DOMAIN_I915_RENDER) + cmd &= ~MI_NO_WRITE_FLUSH; + if (!IS_I965G(dev)) { + /* + * On the 965, the sampler cache always gets flushed + * and this bit is reserved. + */ + if (invalidate_domains & DRM_GEM_DOMAIN_I915_SAMPLER) + cmd |= MI_READ_FLUSH; + } + if (invalidate_domains & DRM_GEM_DOMAIN_I915_INSTRUCTION) + cmd |= MI_EXE_FLUSH; + +#if WATCH_EXEC + DRM_INFO("%s: queue flush %08x to ring\n", __func__, cmd); +#endif + BEGIN_LP_RING(2); + OUT_RING(cmd); + OUT_RING(0); /* noop */ + ADVANCE_LP_RING(); + } +} + +/** + * Ensures that all rendering to the object has completed and the object is + * safe to unbind from the GTT or access from the CPU. + */ +static int +i915_gem_object_wait_rendering(struct drm_gem_object *obj) +{ + struct drm_device *dev = obj->dev; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int ret; + + /* If there are writes queued to the buffer, flush and + * create a new seqno to wait for. + */ + if (obj->write_domain & ~(DRM_GEM_DOMAIN_CPU)) { + uint32_t write_domain = obj->write_domain; +#if WATCH_BUF + DRM_INFO("%s: flushing object %p from write domain %08x\n", + __func__, obj, write_domain); +#endif + i915_gem_flush(dev, 0, write_domain); + obj->write_domain = 0; + + i915_gem_object_move_to_active(obj); + obj_priv->last_rendering_seqno = i915_add_request(dev, + write_domain); + BUG_ON(obj_priv->last_rendering_seqno == 0); +#if WATCH_LRU + DRM_INFO("%s: flush moves to exec list %p\n", __func__, obj); +#endif + } + /* If there is rendering queued on the buffer being evicted, wait for + * it. + */ + if (obj_priv->active) { +#if WATCH_BUF + DRM_INFO("%s: object %p wait for seqno %08x\n", + __func__, obj, obj_priv->last_rendering_seqno); +#endif + ret = i915_wait_request(dev, obj_priv->last_rendering_seqno); + if (ret != 0) + return ret; + } + + return 0; +} + +/** + * Unbinds an object from the GTT aperture. + */ +static void +i915_gem_object_unbind(struct drm_gem_object *obj) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + +#if WATCH_BUF + DRM_INFO("%s:%d %p\n", __func__, __LINE__, obj); + DRM_INFO("gtt_space %p\n", obj_priv->gtt_space); +#endif + if (obj_priv->gtt_space == NULL) + return; + + /* Move the object to the CPU domain to ensure that + * any possible CPU writes while it's not in the GTT + * are flushed when we go to remap it. This will + * also ensure that all pending GPU writes are finished + * before we unbind. + */ + i915_gem_object_set_domain (obj, DRM_GEM_DOMAIN_CPU, + DRM_GEM_DOMAIN_CPU); + + if (obj_priv->agp_mem != NULL) { + drm_unbind_agp(obj_priv->agp_mem); + drm_free_agp(obj_priv->agp_mem, obj->size / PAGE_SIZE); + obj_priv->agp_mem = NULL; + } + + i915_gem_object_free_page_list(obj); + + drm_memrange_put_block(obj_priv->gtt_space); + obj_priv->gtt_space = NULL; + + /* Remove ourselves from the LRU list if present. */ + if (!list_empty(&obj_priv->list)) { + list_del_init(&obj_priv->list); + if (obj_priv->active) { + DRM_ERROR("Failed to wait on buffer when unbinding, " + "continued anyway.\n"); + obj_priv->active = 0; + drm_gem_object_unreference(obj); + } + } +} + +#if WATCH_BUF | WATCH_EXEC +static void +i915_gem_dump_page(struct page *page, uint32_t start, uint32_t end, + uint32_t bias, uint32_t mark) +{ + uint32_t *mem = kmap_atomic(page, KM_USER0); + int i; + for (i = start; i < end; i += 4) + DRM_INFO("%08x: %08x%s\n", + (int) (bias + i), mem[i / 4], + (bias + i == mark) ? " ********" : ""); + kunmap_atomic(mem, KM_USER0); + /* give syslog time to catch up */ + msleep(1); +} + +static void +i915_gem_dump_object(struct drm_gem_object *obj, int len, + const char *where, uint32_t mark) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int page; + + DRM_INFO("%s: object at offset %08x\n", where, obj_priv->gtt_offset); + for (page = 0; page < (len + PAGE_SIZE-1) / PAGE_SIZE; page++) { + int page_len, chunk, chunk_len; + + page_len = len - page * PAGE_SIZE; + if (page_len > PAGE_SIZE) + page_len = PAGE_SIZE; + + for (chunk = 0; chunk < page_len; chunk += 128) { + chunk_len = page_len - chunk; + if (chunk_len > 128) + chunk_len = 128; + i915_gem_dump_page(obj_priv->page_list[page], + chunk, chunk + chunk_len, + obj_priv->gtt_offset + + page * PAGE_SIZE, + mark); + } + } +} +#endif + +#if WATCH_LRU +static void +i915_dump_lru(struct drm_device *dev, const char *where) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv; + + DRM_INFO("active list %s {\n", where); + list_for_each_entry(obj_priv, &dev_priv->mm.active_list, + list) + { + DRM_INFO(" %p: %08x\n", obj_priv, + obj_priv->last_rendering_seqno); + } + DRM_INFO("}\n"); + DRM_INFO("flushing list %s {\n", where); + list_for_each_entry(obj_priv, &dev_priv->mm.flushing_list, + list) + { + DRM_INFO(" %p: %08x\n", obj_priv, + obj_priv->last_rendering_seqno); + } + DRM_INFO("}\n"); + DRM_INFO("inactive %s {\n", where); + list_for_each_entry(obj_priv, &dev_priv->mm.inactive_list, list) { + DRM_INFO(" %p: %08x\n", obj_priv, + obj_priv->last_rendering_seqno); + } + DRM_INFO("}\n"); +} +#endif + +static int +i915_gem_evict_something(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + + for (;;) { + /* If there's an inactive buffer available now, grab it + * and be done. + */ + if (!list_empty(&dev_priv->mm.inactive_list)) { + obj_priv = list_first_entry(&dev_priv->mm.inactive_list, + struct drm_i915_gem_object, + list); + obj = obj_priv->obj; + BUG_ON(obj_priv->pin_count != 0); + break; + } + + /* If we didn't get anything, but the ring is still processing + * things, wait for one of those things to finish and hopefully + * leave us a buffer to evict. + */ + if (!list_empty(&dev_priv->mm.request_list)) { + struct drm_i915_gem_request *request; + int ret; + + request = list_first_entry(&dev_priv->mm.request_list, + struct drm_i915_gem_request, + list); + + ret = i915_wait_request(dev, request->seqno); + if (ret != 0) + return ret; + + continue; + } + + /* If we didn't have anything on the request list but there + * are buffers awaiting a flush, emit one and try again. + * When we wait on it, those buffers waiting for that flush + * will get moved to inactive. + */ + if (!list_empty(&dev_priv->mm.flushing_list)) { + obj_priv = list_first_entry(&dev_priv->mm.flushing_list, + struct drm_i915_gem_object, + list); + obj = obj_priv->obj; + + i915_gem_flush(dev, + obj->write_domain, + obj->write_domain); + i915_add_request(dev, obj->write_domain); + + obj = NULL; + continue; + } + + /* If we didn't do any of the above, there's nothing to be done + * and we just can't fit it in. + */ + return -ENOMEM; + } + +#if WATCH_LRU + DRM_INFO("%s: evicting %p\n", __func__, obj); +#endif + + BUG_ON(obj_priv->active); + + /* Wait on the rendering and unbind the buffer. */ + i915_gem_object_unbind(obj); + + return 0; +} + +static int +i915_gem_object_get_page_list(struct drm_gem_object *obj) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int page_count, i; + if (obj_priv->page_list) + return 0; + + /* Get the list of pages out of our struct file. They'll be pinned + * at this point until we release them. + */ + page_count = obj->size / PAGE_SIZE; + BUG_ON(obj_priv->page_list != NULL); + obj_priv->page_list = drm_calloc(page_count, sizeof(struct page *), + DRM_MEM_DRIVER); + if (obj_priv->page_list == NULL) + return -ENOMEM; + + for (i = 0; i < page_count; i++) { + obj_priv->page_list[i] = + find_or_create_page(obj->filp->f_mapping, i, GFP_HIGHUSER); + + if (obj_priv->page_list[i] == NULL) { + i915_gem_object_free_page_list(obj); + return -ENOMEM; + } + unlock_page(obj_priv->page_list[i]); + } + return 0; +} + +/** + * Finds free space in the GTT aperture and binds the object there. + */ +static int +i915_gem_object_bind_to_gtt(struct drm_gem_object *obj, unsigned alignment) +{ + struct drm_device *dev = obj->dev; + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + struct drm_memrange_node *free_space; + int page_count, ret; + + if (alignment == 0) + alignment = PAGE_SIZE; + if (alignment & (PAGE_SIZE - 1)) { + DRM_ERROR("Invalid object alignment requested %u\n", alignment); + return -EINVAL; + } + + search_free: + free_space = drm_memrange_search_free(&dev_priv->mm.gtt_space, + obj->size, + alignment, 0); + if (free_space != NULL) { + obj_priv->gtt_space = + drm_memrange_get_block(free_space, obj->size, + alignment); + if (obj_priv->gtt_space != NULL) { + obj_priv->gtt_space->private = obj; + obj_priv->gtt_offset = obj_priv->gtt_space->start; + } + } + if (obj_priv->gtt_space == NULL) { + /* If the gtt is empty and we're still having trouble + * fitting our object in, we're out of memory. + */ +#if WATCH_LRU + DRM_INFO("%s: GTT full, evicting something\n", __func__); +#endif + if (list_empty(&dev_priv->mm.inactive_list) && + list_empty(&dev_priv->mm.active_list)) { + DRM_ERROR("GTT full, but LRU list empty\n"); + return -ENOMEM; + } + + ret = i915_gem_evict_something(dev); + if (ret != 0) + return ret; + goto search_free; + } + +#if WATCH_BUF + DRM_INFO("Binding object of size %d at 0x%08x\n", + obj->size, obj_priv->gtt_offset); +#endif + ret = i915_gem_object_get_page_list(obj); + if (ret) { + drm_memrange_put_block(obj_priv->gtt_space); + obj_priv->gtt_space = NULL; + return ret; + } + + page_count = obj->size / PAGE_SIZE; + /* Create an AGP memory structure pointing at our pages, and bind it + * into the GTT. + */ + obj_priv->agp_mem = drm_agp_bind_pages(dev, + obj_priv->page_list, + page_count, + obj_priv->gtt_offset); + if (obj_priv->agp_mem == NULL) { + i915_gem_object_free_page_list(obj); + drm_memrange_put_block(obj_priv->gtt_space); + obj_priv->gtt_space = NULL; + return -ENOMEM; + } + + /* Assert that the object is not currently in any GPU domain. As it + * wasn't in the GTT, there shouldn't be any way it could have been in + * a GPU cache + */ + BUG_ON(obj->read_domains & ~DRM_GEM_DOMAIN_CPU); + BUG_ON(obj->write_domain & ~DRM_GEM_DOMAIN_CPU); + + return 0; +} + +static void +i915_gem_clflush_object(struct drm_gem_object *obj) +{ + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + /* If we don't have a page list set up, then we're not pinned + * to GPU, and we can ignore the cache flush because it'll happen + * again at bind time. + */ + if (obj_priv->page_list == NULL) + return; + + drm_ttm_cache_flush(obj_priv->page_list, obj->size / PAGE_SIZE); +} + +/* + * Set the next domain for the specified object. This + * may not actually perform the necessary flushing/invaliding though, + * as that may want to be batched with other set_domain operations + * + * This is (we hope) the only really tricky part of gem. The goal + * is fairly simple -- track which caches hold bits of the object + * and make sure they remain coherent. A few concrete examples may + * help to explain how it works. For shorthand, we use the notation + * (read_domains, write_domain), e.g. (CPU, CPU) to indicate the + * a pair of read and write domain masks. + * + * Case 1: the batch buffer + * + * 1. Allocated + * 2. Written by CPU + * 3. Mapped to GTT + * 4. Read by GPU + * 5. Unmapped from GTT + * 6. Freed + * + * Let's take these a step at a time + * + * 1. Allocated + * Pages allocated from the kernel may still have + * cache contents, so we set them to (CPU, CPU) always. + * 2. Written by CPU (using pwrite) + * The pwrite function calls set_domain (CPU, CPU) and + * this function does nothing (as nothing changes) + * 3. Mapped by GTT + * This function asserts that the object is not + * currently in any GPU-based read or write domains + * 4. Read by GPU + * i915_gem_execbuffer calls set_domain (COMMAND, 0). + * As write_domain is zero, this function adds in the + * current read domains (CPU+COMMAND, 0). + * flush_domains is set to CPU. + * invalidate_domains is set to COMMAND + * clflush is run to get data out of the CPU caches + * then i915_dev_set_domain calls i915_gem_flush to + * emit an MI_FLUSH and drm_agp_chipset_flush + * 5. Unmapped from GTT + * i915_gem_object_unbind calls set_domain (CPU, CPU) + * flush_domains and invalidate_domains end up both zero + * so no flushing/invalidating happens + * 6. Freed + * yay, done + * + * Case 2: The shared render buffer + * + * 1. Allocated + * 2. Mapped to GTT + * 3. Read/written by GPU + * 4. set_domain to (CPU,CPU) + * 5. Read/written by CPU + * 6. Read/written by GPU + * + * 1. Allocated + * Same as last example, (CPU, CPU) + * 2. Mapped to GTT + * Nothing changes (assertions find that it is not in the GPU) + * 3. Read/written by GPU + * execbuffer calls set_domain (RENDER, RENDER) + * flush_domains gets CPU + * invalidate_domains gets GPU + * clflush (obj) + * MI_FLUSH and drm_agp_chipset_flush + * 4. set_domain (CPU, CPU) + * flush_domains gets GPU + * invalidate_domains gets CPU + * wait_rendering (obj) to make sure all drawing is complete. + * This will include an MI_FLUSH to get the data from GPU + * to memory + * clflush (obj) to invalidate the CPU cache + * Another MI_FLUSH in i915_gem_flush (eliminate this somehow?) + * 5. Read/written by CPU + * cache lines are loaded and dirtied + * 6. Read written by GPU + * Same as last GPU access + * + * Case 3: The constant buffer + * + * 1. Allocated + * 2. Written by CPU + * 3. Read by GPU + * 4. Updated (written) by CPU again + * 5. Read by GPU + * + * 1. Allocated + * (CPU, CPU) + * 2. Written by CPU + * (CPU, CPU) + * 3. Read by GPU + * (CPU+RENDER, 0) + * flush_domains = CPU + * invalidate_domains = RENDER + * clflush (obj) + * MI_FLUSH + * drm_agp_chipset_flush + * 4. Updated (written) by CPU again + * (CPU, CPU) + * flush_domains = 0 (no previous write domain) + * invalidate_domains = 0 (no new read domains) + * 5. Read by GPU + * (CPU+RENDER, 0) + * flush_domains = CPU + * invalidate_domains = RENDER + * clflush (obj) + * MI_FLUSH + * drm_agp_chipset_flush + */ +static void +i915_gem_object_set_domain(struct drm_gem_object *obj, + uint32_t read_domains, + uint32_t write_domain) +{ + struct drm_device *dev = obj->dev; + uint32_t invalidate_domains = 0; + uint32_t flush_domains = 0; + +#if WATCH_BUF + DRM_INFO("%s: object %p read %08x write %08x\n", + __func__, obj, read_domains, write_domain); +#endif + /* + * If the object isn't moving to a new write domain, + * let the object stay in multiple read domains + */ + if (write_domain == 0) + read_domains |= obj->read_domains; + + /* + * Flush the current write domain if + * the new read domains don't match. Invalidate + * any read domains which differ from the old + * write domain + */ + if (obj->write_domain && obj->write_domain != read_domains) { + flush_domains |= obj->write_domain; + invalidate_domains |= read_domains & ~obj->write_domain; + } + /* + * Invalidate any read caches which may have + * stale data. That is, any new read domains. + */ + invalidate_domains |= read_domains & ~obj->read_domains; + if ((flush_domains | invalidate_domains) & DRM_GEM_DOMAIN_CPU) { +#if WATCH_BUF + DRM_INFO("%s: CPU domain flush %08x invalidate %08x\n", + __func__, flush_domains, invalidate_domains); +#endif + /* + * If we're invaliding the CPU cache and flushing a GPU cache, + * then pause for rendering so that the GPU caches will be + * flushed before the cpu cache is invalidated + */ + if ((invalidate_domains & DRM_GEM_DOMAIN_CPU) && + (flush_domains & ~DRM_GEM_DOMAIN_CPU)) + i915_gem_object_wait_rendering(obj); + i915_gem_clflush_object(obj); + } + + if ((write_domain | flush_domains) != 0) + obj->write_domain = write_domain; + obj->read_domains = read_domains; + dev->invalidate_domains |= invalidate_domains; + dev->flush_domains |= flush_domains; +} + +/** + * Once all of the objects have been set in the proper domain, + * perform the necessary flush and invalidate operations. + * + * Returns the write domains flushed, for use in flush tracking. + */ +static uint32_t +i915_gem_dev_set_domain(struct drm_device *dev) +{ + uint32_t flush_domains = dev->flush_domains; + + /* + * Now that all the buffers are synced to the proper domains, + * flush and invalidate the collected domains + */ + if (dev->invalidate_domains | dev->flush_domains) { +#if WATCH_EXEC + DRM_INFO("%s: invalidate_domains %08x flush_domains %08x\n", + __func__, + dev->invalidate_domains, + dev->flush_domains); +#endif + i915_gem_flush(dev, + dev->invalidate_domains, + dev->flush_domains); + dev->invalidate_domains = 0; + dev->flush_domains = 0; + } + + return flush_domains; +} + +#if WATCH_COHERENCY +static void +i915_gem_object_check_coherency(struct drm_gem_object *obj, int handle) +{ + struct drm_device *dev = obj->dev; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int page; + uint32_t *gtt_mapping; + uint32_t *backing_map = NULL; + int bad_count = 0; + + DRM_INFO("%s: checking coherency of object %p@0x%08x (%d, %dkb):\n", + __FUNCTION__, obj, obj_priv->gtt_offset, handle, + obj->size / 1024); + + gtt_mapping = ioremap(dev->agp->base + obj_priv->gtt_offset, + obj->size); + if (gtt_mapping == NULL) { + DRM_ERROR("failed to map GTT space\n"); + return; + } + + for (page = 0; page < obj->size / PAGE_SIZE; page++) { + int i; + + backing_map = kmap_atomic(obj_priv->page_list[page], KM_USER0); + + if (backing_map == NULL) { + DRM_ERROR("failed to map backing page\n"); + goto out; + } + + for (i = 0; i < PAGE_SIZE / 4; i++) { + uint32_t cpuval = backing_map[i]; + uint32_t gttval = readl(gtt_mapping + + page * 1024 + i); + + if (cpuval != gttval) { + DRM_INFO("incoherent CPU vs GPU at 0x%08x: " + "0x%08x vs 0x%08x\n", + (int)(obj_priv->gtt_offset + + page * PAGE_SIZE + i * 4), + cpuval, gttval); + if (bad_count++ >= 8) { + DRM_INFO("...\n"); + goto out; + } + } + } + kunmap_atomic(backing_map, KM_USER0); + backing_map = NULL; + } + + out: + if (backing_map != NULL) + kunmap_atomic(backing_map, KM_USER0); + iounmap(gtt_mapping); + + /* give syslog time to catch up */ + msleep(1); + + /* Directly flush the object, since we just loaded values with the CPU + * from thebacking pages and we don't want to disturb the cache + * management that we're trying to observe. + */ + + i915_gem_clflush_object(obj); +} +#endif + +static int +i915_gem_reloc_and_validate_object(struct drm_gem_object *obj, + struct drm_file *file_priv, + struct drm_i915_gem_exec_object *entry) +{ + struct drm_device *dev = obj->dev; + struct drm_i915_gem_relocation_entry reloc; + struct drm_i915_gem_relocation_entry __user *relocs; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + int i; + uint32_t last_reloc_offset = -1; + void *reloc_page = NULL; + + /* Choose the GTT offset for our buffer and put it there. */ + if (obj_priv->gtt_space == NULL) { + i915_gem_object_bind_to_gtt(obj, (unsigned) entry->alignment); + if (obj_priv->gtt_space == NULL) + return -ENOMEM; + } + + entry->offset = obj_priv->gtt_offset; + + relocs = (struct drm_i915_gem_relocation_entry __user *) + (uintptr_t) entry->relocs_ptr; + /* Apply the relocations, using the GTT aperture to avoid cache + * flushing requirements. + */ + for (i = 0; i < entry->relocation_count; i++) { + struct drm_gem_object *target_obj; + struct drm_i915_gem_object *target_obj_priv; + uint32_t reloc_val, reloc_offset, *reloc_entry; + int ret; + + ret = copy_from_user(&reloc, relocs + i, sizeof(reloc)); + if (ret != 0) + return ret; + + target_obj = drm_gem_object_lookup(obj->dev, file_priv, + reloc.target_handle); + if (target_obj == NULL) + return -EINVAL; + target_obj_priv = target_obj->driver_private; + + /* The target buffer should have appeared before us in the + * validate list, so it should have a GTT space bound by now. + */ + if (target_obj_priv->gtt_space == NULL) { + DRM_ERROR("No GTT space found for object %d\n", + reloc.target_handle); + drm_gem_object_unreference(target_obj); + return -EINVAL; + } + + if (reloc.offset > obj->size - 4) { + DRM_ERROR("Relocation beyond object bounds: " + "obj %p target %d offset %d size %d.\n", + obj, reloc.target_handle, + (int) reloc.offset, (int) obj->size); + drm_gem_object_unreference(target_obj); + return -EINVAL; + } + if (reloc.offset & 3) { + DRM_ERROR("Relocation not 4-byte aligned: " + "obj %p target %d offset %d.\n", + obj, reloc.target_handle, + (int) reloc.offset); + drm_gem_object_unreference(target_obj); + return -EINVAL; + } + + if (reloc.write_domain && target_obj->pending_write_domain && + reloc.write_domain != target_obj->pending_write_domain) { + DRM_ERROR("Write domain conflict: " + "obj %p target %d offset %d " + "new %08x old %08x\n", + obj, reloc.target_handle, + (int) reloc.offset, + reloc.write_domain, + target_obj->pending_write_domain); + drm_gem_object_unreference(target_obj); + return -EINVAL; + } + +#if WATCH_RELOC + DRM_INFO("%s: obj %p offset %08x target %d " + "read %08x write %08x gtt %08x " + "presumed %08x delta %08x\n", + __func__, + obj, + (int) reloc.offset, + (int) reloc.target_handle, + (int) reloc.read_domains, + (int) reloc.write_domain, + (int) target_obj_priv->gtt_offset, + (int) reloc.presumed_offset, + reloc.delta); +#endif + + target_obj->pending_read_domains |= reloc.read_domains; + target_obj->pending_write_domain |= reloc.write_domain; + + /* If the relocation already has the right value in it, no + * more work needs to be done. + */ + if (target_obj_priv->gtt_offset == reloc.presumed_offset) { + drm_gem_object_unreference(target_obj); + continue; + } + + /* Now that we're going to actually write some data in, + * make sure that any rendering using this buffer's contents + * is completed. + */ + i915_gem_object_wait_rendering(obj); + + /* As we're writing through the gtt, flush + * any CPU writes before we write the relocations + */ + if (obj->write_domain & DRM_GEM_DOMAIN_CPU) { + i915_gem_clflush_object(obj); + drm_agp_chipset_flush(dev); + obj->write_domain = 0; + } + + /* Map the page containing the relocation we're going to + * perform. + */ + reloc_offset = obj_priv->gtt_offset + reloc.offset; + if (reloc_page == NULL || + (last_reloc_offset & ~(PAGE_SIZE - 1)) != + (reloc_offset & ~(PAGE_SIZE - 1))) { + if (reloc_page != NULL) + iounmap(reloc_page); + + reloc_page = ioremap(dev->agp->base + + (reloc_offset & ~(PAGE_SIZE - 1)), + PAGE_SIZE); + last_reloc_offset = reloc_offset; + if (reloc_page == NULL) { + drm_gem_object_unreference(target_obj); + return -ENOMEM; + } + } + + reloc_entry = (uint32_t *)((char *)reloc_page + + (reloc_offset & (PAGE_SIZE - 1))); + reloc_val = target_obj_priv->gtt_offset + reloc.delta; + +#if WATCH_BUF + DRM_INFO("Applied relocation: %p@0x%08x %08x -> %08x\n", + obj, (unsigned int) reloc.offset, + readl(reloc_entry), reloc_val); +#endif + writel(reloc_val, reloc_entry); + + /* Write the updated presumed offset for this entry back out + * to the user. + */ + reloc.presumed_offset = target_obj_priv->gtt_offset; + ret = copy_to_user(relocs + i, &reloc, sizeof(reloc)); + if (ret != 0) { + drm_gem_object_unreference(target_obj); + return ret; + } + + drm_gem_object_unreference(target_obj); + } + + if (reloc_page != NULL) + iounmap(reloc_page); + +#if WATCH_BUF + if (0) + i915_gem_dump_object(obj, 128, __func__, ~0); +#endif + return 0; +} + +static int +i915_dispatch_gem_execbuffer(struct drm_device *dev, + struct drm_i915_gem_execbuffer *exec, + uint64_t exec_offset) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + struct drm_clip_rect __user *boxes = (struct drm_clip_rect __user *) + (uintptr_t) exec->cliprects_ptr; + int nbox = exec->num_cliprects; + int i = 0, count; + uint32_t exec_start, exec_len; + RING_LOCALS; + + exec_start = (uint32_t) exec_offset + exec->batch_start_offset; + exec_len = (uint32_t) exec->batch_len; + + if ((exec_start | exec_len) & 0x7) { + DRM_ERROR("alignment\n"); + return -EINVAL; + } + + if (!exec_start) + return -EINVAL; + + count = nbox ? nbox : 1; + + for (i = 0; i < count; i++) { + if (i < nbox) { + int ret = i915_emit_box(dev, boxes, i, + exec->DR1, exec->DR4); + if (ret) + return ret; + } + + if (dev_priv->use_mi_batchbuffer_start) { + BEGIN_LP_RING(2); + if (IS_I965G(dev)) { + OUT_RING(MI_BATCH_BUFFER_START | + (2 << 6) | + MI_BATCH_NON_SECURE_I965); + OUT_RING(exec_start); + } else { + OUT_RING(MI_BATCH_BUFFER_START | + (2 << 6)); + OUT_RING(exec_start | MI_BATCH_NON_SECURE); + } + ADVANCE_LP_RING(); + + } else { + BEGIN_LP_RING(4); + OUT_RING(MI_BATCH_BUFFER); + OUT_RING(exec_start | MI_BATCH_NON_SECURE); + OUT_RING(exec_start + exec_len - 4); + OUT_RING(0); + ADVANCE_LP_RING(); + } + } + + /* XXX breadcrumb */ + return 0; +} + +/* Throttle our rendering by waiting until the ring has completed our requests + * emitted over 20 msec ago. + * + * This should get us reasonable parallelism between CPU and GPU but also + * relatively low latency when blocking on a particular request to finish. + */ +static int +i915_gem_ring_throttle(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + int ret = 0; + + mutex_lock(&dev->struct_mutex); + while (!list_empty(&dev_priv->mm.request_list)) { + struct drm_i915_gem_request *request; + + request = list_first_entry(&dev_priv->mm.request_list, + struct drm_i915_gem_request, + list); + + /* Break out if we're close enough. */ + if ((long) (jiffies - request->emitted_jiffies) <= (20 * HZ) / 1000) { + mutex_unlock(&dev->struct_mutex); + return 0; + } + + /* Wait on the last request if not. */ + ret = i915_wait_request(dev, request->seqno); + if (ret != 0) { + mutex_unlock(&dev->struct_mutex); + return ret; + } + } + mutex_unlock(&dev->struct_mutex); + return ret; +} + +int +i915_gem_execbuffer(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_execbuffer *args = data; + struct drm_i915_gem_exec_object *validate_list = NULL; + struct drm_gem_object **object_list = NULL; + struct drm_gem_object *batch_obj; + int ret, i; + uint64_t exec_offset; + uint32_t seqno, flush_domains; + + LOCK_TEST_WITH_RETURN(dev, file_priv); + +#if WATCH_EXEC + DRM_INFO("buffers_ptr %d buffer_count %d len %08x\n", + (int) args->buffers_ptr, args->buffer_count, args->batch_len); +#endif + i915_kernel_lost_context(dev); + + ret = i915_gem_ring_throttle(dev); + if (ret) + return ret; + + /* Copy in the validate list from userland */ + validate_list = drm_calloc(sizeof(*validate_list), args->buffer_count, + DRM_MEM_DRIVER); + object_list = drm_calloc(sizeof(*object_list), args->buffer_count, + DRM_MEM_DRIVER); + if (validate_list == NULL || object_list == NULL) { + DRM_ERROR("Failed to allocate validate or object list " + "for %d buffers\n", + args->buffer_count); + ret = -ENOMEM; + goto err; + } + ret = copy_from_user(validate_list, + (struct drm_i915_relocation_entry __user *) + (uintptr_t) args->buffers_ptr, + sizeof(*validate_list) * args->buffer_count); + if (ret != 0) { + DRM_ERROR("copy %d validate entries failed %d\n", + args->buffer_count, ret); + goto err; + } + + mutex_lock(&dev->struct_mutex); + /* Look up object handles and perform the relocations */ + for (i = 0; i < args->buffer_count; i++) { + object_list[i] = drm_gem_object_lookup(dev, file_priv, + validate_list[i].handle); + if (object_list[i] == NULL) { + DRM_ERROR("Invalid object handle %d at index %d\n", + validate_list[i].handle, i); + ret = -EINVAL; + goto err; + } + + ret = i915_gem_reloc_and_validate_object(object_list[i], + file_priv, + &validate_list[i]); + if (ret) { + DRM_ERROR("reloc and validate failed %d\n", ret); + goto err; + } + } + + /* Set the pending read domains for the batch buffer to COMMAND */ + batch_obj = object_list[args->buffer_count-1]; + batch_obj->pending_read_domains = DRM_GEM_DOMAIN_I915_COMMAND; + batch_obj->pending_write_domain = 0; + + for (i = 0; i < args->buffer_count; i++) { + struct drm_gem_object *obj = object_list[i]; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + if (obj_priv->gtt_space == NULL) { + /* We evicted the buffer in the process of validating + * our set of buffers in. We could try to recover by + * kicking them everything out and trying again from + * the start. + */ + ret = -ENOMEM; + goto err; + } + + /* make sure all previous memory operations have passed */ + i915_gem_object_set_domain(obj, + obj->pending_read_domains, + obj->pending_write_domain); + obj->pending_read_domains = 0; + obj->pending_write_domain = 0; + } + + /* Flush/invalidate caches and chipset buffer */ + flush_domains = i915_gem_dev_set_domain(dev); + +#if WATCH_COHERENCY + for (i = 0; i < args->buffer_count; i++) { + i915_gem_object_check_coherency(object_list[i], + validate_list[i].handle); + } +#endif + + exec_offset = validate_list[args->buffer_count - 1].offset; + +#if WATCH_EXEC + i915_gem_dump_object(object_list[args->buffer_count - 1], + args->batch_len, + __func__, + ~0); +#endif + + /* Exec the batchbuffer */ + ret = i915_dispatch_gem_execbuffer(dev, args, exec_offset); + if (ret) { + DRM_ERROR("dispatch failed %d\n", ret); + goto err; + } + + /* + * Ensure that the commands in the batch buffer are + * finished before the interrupt fires + */ + flush_domains |= i915_retire_commands(dev); + + /* + * Get a seqno representing the execution of the current buffer, + * which we can wait on. We would like to mitigate these interrupts, + * likely by only creating seqnos occasionally (so that we have + * *some* interrupts representing completion of buffers that we can + * wait on when trying to clear up gtt space). + */ + seqno = i915_add_request(dev, flush_domains); + BUG_ON(seqno == 0); + for (i = 0; i < args->buffer_count; i++) { + struct drm_gem_object *obj = object_list[i]; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + i915_gem_object_move_to_active(obj); + obj_priv->last_rendering_seqno = seqno; +#if WATCH_LRU + DRM_INFO("%s: move to exec list %p\n", __func__, obj); +#endif + } +#if WATCH_LRU + i915_dump_lru(dev, __func__); +#endif + + /* Copy the new buffer offsets back to the user's validate list. */ + ret = copy_to_user((struct drm_i915_relocation_entry __user *) + (uintptr_t) args->buffers_ptr, + validate_list, + sizeof(*validate_list) * args->buffer_count); + if (ret) + DRM_ERROR("failed to copy %d validate entries " + "back to user (%d)\n", + args->buffer_count, ret); +err: + if (object_list != NULL) { + for (i = 0; i < args->buffer_count; i++) + drm_gem_object_unreference(object_list[i]); + } + mutex_unlock(&dev->struct_mutex); + + drm_free(object_list, sizeof(*object_list) * args->buffer_count, + DRM_MEM_DRIVER); + drm_free(validate_list, sizeof(*validate_list) * args->buffer_count, + DRM_MEM_DRIVER); + + return ret; +} + +int +i915_gem_pin_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_pin *args = data; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + int ret; + + mutex_lock(&dev->struct_mutex); + + i915_kernel_lost_context(dev); + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) { + DRM_ERROR("Bad handle in i915_gem_pin_ioctl(): %d\n", + args->handle); + mutex_unlock(&dev->struct_mutex); + return -EINVAL; + } + + obj_priv = obj->driver_private; + if (obj_priv->gtt_space == NULL) { + ret = i915_gem_object_bind_to_gtt(obj, + (unsigned) args->alignment); + if (ret != 0) { + DRM_ERROR("Failure to bind in " + "i915_gem_pin_ioctl(): %d\n", + ret); + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return ret; + } + } + + obj_priv->pin_count++; + args->offset = obj_priv->gtt_offset; + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + + return 0; +} + +int +i915_gem_unpin_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_pin *args = data; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + + mutex_lock(&dev->struct_mutex); + + i915_kernel_lost_context(dev); + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) { + DRM_ERROR("Bad handle in i915_gem_unpin_ioctl(): %d\n", + args->handle); + mutex_unlock(&dev->struct_mutex); + return -EINVAL; + } + + obj_priv = obj->driver_private; + obj_priv->pin_count--; + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return 0; +} + +int +i915_gem_busy_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv) +{ + struct drm_i915_gem_busy *args = data; + struct drm_gem_object *obj; + struct drm_i915_gem_object *obj_priv; + + mutex_lock(&dev->struct_mutex); + obj = drm_gem_object_lookup(dev, file_priv, args->handle); + if (obj == NULL) { + DRM_ERROR("Bad handle in i915_gem_busy_ioctl(): %d\n", + args->handle); + mutex_unlock(&dev->struct_mutex); + return -EINVAL; + } + + obj_priv = obj->driver_private; + args->busy = obj_priv->active; + + drm_gem_object_unreference(obj); + mutex_unlock(&dev->struct_mutex); + return 0; +} + +int i915_gem_init_object(struct drm_gem_object *obj) +{ + struct drm_i915_gem_object *obj_priv; + + obj_priv = drm_calloc(1, sizeof(*obj_priv), DRM_MEM_DRIVER); + if (obj_priv == NULL) + return -ENOMEM; + + obj->driver_private = obj_priv; + obj_priv->obj = obj; + INIT_LIST_HEAD(&obj_priv->list); + return 0; +} + +void i915_gem_free_object(struct drm_gem_object *obj) +{ + i915_kernel_lost_context(obj->dev); + i915_gem_object_unbind(obj); + + drm_free(obj->driver_private, 1, DRM_MEM_DRIVER); +} + +int +i915_gem_set_domain(struct drm_gem_object *obj, + struct drm_file *file_priv, + uint32_t read_domains, + uint32_t write_domain) +{ + struct drm_device *dev = obj->dev; + + BUG_ON(!mutex_is_locked(&dev->struct_mutex)); + + drm_client_lock_take(dev, file_priv); + i915_kernel_lost_context(dev); + i915_gem_object_set_domain(obj, read_domains, write_domain); + i915_gem_dev_set_domain(obj->dev); + drm_client_lock_release(dev); + + return 0; +} + +int +i915_gem_flush_pwrite(struct drm_gem_object *obj, + uint64_t offset, uint64_t size) +{ +#if 0 + struct drm_device *dev = obj->dev; + struct drm_i915_gem_object *obj_priv = obj->driver_private; + + /* + * For writes much less than the size of the object and + * which are already pinned in memory, do the flush right now + */ + + if ((size < obj->size >> 1) && obj_priv->page_list != NULL) { + unsigned long first_page = offset / PAGE_SIZE; + unsigned long beyond_page = roundup(offset + size, PAGE_SIZE) / PAGE_SIZE; + + drm_ttm_cache_flush(obj_priv->page_list + first_page, + beyond_page - first_page); + drm_agp_chipset_flush(dev); + obj->write_domain = 0; + } +#endif + return 0; +} + +void +i915_gem_lastclose(struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + + mutex_lock(&dev->struct_mutex); + + /* Assume that the chip has been idled at this point. Just pull them + * off the execution list and unref them. Since this is the last + * close, this is also the last ref and they'll go away. + */ + + while (!list_empty(&dev_priv->mm.active_list)) { + struct drm_i915_gem_object *obj_priv; + + obj_priv = list_first_entry(&dev_priv->mm.active_list, + struct drm_i915_gem_object, + list); + + list_del_init(&obj_priv->list); + obj_priv->active = 0; + obj_priv->obj->write_domain = 0; + drm_gem_object_unreference(obj_priv->obj); + } + + mutex_unlock(&dev->struct_mutex); +} diff --git a/linux-core/nouveau_bo.c b/linux-core/nouveau_bo.c index ab3b23a4..86347e03 100644 --- a/linux-core/nouveau_bo.c +++ b/linux-core/nouveau_bo.c @@ -229,7 +229,7 @@ out_cleanup: if (tmp_mem.mm_node) { mutex_lock(&dev->struct_mutex); if (tmp_mem.mm_node != bo->pinned_node) - drm_mm_put_block(tmp_mem.mm_node); + drm_memrange_put_block(tmp_mem.mm_node); tmp_mem.mm_node = NULL; mutex_unlock(&dev->struct_mutex); } diff --git a/linux-core/nouveau_sgdma.c b/linux-core/nouveau_sgdma.c index cc4d5a92..81704ea1 100644 --- a/linux-core/nouveau_sgdma.c +++ b/linux-core/nouveau_sgdma.c @@ -280,7 +280,7 @@ nouveau_sgdma_nottm_hack_init(struct drm_device *dev) struct drm_nouveau_private *dev_priv = dev->dev_private; struct drm_ttm_backend *be; struct drm_scatter_gather sgreq; - struct drm_mm_node mm_node; + struct drm_memrange_node mm_node; struct drm_bo_mem_reg mem; int ret; diff --git a/shared-core/drm.h b/shared-core/drm.h index 52b01cd1..2373a22e 100644 --- a/shared-core/drm.h +++ b/shared-core/drm.h @@ -959,6 +959,101 @@ struct drm_mm_info_arg { uint64_t p_size; }; + +struct drm_gem_create { + /** + * Requested size for the object. + * + * The (page-aligned) allocated size for the object will be returned. + */ + uint64_t size; + /** + * Returned handle for the object. + * + * Object handles are nonzero. + */ + uint32_t handle; + uint32_t pad; +}; + +struct drm_gem_close { + /** Handle of the object to be closed. */ + uint32_t handle; + uint32_t pad; +}; + +struct drm_gem_pread { + /** Handle for the object being read. */ + uint32_t handle; + uint32_t pad; + /** Offset into the object to read from */ + uint64_t offset; + /** Length of data to read */ + uint64_t size; + /** Pointer to write the data into. */ + uint64_t data_ptr; /* void *, but pointers are not 32/64 compatible */ +}; + +struct drm_gem_pwrite { + /** Handle for the object being written to. */ + uint32_t handle; + uint32_t pad; + /** Offset into the object to write to */ + uint64_t offset; + /** Length of data to write */ + uint64_t size; + /** Pointer to read the data from. */ + uint64_t data_ptr; /* void *, but pointers are not 32/64 compatible */ +}; + +struct drm_gem_mmap { + /** Handle for the object being mapped. */ + uint32_t handle; + uint32_t pad; + /** Offset in the object to map. */ + uint64_t offset; + /** + * Length of data to map. + * + * The value will be page-aligned. + */ + uint64_t size; + /** Returned pointer the data was mapped at */ + uint64_t addr_ptr; /* void *, but pointers are not 32/64 compatible */ +}; + +struct drm_gem_flink { + /** Handle for the object being named */ + uint32_t handle; + + /** Returned global name */ + uint32_t name; +}; + +struct drm_gem_open { + /** Name of object being opened */ + uint32_t name; + + /** Returned handle for the object */ + uint32_t handle; + + /** Returned size of the object */ + uint64_t size; +}; + +struct drm_gem_set_domain { + /** Handle for the object */ + uint32_t handle; + + /** New read domains */ + uint32_t read_domains; + + /** New write domain */ + uint32_t write_domain; +}; + +#define DRM_GEM_DOMAIN_CPU 0x00000001 + /** * \name Ioctls Definitions */ @@ -978,7 +1073,7 @@ struct drm_mm_info_arg { #define DRM_IOCTL_GET_CLIENT DRM_IOWR(0x05, struct drm_client) #define DRM_IOCTL_GET_STATS DRM_IOR( 0x06, struct drm_stats) #define DRM_IOCTL_SET_VERSION DRM_IOWR(0x07, struct drm_set_version) -#define DRM_IOCTL_MODESET_CTL DRM_IOW(0x08, struct drm_modeset_ctl) +#define DRM_IOCTL_MODESET_CTL DRM_IOW(0x08, struct drm_modeset_ctl) #define DRM_IOCTL_SET_UNIQUE DRM_IOW( 0x10, struct drm_unique) #define DRM_IOCTL_AUTH_MAGIC DRM_IOW( 0x11, struct drm_auth) @@ -1027,6 +1122,15 @@ struct drm_mm_info_arg { #define DRM_IOCTL_UPDATE_DRAW DRM_IOW(0x3f, struct drm_update_draw) +#define DRM_IOCTL_GEM_CREATE DRM_IOWR(0x09, struct drm_gem_create) +#define DRM_IOCTL_GEM_CLOSE DRM_IOW (0x0a, struct drm_gem_close) +#define DRM_IOCTL_GEM_PREAD DRM_IOW (0x0b, struct drm_gem_pread) +#define DRM_IOCTL_GEM_PWRITE DRM_IOW (0x0c, struct drm_gem_pwrite) +#define DRM_IOCTL_GEM_MMAP DRM_IOWR(0x0d, struct drm_gem_mmap) +#define DRM_IOCTL_GEM_FLINK DRM_IOWR(0x0e, struct drm_gem_flink) +#define DRM_IOCTL_GEM_OPEN DRM_IOWR(0x0f, struct drm_gem_open) +#define DRM_IOCTL_GEM_SET_DOMAIN DRM_IOW (0xb7, struct drm_gem_set_domain) + #define DRM_IOCTL_MM_INIT DRM_IOWR(0xc0, struct drm_mm_init_arg) #define DRM_IOCTL_MM_TAKEDOWN DRM_IOWR(0xc1, struct drm_mm_type_arg) #define DRM_IOCTL_MM_LOCK DRM_IOWR(0xc2, struct drm_mm_type_arg) diff --git a/shared-core/i915_dma.c b/shared-core/i915_dma.c index 60b405d4..a948834a 100644 --- a/shared-core/i915_dma.c +++ b/shared-core/i915_dma.c @@ -41,10 +41,14 @@ int i915_wait_ring(struct drm_device * dev, int n, const char *caller) drm_i915_private_t *dev_priv = dev->dev_private; drm_i915_ring_buffer_t *ring = &(dev_priv->ring); u32 last_head = I915_READ(LP_RING + RING_HEAD) & HEAD_ADDR; + u32 acthd_reg = IS_I965G(dev) ? I965REG_ACTHD : I915REG_ACTHD; + u32 last_acthd = I915_READ(acthd_reg); + u32 acthd; int i; - for (i = 0; i < 10000; i++) { + for (i = 0; i < 100000; i++) { ring->head = I915_READ(LP_RING + RING_HEAD) & HEAD_ADDR; + acthd = I915_READ(acthd_reg); ring->space = ring->head - (ring->tail + 8); if (ring->space < 0) ring->space += ring->Size; @@ -54,13 +58,41 @@ int i915_wait_ring(struct drm_device * dev, int n, const char *caller) if (ring->head != last_head) i = 0; + if (acthd != last_acthd) + i = 0; + last_head = ring->head; - DRM_UDELAY(1); + last_acthd = acthd; + msleep_interruptible (10); } return -EBUSY; } +#if I915_RING_VALIDATE +/** + * Validate the cached ring tail value + * + * If the X server writes to the ring and DRM doesn't + * reload the head and tail pointers, it will end up writing + * data to the wrong place in the ring, causing havoc. + */ +void i915_ring_validate(struct drm_device *dev, const char *func, int line) +{ + drm_i915_private_t *dev_priv = dev->dev_private; + drm_i915_ring_buffer_t *ring = &(dev_priv->ring); + u32 tail = I915_READ(LP_RING+RING_TAIL) & HEAD_ADDR; + u32 head = I915_READ(LP_RING+RING_HEAD) & HEAD_ADDR; + + if (tail != ring->tail) { + DRM_ERROR("%s:%d head sw %x, hw %x. tail sw %x hw %x\n", + func, line, + ring->head, head, ring->tail, tail); + BUG_ON(1); + } +} +#endif + void i915_kernel_lost_context(struct drm_device * dev) { drm_i915_private_t *dev_priv = dev->dev_private; @@ -459,9 +491,9 @@ static int i915_emit_cmds(struct drm_device *dev, int __user *buffer, return 0; } -static int i915_emit_box(struct drm_device * dev, - struct drm_clip_rect __user * boxes, - int i, int DR1, int DR4) +int i915_emit_box(struct drm_device * dev, + struct drm_clip_rect __user * boxes, + int i, int DR1, int DR4) { drm_i915_private_t *dev_priv = dev->dev_private; struct drm_clip_rect box; @@ -517,7 +549,7 @@ void i915_emit_breadcrumb(struct drm_device *dev) BEGIN_LP_RING(4); OUT_RING(CMD_STORE_DWORD_IDX); - OUT_RING(20); + OUT_RING(5 << STORE_DWORD_INDEX_SHIFT); OUT_RING(dev_priv->counter); OUT_RING(0); ADVANCE_LP_RING(); @@ -715,9 +747,19 @@ void i915_dispatch_flip(struct drm_device * dev, int planes, int sync) int i915_quiescent(struct drm_device *dev) { drm_i915_private_t *dev_priv = dev->dev_private; + int ret; i915_kernel_lost_context(dev); - return i915_wait_ring(dev, dev_priv->ring.Size - 8, __FUNCTION__); + ret = i915_wait_ring(dev, dev_priv->ring.Size - 8, __FUNCTION__); + if (ret) + { + i915_kernel_lost_context (dev); + DRM_ERROR ("not quiescent head %08x tail %08x space %08x\n", + dev_priv->ring.head, + dev_priv->ring.tail, + dev_priv->ring.space); + } + return ret; } static int i915_flush_ioctl(struct drm_device *dev, void *data, @@ -1026,6 +1068,7 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags) memset(dev_priv, 0, sizeof(drm_i915_private_t)); dev->dev_private = (void *)dev_priv; + dev_priv->dev = dev; /* Add register map (needed for suspend/resume) */ base = drm_get_resource_start(dev, mmio_bar); @@ -1034,6 +1077,14 @@ int i915_driver_load(struct drm_device *dev, unsigned long flags) ret = drm_addmap(dev, base, size, _DRM_REGISTERS, _DRM_KERNEL | _DRM_DRIVER, &dev_priv->mmio_map); + INIT_LIST_HEAD(&dev_priv->mm.active_list); + INIT_LIST_HEAD(&dev_priv->mm.flushing_list); + INIT_LIST_HEAD(&dev_priv->mm.inactive_list); + INIT_LIST_HEAD(&dev_priv->mm.request_list); + INIT_WORK(&dev_priv->user_interrupt_task, + i915_user_interrupt_handler); + dev_priv->mm.next_gem_seqno = 1; + #ifdef __linux__ #if LINUX_VERSION_CODE < KERNEL_VERSION(2,6,25) intel_init_chipset_flush_compat(dev); @@ -1074,6 +1125,7 @@ void i915_driver_lastclose(struct drm_device * dev) dev_priv->val_bufs = NULL; } #endif + i915_gem_lastclose(dev); if (drm_getsarea(dev) && dev_priv->sarea_priv) i915_do_cleanup_pageflip(dev); @@ -1125,6 +1177,11 @@ struct drm_ioctl_desc i915_ioctls[] = { #ifdef I915_HAVE_BUFFER DRM_IOCTL_DEF(DRM_I915_EXECBUFFER, i915_execbuffer, DRM_AUTH), #endif + DRM_IOCTL_DEF(DRM_I915_GEM_INIT, i915_gem_init_ioctl, DRM_AUTH), + DRM_IOCTL_DEF(DRM_I915_GEM_EXECBUFFER, i915_gem_execbuffer, DRM_AUTH), + DRM_IOCTL_DEF(DRM_I915_GEM_PIN, i915_gem_pin_ioctl, DRM_AUTH|DRM_ROOT_ONLY), + DRM_IOCTL_DEF(DRM_I915_GEM_UNPIN, i915_gem_unpin_ioctl, DRM_AUTH|DRM_ROOT_ONLY), + DRM_IOCTL_DEF(DRM_I915_GEM_BUSY, i915_gem_busy_ioctl, DRM_AUTH), }; int i915_max_ioctl = DRM_ARRAY_SIZE(i915_ioctls); diff --git a/shared-core/i915_drm.h b/shared-core/i915_drm.h index 97e77428..4712ea4f 100644 --- a/shared-core/i915_drm.h +++ b/shared-core/i915_drm.h @@ -176,6 +176,11 @@ typedef struct drm_i915_sarea { #define DRM_I915_MMIO 0x10 #define DRM_I915_HWS_ADDR 0x11 #define DRM_I915_EXECBUFFER 0x12 +#define DRM_I915_GEM_INIT 0x13 +#define DRM_I915_GEM_EXECBUFFER 0x14 +#define DRM_I915_GEM_PIN 0x15 +#define DRM_I915_GEM_UNPIN 0x16 +#define DRM_I915_GEM_BUSY 0x17 #define DRM_IOCTL_I915_INIT DRM_IOW( DRM_COMMAND_BASE + DRM_I915_INIT, drm_i915_init_t) #define DRM_IOCTL_I915_FLUSH DRM_IO ( DRM_COMMAND_BASE + DRM_I915_FLUSH) @@ -195,6 +200,11 @@ typedef struct drm_i915_sarea { #define DRM_IOCTL_I915_VBLANK_SWAP DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_VBLANK_SWAP, drm_i915_vblank_swap_t) #define DRM_IOCTL_I915_MMIO DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_MMIO, drm_i915_mmio) #define DRM_IOCTL_I915_EXECBUFFER DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_EXECBUFFER, struct drm_i915_execbuffer) +#define DRM_IOCTL_I915_GEM_INIT DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_INIT, struct drm_i915_gem_init) +#define DRM_IOCTL_I915_GEM_EXECBUFFER DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_EXECBUFFER, struct drm_i915_gem_execbuffer) +#define DRM_IOCTL_I915_GEM_PIN DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_PIN, struct drm_i915_gem_pin) +#define DRM_IOCTL_I915_GEM_UNPIN DRM_IOW(DRM_COMMAND_BASE + DRM_I915_GEM_UNPIN, struct drm_i915_gem_unpin) +#define DRM_IOCTL_I915_GEM_BUSY DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_BUSY, struct drm_i915_gem_busy) /* Asynchronous page flipping: */ @@ -395,4 +405,146 @@ struct drm_i915_execbuffer { struct drm_fence_arg fence_arg; }; +struct drm_i915_gem_init { + /** + * Beginning offset in the GTT to be managed by the DRM memory + * manager. + */ + uint64_t gtt_start; + /** + * Ending offset in the GTT to be managed by the DRM memory + * manager. + */ + uint64_t gtt_end; +}; + +struct drm_i915_gem_relocation_entry { + /** + * Handle of the buffer being pointed to by this relocation entry. + * + * It's appealing to make this be an index into the mm_validate_entry + * list to refer to the buffer, but this allows the driver to create + * a relocation list for state buffers and not re-write it per + * exec using the buffer. + */ + uint32_t target_handle; + + /** + * Value to be added to the offset of the target buffer to make up + * the relocation entry. + */ + uint32_t delta; + + /** Offset in the buffer the relocation entry will be written into */ + uint64_t offset; + + /** + * Offset value of the target buffer that the relocation entry was last + * written as. + * + * If the buffer has the same offset as last time, we can skip syncing + * and writing the relocation. This value is written back out by + * the execbuffer ioctl when the relocation is written. + */ + uint64_t presumed_offset; + + /** + * Target memory domains read by this operation. + */ + uint32_t read_domains; + + /** + * Target memory domains written by this operation. + * + * Note that only one domain may be written by the whole + * execbuffer operation, so that where there are conflicts, + * the application will get -EINVAL back. + */ + uint32_t write_domain; +}; + +/** + * Intel memory domains + * + * Most of these just align with the various caches in + * the system and are used to flush and invalidate as + * objects end up cached in different domains. + */ + +/* 0x00000001 is DRM_GEM_DOMAIN_CPU */ +#define DRM_GEM_DOMAIN_I915_RENDER 0x00000002 /* Render cache, used by 2D and 3D drawing */ +#define DRM_GEM_DOMAIN_I915_SAMPLER 0x00000004 /* Sampler cache, used by texture engine */ +#define DRM_GEM_DOMAIN_I915_COMMAND 0x00000008 /* Command queue, used to load batch buffers */ +#define DRM_GEM_DOMAIN_I915_INSTRUCTION 0x00000010 /* Instruction cache, used by shader programs */ +#define DRM_GEM_DOMAIN_I915_VERTEX 0x00000020 /* Vertex address cache */ + +struct drm_i915_gem_exec_object { + /** + * User's handle for a buffer to be bound into the GTT for this + * operation. + */ + uint32_t handle; + + /** List of relocations to be performed on this buffer */ + uint32_t relocation_count; + uint64_t relocs_ptr; /* struct drm_i915_gem_relocation_entry *relocs */ + + /** Required alignment in graphics aperture */ + uint64_t alignment; + + /** + * Returned value of the updated offset of the object, for future + * presumed_offset writes. + */ + uint64_t offset; +}; + +struct drm_i915_gem_execbuffer { + /** + * List of buffers to be validated with their relocations to be + * performend on them. + * + * These buffers must be listed in an order such that all relocations + * a buffer is performing refer to buffers that have already appeared + * in the validate list. + */ + uint64_t buffers_ptr; /* struct drm_i915_gem_validate_entry *buffers */ + uint32_t buffer_count; + + /** Offset in the batchbuffer to start execution from. */ + uint32_t batch_start_offset; + /** Bytes used in batchbuffer from batch_start_offset */ + uint32_t batch_len; + uint32_t DR1; + uint32_t DR4; + uint32_t num_cliprects; + uint64_t cliprects_ptr; /* struct drm_clip_rect *cliprects */ +}; + +struct drm_i915_gem_pin { + /** Handle of the buffer to be pinned. */ + uint32_t handle; + uint32_t pad; + + /** alignment required within the aperture */ + uint64_t alignment; + + /** Returned GTT offset of the buffer. */ + uint64_t offset; +}; + +struct drm_i915_gem_unpin { + /** Handle of the buffer to be unpinned. */ + uint32_t handle; + uint32_t pad; +}; + +struct drm_i915_gem_busy { + /** Handle of the buffer to check for busy */ + uint32_t handle; + + /** Return busy status (1 if busy, 0 if idle) */ + uint32_t busy; +}; + #endif /* _I915_DRM_H_ */ diff --git a/shared-core/i915_drv.h b/shared-core/i915_drv.h index 6421f689..d646177b 100644 --- a/shared-core/i915_drv.h +++ b/shared-core/i915_drv.h @@ -101,6 +101,8 @@ typedef struct _drm_i915_vbl_swap { } drm_i915_vbl_swap_t; typedef struct drm_i915_private { + struct drm_device *dev; + drm_local_map_t *sarea; drm_local_map_t *mmio_map; @@ -129,7 +131,7 @@ typedef struct drm_i915_private { DRM_SPINTYPE user_irq_lock; int user_irq_refcount; int fence_irq_on; - uint32_t irq_enable_reg; + uint32_t irq_mask_reg; int irq_enabled; #ifdef I915_HAVE_FENCE @@ -241,6 +243,47 @@ typedef struct drm_i915_private { u8 saveDACMASK; u8 saveDACDATA[256*3]; /* 256 3-byte colors */ u8 saveCR[37]; + + struct { + struct drm_memrange gtt_space; + + /** + * List of objects currently involved in rendering from the + * ringbuffer. + * + * A reference is held on the buffer while on this list. + */ + struct list_head active_list; + + /** + * List of objects which are not in the ringbuffer but which + * still have a write_domain which needs to be flushed before + * unbinding. + * + * A reference is held on the buffer while on this list. + */ + struct list_head flushing_list; + + /** + * LRU list of objects which are not in the ringbuffer and + * are ready to unbind, but are still in the GTT. + * + * A reference is not held on the buffer while on this list, + * as merely being GTT-bound shouldn't prevent its being + * freed, and we'll pull it off the list in the free path. + */ + struct list_head inactive_list; + + /** + * List of breadcrumbs associated with GPU requests currently + * outstanding. + */ + struct list_head request_list; + + uint32_t next_gem_seqno; + } mm; + + struct work_struct user_interrupt_task; } drm_i915_private_t; enum intel_chip_family { @@ -250,6 +293,68 @@ enum intel_chip_family { CHIP_I965 = 0x08, }; +/** driver private structure attached to each drm_gem_object */ +struct drm_i915_gem_object { + struct drm_gem_object *obj; + + /** Current space allocated to this object in the GTT, if any. */ + struct drm_memrange_node *gtt_space; + + /** This object's place on the active/flushing/inactive lists */ + struct list_head list; + + /** + * This is set if the object is on the active or flushing lists + * (has pending rendering), and is not set if it's on inactive (ready + * to be unbound). + */ + int active; + + /** AGP memory structure for our GTT binding. */ + DRM_AGP_MEM *agp_mem; + + struct page **page_list; + + /** + * Current offset of the object in GTT space. + * + * This is the same as gtt_space->start + */ + uint32_t gtt_offset; + + /** Boolean whether this object has a valid gtt offset. */ + int gtt_bound; + + /** How many users have pinned this object in GTT space */ + int pin_count; + + /** Breadcrumb of last rendering to the buffer. */ + uint32_t last_rendering_seqno; +}; + +/** + * Request queue structure. + * + * The request queue allows us to note sequence numbers that have been emitted + * and may be associated with active buffers to be retired. + * + * By keeping this list, we can avoid having to do questionable + * sequence-number comparisons on buffer last_rendering_seqnos, and associate + * an emission time with seqnos for tracking how far ahead of the GPU we are. + */ +struct drm_i915_gem_request { + /** GEM sequence number associated with this request. */ + uint32_t seqno; + + /** Time at which this request was emitted, in jiffies. */ + unsigned long emitted_jiffies; + + /** Cache domains that were flushed at the start of the request. */ + uint32_t flush_domains; + + struct list_head list; +}; + extern struct drm_ioctl_desc i915_ioctls[]; extern int i915_max_ioctl; @@ -271,6 +376,10 @@ extern int i915_dispatch_batchbuffer(struct drm_device * dev, drm_i915_batchbuffer_t * batch); extern int i915_quiescent(struct drm_device *dev); +int i915_emit_box(struct drm_device * dev, + struct drm_clip_rect __user * boxes, + int i, int DR1, int DR4); + /* i915_irq.c */ extern int i915_irq_emit(struct drm_device *dev, void *data, struct drm_file *file_priv); @@ -286,6 +395,7 @@ extern int i915_vblank_pipe_set(struct drm_device *dev, void *data, extern int i915_vblank_pipe_get(struct drm_device *dev, void *data, struct drm_file *file_priv); extern int i915_emit_irq(struct drm_device * dev); +extern int i915_wait_irq(struct drm_device * dev, int irq_nr); extern int i915_enable_vblank(struct drm_device *dev, int crtc); extern void i915_disable_vblank(struct drm_device *dev, int crtc); extern u32 i915_get_vblank_counter(struct drm_device *dev, int crtc); @@ -293,6 +403,7 @@ extern int i915_vblank_swap(struct drm_device *dev, void *data, struct drm_file *file_priv); extern void i915_user_irq_on(drm_i915_private_t *dev_priv); extern void i915_user_irq_off(drm_i915_private_t *dev_priv); +extern void i915_user_interrupt_handler(struct work_struct *work); /* i915_mem.c */ extern int i915_mem_alloc(struct drm_device *dev, void *data, @@ -329,7 +440,27 @@ void i915_flush_ttm(struct drm_ttm *ttm); /* i915_execbuf.c */ int i915_execbuffer(struct drm_device *dev, void *data, struct drm_file *file_priv); - +/* i915_gem.c */ +int i915_gem_init_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_execbuffer(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_pin_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_unpin_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_busy_ioctl(struct drm_device *dev, void *data, + struct drm_file *file_priv); +int i915_gem_init_object(struct drm_gem_object *obj); +void i915_gem_free_object(struct drm_gem_object *obj); +int i915_gem_set_domain(struct drm_gem_object *obj, + struct drm_file *file_priv, + uint32_t read_domains, + uint32_t write_domain); +int i915_gem_flush_pwrite(struct drm_gem_object *obj, + uint64_t offset, uint64_t size); +void i915_gem_lastclose(struct drm_device *dev); +void i915_gem_retire_requests(struct drm_device *dev); #endif #ifdef __linux__ @@ -345,14 +476,23 @@ extern void intel_fini_chipset_flush_compat(struct drm_device *dev); #define I915_WRITE16(reg,val) DRM_WRITE16(dev_priv->mmio_map, (reg), (val)) #define I915_VERBOSE 0 +#define I915_RING_VALIDATE 0 #define RING_LOCALS unsigned int outring, ringmask, outcount; \ volatile char *virt; +#if I915_RING_VALIDATE +void i915_ring_validate(struct drm_device *dev, const char *func, int line); +#define I915_RING_DO_VALIDATE(dev) i915_ring_validate(dev, __FUNCTION__, __LINE__) +#else +#define I915_RING_DO_VALIDATE(dev) +#endif + #define BEGIN_LP_RING(n) do { \ if (I915_VERBOSE) \ DRM_DEBUG("BEGIN_LP_RING(%d)\n", \ (n)); \ + I915_RING_DO_VALIDATE(dev); \ if (dev_priv->ring.space < (n)*4) \ i915_wait_ring(dev, (n)*4, __FUNCTION__); \ outcount = 0; \ @@ -371,6 +511,7 @@ extern void intel_fini_chipset_flush_compat(struct drm_device *dev); #define ADVANCE_LP_RING() do { \ if (I915_VERBOSE) DRM_DEBUG("ADVANCE_LP_RING %x\n", outring); \ + I915_RING_DO_VALIDATE(dev); \ dev_priv->ring.tail = outring; \ dev_priv->ring.space -= outcount * 4; \ I915_WRITE(LP_RING + RING_TAIL, outring); \ @@ -426,7 +567,12 @@ extern int i915_wait_ring(struct drm_device * dev, int n, const char *caller); #define GFX_OP_BREAKPOINT_INTERRUPT ((0<<29)|(1<<23)) #define CMD_REPORT_HEAD (7<<23) #define CMD_STORE_DWORD_IMM ((0x20<<23) | (0x1 << 22) | 0x1) +/** + * Stores a 32-bit integer to the status page at the dword index given. + */ #define CMD_STORE_DWORD_IDX ((0x21<<23) | 0x1) +# define STORE_DWORD_INDEX_SHIFT 2 + #define CMD_OP_BATCH_BUFFER ((0x0<<29)|(0x30<<23)|0x1) #define CMD_MI_FLUSH (0x04 << 23) @@ -501,6 +647,8 @@ extern int i915_wait_ring(struct drm_device * dev, int n, const char *caller); #define I915REG_INT_MASK_R 0x020a8 #define I915REG_INT_ENABLE_R 0x020a0 #define I915REG_INSTPM 0x020c0 +#define I965REG_ACTHD 0x02074 +#define I915REG_ACTHD 0x020C8 #define PIPEADSL 0x70000 #define PIPEBDSL 0x71000 @@ -773,7 +921,22 @@ extern int i915_wait_ring(struct drm_device * dev, int n, const char *caller); #define BREADCRUMB_MASK ((1U << BREADCRUMB_BITS) - 1) #define READ_BREADCRUMB(dev_priv) (((volatile u32*)(dev_priv->hw_status_page))[5]) + +/** + * Reads a dword out of the status page, which is written to from the command + * queue by automatic updates, MI_REPORT_HEAD, MI_STORE_DATA_INDEX, or + * MI_STORE_DATA_IMM. + * + * The following dwords have a reserved meaning: + * 0: ISR copy, updated when an ISR bit not set in the HWSTAM changes. + * 4: ring 0 head pointer + * 5: ring 1 head pointer (915-class) + * 6: ring 2 head pointer (915-class) + * + * The area from dword 0x10 to 0x3ff is available for driver usage. + */ #define READ_HWSP(dev_priv, reg) (((volatile u32*)(dev_priv->hw_status_page))[reg]) +#define I915_GEM_HWS_INDEX 0x10 #define BLC_PWM_CTL 0x61254 #define BACKLIGHT_MODULATION_FREQ_SHIFT (17) diff --git a/shared-core/i915_irq.c b/shared-core/i915_irq.c index 2287cd0c..37d85f40 100644 --- a/shared-core/i915_irq.c +++ b/shared-core/i915_irq.c @@ -435,6 +435,28 @@ u32 i915_get_vblank_counter(struct drm_device *dev, int plane) return count; } +/** + * Handler for user interrupts in process context (able to sleep, do VFS + * operations, etc. + * + * If another IRQ comes in while we're in this handler, it will still get put + * on the queue again to be rerun when we finish. + */ +void +i915_user_interrupt_handler(struct work_struct *work) +{ + drm_i915_private_t *dev_priv; + struct drm_device *dev; + + dev_priv = container_of(work, drm_i915_private_t, + user_interrupt_task); + dev = dev_priv->dev; + + mutex_lock(&dev->struct_mutex); + i915_gem_retire_requests(dev); + mutex_unlock(&dev->struct_mutex); +} + irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) { struct drm_device *dev = (struct drm_device *) arg; @@ -485,13 +507,14 @@ irqreturn_t i915_driver_irq_handler(DRM_IRQ_ARGS) if (dev_priv->sarea_priv) dev_priv->sarea_priv->last_dispatch = READ_BREADCRUMB(dev_priv); - I915_WRITE(I915REG_INT_IDENTITY_R, iir); + I915_WRITE(I915REG_INT_IDENTITY_R, iir | I915_USER_INTERRUPT); (void) I915_READ(I915REG_INT_IDENTITY_R); /* Flush posted write */ if (iir & I915_USER_INTERRUPT) { DRM_WAKEUP(&dev_priv->irq_queue); #ifdef I915_HAVE_FENCE i915_fence_handler(dev); + schedule_work(&dev_priv->user_interrupt_task); #endif } @@ -526,8 +549,9 @@ void i915_user_irq_on(drm_i915_private_t *dev_priv) { DRM_SPINLOCK(&dev_priv->user_irq_lock); if (dev_priv->irq_enabled && (++dev_priv->user_irq_refcount == 1)){ - dev_priv->irq_enable_reg |= I915_USER_INTERRUPT; - I915_WRITE(I915REG_INT_ENABLE_R, dev_priv->irq_enable_reg); + dev_priv->irq_mask_reg &= ~I915_USER_INTERRUPT; + I915_WRITE(I915REG_INT_MASK_R, dev_priv->irq_mask_reg); + (void) I915_READ (I915REG_INT_ENABLE_R); } DRM_SPINUNLOCK(&dev_priv->user_irq_lock); @@ -536,15 +560,17 @@ void i915_user_irq_on(drm_i915_private_t *dev_priv) void i915_user_irq_off(drm_i915_private_t *dev_priv) { DRM_SPINLOCK(&dev_priv->user_irq_lock); + BUG_ON(dev_priv->user_irq_refcount <= 0); if (dev_priv->irq_enabled && (--dev_priv->user_irq_refcount == 0)) { - // dev_priv->irq_enable_reg &= ~USER_INT_FLAG; - // I915_WRITE(I915REG_INT_ENABLE_R, dev_priv->irq_enable_reg); + dev_priv->irq_mask_reg |= I915_USER_INTERRUPT; + I915_WRITE(I915REG_INT_MASK_R, dev_priv->irq_mask_reg); + (void) I915_READ(I915REG_INT_MASK_R); } DRM_SPINUNLOCK(&dev_priv->user_irq_lock); } -static int i915_wait_irq(struct drm_device * dev, int irq_nr) +int i915_wait_irq(struct drm_device * dev, int irq_nr) { drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private; int ret = 0; @@ -618,16 +644,17 @@ int i915_enable_vblank(struct drm_device *dev, int plane) drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private; int pipe = i915_get_pipe(dev, plane); u32 pipestat_reg = 0; + u32 mask_reg = 0; u32 pipestat; switch (pipe) { case 0: pipestat_reg = I915REG_PIPEASTAT; - dev_priv->irq_enable_reg |= I915_DISPLAY_PIPE_A_EVENT_INTERRUPT; + mask_reg |= I915_DISPLAY_PIPE_A_EVENT_INTERRUPT; break; case 1: pipestat_reg = I915REG_PIPEBSTAT; - dev_priv->irq_enable_reg |= I915_DISPLAY_PIPE_B_EVENT_INTERRUPT; + mask_reg |= I915_DISPLAY_PIPE_B_EVENT_INTERRUPT; break; default: DRM_ERROR("tried to enable vblank on non-existent pipe %d\n", @@ -653,7 +680,11 @@ int i915_enable_vblank(struct drm_device *dev, int plane) I915_VBLANK_INTERRUPT_STATUS); I915_WRITE(pipestat_reg, pipestat); } - I915_WRITE(I915REG_INT_ENABLE_R, dev_priv->irq_enable_reg); + DRM_SPINLOCK(&dev_priv->user_irq_lock); + dev_priv->irq_mask_reg &= ~mask_reg; + I915_WRITE(I915REG_INT_MASK_R, dev_priv->irq_mask_reg); + I915_READ(I915REG_INT_MASK_R); + DRM_SPINUNLOCK(&dev_priv->user_irq_lock); return 0; } @@ -663,16 +694,17 @@ void i915_disable_vblank(struct drm_device *dev, int plane) drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private; int pipe = i915_get_pipe(dev, plane); u32 pipestat_reg = 0; + u32 mask_reg = 0; u32 pipestat; switch (pipe) { case 0: pipestat_reg = I915REG_PIPEASTAT; - dev_priv->irq_enable_reg &= ~I915_DISPLAY_PIPE_A_EVENT_INTERRUPT; + mask_reg |= I915_DISPLAY_PIPE_A_EVENT_INTERRUPT; break; case 1: pipestat_reg = I915REG_PIPEBSTAT; - dev_priv->irq_enable_reg &= ~I915_DISPLAY_PIPE_B_EVENT_INTERRUPT; + mask_reg |= I915_DISPLAY_PIPE_B_EVENT_INTERRUPT; break; default: DRM_ERROR("tried to disable vblank on non-existent pipe %d\n", @@ -680,7 +712,11 @@ void i915_disable_vblank(struct drm_device *dev, int plane) break; } - I915_WRITE(I915REG_INT_ENABLE_R, dev_priv->irq_enable_reg); + DRM_SPINLOCK(&dev_priv->user_irq_lock); + dev_priv->irq_mask_reg |= mask_reg; + I915_WRITE(I915REG_INT_MASK_R, dev_priv->irq_mask_reg); + (void) I915_READ (I915REG_INT_MASK_R); + DRM_SPINUNLOCK(&dev_priv->user_irq_lock); if (pipestat_reg) { pipestat = I915_READ (pipestat_reg); @@ -692,6 +728,7 @@ void i915_disable_vblank(struct drm_device *dev, int plane) pipestat |= (I915_START_VBLANK_INTERRUPT_STATUS | I915_VBLANK_INTERRUPT_STATUS); I915_WRITE(pipestat_reg, pipestat); + (void) I915_READ(pipestat_reg); } } @@ -699,12 +736,27 @@ static void i915_enable_interrupt (struct drm_device *dev) { drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private; - dev_priv->irq_enable_reg |= I915_USER_INTERRUPT; - - I915_WRITE(I915REG_INT_ENABLE_R, dev_priv->irq_enable_reg); + dev_priv->irq_mask_reg = (I915_USER_INTERRUPT | + I915_DISPLAY_PIPE_A_EVENT_INTERRUPT | + I915_DISPLAY_PIPE_B_EVENT_INTERRUPT); + I915_WRITE(I915REG_INT_MASK_R, dev_priv->irq_mask_reg); + I915_WRITE(I915REG_INT_ENABLE_R, dev_priv->irq_mask_reg); + (void) I915_READ (I915REG_INT_ENABLE_R); dev_priv->irq_enabled = 1; } +static void i915_disable_interrupt (struct drm_device *dev) +{ + drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private; + + I915_WRITE(I915REG_HWSTAM, 0xffffffff); + I915_WRITE(I915REG_INT_MASK_R, 0xffffffff); + I915_WRITE(I915REG_INT_ENABLE_R, 0); + I915_WRITE(I915REG_INT_IDENTITY_R, 0xffffffff); + (void) I915_READ (I915REG_INT_IDENTITY_R); + dev_priv->irq_enabled = 0; +} + /* Set the vblank monitor pipe */ int i915_vblank_pipe_set(struct drm_device *dev, void *data, @@ -910,9 +962,11 @@ void i915_driver_irq_preinstall(struct drm_device * dev) { drm_i915_private_t *dev_priv = (drm_i915_private_t *) dev->dev_private; - I915_WRITE16(I915REG_HWSTAM, 0xeffe); - I915_WRITE16(I915REG_INT_MASK_R, 0x0); - I915_WRITE16(I915REG_INT_ENABLE_R, 0x0); + I915_WRITE(I915REG_HWSTAM, 0xffff); + I915_WRITE(I915REG_INT_ENABLE_R, 0x0); + I915_WRITE(I915REG_INT_MASK_R, 0xffffffff); + I915_WRITE(I915REG_INT_IDENTITY_R, 0xffffffff); + (void) I915_READ(I915REG_INT_IDENTITY_R); } int i915_driver_irq_postinstall(struct drm_device * dev) @@ -926,7 +980,7 @@ int i915_driver_irq_postinstall(struct drm_device * dev) DRM_SPININIT(&dev_priv->user_irq_lock, "userirq"); dev_priv->user_irq_refcount = 0; - dev_priv->irq_enable_reg = 0; + dev_priv->irq_mask_reg = 0; ret = drm_vblank_init(dev, num_pipes); if (ret) @@ -953,15 +1007,10 @@ void i915_driver_irq_uninstall(struct drm_device * dev) if (!dev_priv) return; - dev_priv->irq_enabled = 0; - I915_WRITE(I915REG_HWSTAM, 0xffffffff); - I915_WRITE(I915REG_INT_MASK_R, 0xffffffff); - I915_WRITE(I915REG_INT_ENABLE_R, 0x0); + i915_disable_interrupt (dev); temp = I915_READ(I915REG_PIPEASTAT); I915_WRITE(I915REG_PIPEASTAT, temp); temp = I915_READ(I915REG_PIPEBSTAT); I915_WRITE(I915REG_PIPEBSTAT, temp); - temp = I915_READ(I915REG_INT_IDENTITY_R); - I915_WRITE(I915REG_INT_IDENTITY_R, temp); } diff --git a/tests/Makefile.am b/tests/Makefile.am index dce1754e..718cc436 100644 --- a/tests/Makefile.am +++ b/tests/Makefile.am @@ -22,7 +22,10 @@ TESTS = auth \ getstats \ lock \ setversion \ - updatedraw + updatedraw \ + gem_basic \ + gem_readwrite \ + gem_mmap EXTRA_PROGRAMS = $(TESTS) CLEANFILES = $(EXTRA_PROGRAMS) $(EXTRA_LTLIBRARIES) diff --git a/tests/drmtest.c b/tests/drmtest.c index cae99a0c..5453b105 100644 --- a/tests/drmtest.c +++ b/tests/drmtest.c @@ -26,6 +26,7 @@ */ #include <fcntl.h> +#include <sys/stat.h> #include "drmtest.h" /** Open the first DRM device we can find, searching up to 16 device nodes */ @@ -80,4 +81,3 @@ int drm_open_any_master(void) fprintf(stderr, "Couldn't find an un-controlled DRM device\n"); abort(); } - diff --git a/tests/gem_basic.c b/tests/gem_basic.c new file mode 100644 index 00000000..8b8b63d0 --- /dev/null +++ b/tests/gem_basic.c @@ -0,0 +1,97 @@ +/* + * Copyright © 2008 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt <eric@anholt.net> + * + */ + +#include <stdlib.h> +#include <stdio.h> +#include <string.h> +#include <assert.h> +#include <fcntl.h> +#include <inttypes.h> +#include <errno.h> +#include <sys/stat.h> +#include "drm.h" + +static void +test_bad_close(int fd) +{ + struct drm_gem_close close; + int ret; + + printf("Testing error return on bad close ioctl.\n"); + + close.handle = 0x10101010; + ret = ioctl(fd, DRM_IOCTL_GEM_CLOSE, &close); + + assert(ret == -1 && errno == EINVAL); +} + +static void +test_create_close(int fd) +{ + struct drm_gem_create create; + struct drm_gem_close close; + int ret; + + printf("Testing creating and closing an object.\n"); + + memset(&create, 0, sizeof(create)); + create.size = 16 * 1024; + ret = ioctl(fd, DRM_IOCTL_GEM_CREATE, &create); + assert(ret == 0); + + close.handle = create.handle; + ret = ioctl(fd, DRM_IOCTL_GEM_CLOSE, &close); +} + +static void +test_create_fd_close(int fd) +{ + struct drm_gem_create create; + int ret; + + printf("Testing closing with an object allocated.\n"); + + memset(&create, 0, sizeof(create)); + create.size = 16 * 1024; + ret = ioctl(fd, DRM_IOCTL_GEM_CREATE, &create); + assert(ret == 0); + + close(fd); +} + +int main(int argc, char **argv) +{ + int fd; + + fd = drm_open_any(); + + test_bad_close(fd); + test_create_close(fd); + test_create_fd_close(fd); + + return 0; +} diff --git a/tests/gem_mmap.c b/tests/gem_mmap.c new file mode 100644 index 00000000..3f8e27a0 --- /dev/null +++ b/tests/gem_mmap.c @@ -0,0 +1,131 @@ +/* + * Copyright © 2008 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt <eric@anholt.net> + * + */ + +#include <stdlib.h> +#include <stdio.h> +#include <string.h> +#include <assert.h> +#include <fcntl.h> +#include <inttypes.h> +#include <errno.h> +#include <sys/stat.h> +#include "drm.h" + +#define OBJECT_SIZE 16384 + +int do_read(int fd, int handle, void *buf, int offset, int size) +{ + struct drm_gem_pread read; + + /* Ensure that we don't have any convenient data in buf in case + * we fail. + */ + memset(buf, 0xd0, size); + + memset(&read, 0, sizeof(read)); + read.handle = handle; + read.data_ptr = (uintptr_t)buf; + read.size = size; + read.offset = offset; + + return ioctl(fd, DRM_IOCTL_GEM_PREAD, &read); +} + +int do_write(int fd, int handle, void *buf, int offset, int size) +{ + struct drm_gem_pwrite write; + + memset(&write, 0, sizeof(write)); + write.handle = handle; + write.data_ptr = (uintptr_t)buf; + write.size = size; + write.offset = offset; + + return ioctl(fd, DRM_IOCTL_GEM_PWRITE, &write); +} + +int main(int argc, char **argv) +{ + int fd; + struct drm_gem_create create; + struct drm_gem_mmap mmap; + struct drm_gem_close unref; + uint8_t expected[OBJECT_SIZE]; + uint8_t buf[OBJECT_SIZE]; + uint8_t *addr; + int ret; + int handle; + + fd = drm_open_any(); + + memset(&mmap, 0, sizeof(mmap)); + mmap.handle = 0x10101010; + mmap.offset = 0; + mmap.size = 4096; + printf("Testing mmaping of bad object.\n"); + ret = ioctl(fd, DRM_IOCTL_GEM_MMAP, &mmap); + assert(ret == -1 && errno == EINVAL); + + memset(&create, 0, sizeof(create)); + create.size = OBJECT_SIZE; + ret = ioctl(fd, DRM_IOCTL_GEM_CREATE, &create); + assert(ret == 0); + handle = create.handle; + + printf("Testing mmaping of newly created object.\n"); + mmap.handle = handle; + mmap.offset = 0; + mmap.size = OBJECT_SIZE; + ret = ioctl(fd, DRM_IOCTL_GEM_MMAP, &mmap); + assert(ret == 0); + addr = (uint8_t *)(uintptr_t)mmap.addr_ptr; + + printf("Testing contents of newly created object.\n"); + memset(expected, 0, sizeof(expected)); + assert(memcmp(addr, expected, sizeof(expected)) == 0); + + printf("Testing coherency of writes and mmap reads.\n"); + memset(buf, 0, sizeof(buf)); + memset(buf + 1024, 0x01, 1024); + memset(expected + 1024, 0x01, 1024); + ret = do_write(fd, handle, buf, 0, OBJECT_SIZE); + assert(ret == 0); + assert(memcmp(buf, addr, sizeof(buf)) == 0); + + printf("Testing that mapping stays after close\n"); + unref.handle = handle; + ret = ioctl(fd, DRM_IOCTL_GEM_CLOSE, &unref); + assert(ret == 0); + assert(memcmp(buf, addr, sizeof(buf)) == 0); + + printf("Testing unmapping\n"); + munmap(addr, OBJECT_SIZE); + + close(fd); + + return 0; +} diff --git a/tests/gem_readwrite.c b/tests/gem_readwrite.c new file mode 100644 index 00000000..a48f9847 --- /dev/null +++ b/tests/gem_readwrite.c @@ -0,0 +1,125 @@ +/* + * Copyright © 2008 Intel Corporation + * + * Permission is hereby granted, free of charge, to any person obtaining a + * copy of this software and associated documentation files (the "Software"), + * to deal in the Software without restriction, including without limitation + * the rights to use, copy, modify, merge, publish, distribute, sublicense, + * and/or sell copies of the Software, and to permit persons to whom the + * Software is furnished to do so, subject to the following conditions: + * + * The above copyright notice and this permission notice (including the next + * paragraph) shall be included in all copies or substantial portions of the + * Software. + * + * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR + * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, + * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL + * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER + * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING + * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS + * IN THE SOFTWARE. + * + * Authors: + * Eric Anholt <eric@anholt.net> + * + */ + +#include <stdlib.h> +#include <stdio.h> +#include <string.h> +#include <assert.h> +#include <fcntl.h> +#include <inttypes.h> +#include <errno.h> +#include <sys/stat.h> +#include "drm.h" + +#define OBJECT_SIZE 16384 + +int do_read(int fd, int handle, void *buf, int offset, int size) +{ + struct drm_gem_pread read; + + /* Ensure that we don't have any convenient data in buf in case + * we fail. + */ + memset(buf, 0xd0, size); + + memset(&read, 0, sizeof(read)); + read.handle = handle; + read.data_ptr = (uintptr_t)buf; + read.size = size; + read.offset = offset; + + return ioctl(fd, DRM_IOCTL_GEM_PREAD, &read); +} + +int do_write(int fd, int handle, void *buf, int offset, int size) +{ + struct drm_gem_pwrite write; + + memset(&write, 0, sizeof(write)); + write.handle = handle; + write.data_ptr = (uintptr_t)buf; + write.size = size; + write.offset = offset; + + return ioctl(fd, DRM_IOCTL_GEM_PWRITE, &write); +} + +int main(int argc, char **argv) +{ + int fd; + struct drm_gem_create create; + uint8_t expected[OBJECT_SIZE]; + uint8_t buf[OBJECT_SIZE]; + int ret; + int handle; + + fd = drm_open_any(); + + memset(&create, 0, sizeof(create)); + create.size = OBJECT_SIZE; + ret = ioctl(fd, DRM_IOCTL_GEM_CREATE, &create); + assert(ret == 0); + handle = create.handle; + + printf("Testing contents of newly created object.\n"); + ret = do_read(fd, handle, buf, 0, OBJECT_SIZE); + assert(ret == 0); + memset(&expected, 0, sizeof(expected)); + assert(memcmp(expected, buf, sizeof(expected)) == 0); + + printf("Testing read beyond end of buffer.\n"); + ret = do_read(fd, handle, buf, OBJECT_SIZE / 2, OBJECT_SIZE); + assert(ret == -1 && errno == EINVAL); + + printf("Testing full write of buffer\n"); + memset(buf, 0, sizeof(buf)); + memset(buf + 1024, 0x01, 1024); + memset(expected + 1024, 0x01, 1024); + ret = do_write(fd, handle, buf, 0, OBJECT_SIZE); + assert(ret == 0); + ret = do_read(fd, handle, buf, 0, OBJECT_SIZE); + assert(ret == 0); + assert(memcmp(buf, expected, sizeof(buf)) == 0); + + printf("Testing partial write of buffer\n"); + memset(buf + 4096, 0x02, 1024); + memset(expected + 4096, 0x02, 1024); + ret = do_write(fd, handle, buf + 4096, 4096, 1024); + assert(ret == 0); + ret = do_read(fd, handle, buf, 0, OBJECT_SIZE); + assert(ret == 0); + assert(memcmp(buf, expected, sizeof(buf)) == 0); + + printf("Testing partial read of buffer\n"); + ret = do_read(fd, handle, buf, 512, 1024); + assert(ret == 0); + assert(memcmp(buf, expected + 512, 1024) == 0); + + close(fd); + + return 0; +} |