libdrm: add support for i915 GTT mapping ioctl

Add a drm_intel_gem_bo_map_gtt() function for mapping a buffer object
through the aperture rather than directly to its CPU cacheable memory.

4 files changed, 104 insertions, 1 deletions

diff --git a/libdrm/intel/intel_bufmgr.h b/libdrm/intel/intel_bufmgr.h
index e3af886b..361e4345 100644
--- a/libdrm/intel/intel_bufmgr.h
+++ b/libdrm/intel/intel_bufmgr.h
@@ -48,6 +48,13 @@ struct _drm_intel_bo {
      */
     unsigned long size;
     /**
+     * Alignment requirement for object
+     *
+     * Used for GTT mapping & pinning the object.
+     */
+    unsigned long align;
+
+    /**
      * Card virtual address (offset from the beginning of the aperture) for the
      * object.  Only valid while validated.
      */
@@ -98,6 +105,7 @@ drm_intel_bo *drm_intel_bo_gem_create_from_name(drm_intel_bufmgr *bufmgr,
 						const char *name,
 						unsigned int handle);
 void drm_intel_bufmgr_gem_enable_reuse(drm_intel_bufmgr *bufmgr);
+int drm_intel_gem_bo_map_gtt(drm_intel_bo *bo);
 
 /* drm_intel_bufmgr_fake.c */
 drm_intel_bufmgr *drm_intel_bufmgr_fake_init(int fd,
diff --git a/libdrm/intel/intel_bufmgr_gem.c b/libdrm/intel/intel_bufmgr_gem.c
index 9cb33598..84768b46 100644
--- a/libdrm/intel/intel_bufmgr_gem.c
+++ b/libdrm/intel/intel_bufmgr_gem.c
@@ -39,6 +39,7 @@
 #endif
 
 #include <xf86drm.h>
+#include <fcntl.h>
 #include <stdio.h>
 #include <stdlib.h>
 #include <string.h>
@@ -47,6 +48,8 @@
 #include <pthread.h>
 #include <sys/ioctl.h>
 #include <sys/mman.h>
+#include <sys/stat.h>
+#include <sys/types.h>
 
 #include "errno.h"
 #include "intel_bufmgr.h"
@@ -569,6 +572,84 @@ drm_intel_gem_bo_map(drm_intel_bo *bo, int write_enable)
     return 0;
 }
 
+int
+drm_intel_gem_bo_map_gtt(drm_intel_bo *bo)
+{
+    drm_intel_bufmgr_gem *bufmgr_gem = (drm_intel_bufmgr_gem *)bo->bufmgr;
+    drm_intel_bo_gem *bo_gem = (drm_intel_bo_gem *)bo;
+    struct drm_i915_gem_set_domain set_domain;
+    int ret;
+
+    pthread_mutex_lock(&bufmgr_gem->lock);
+
+    /* Allow recursive mapping. Mesa may recursively map buffers with
+     * nested display loops.
+     */
+    if (!bo_gem->mapped) {
+
+	assert(bo->virtual == NULL);
+
+	DBG("bo_map_gtt: %d (%s)\n", bo_gem->gem_handle, bo_gem->name);
+
+	if (bo_gem->virtual == NULL) {
+		struct drm_i915_gem_mmap_gtt mmap_arg;
+
+		memset(&mmap_arg, 0, sizeof(mmap_arg));
+		mmap_arg.handle = bo_gem->gem_handle;
+
+		/* Get the fake offset back... */
+		ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_MMAP_GTT,
+			    &mmap_arg);
+		if (ret != 0) {
+			fprintf(stderr,
+				"%s:%d: Error preparing buffer map %d (%s): %s .\n",
+				__FILE__, __LINE__,
+				bo_gem->gem_handle, bo_gem->name,
+				strerror(errno));
+			pthread_mutex_unlock(&bufmgr_gem->lock);
+			return ret;
+		}
+
+		/* and mmap it */
+		bo_gem->virtual = mmap(0, bo->size, PROT_READ | PROT_WRITE,
+				       MAP_SHARED, bufmgr_gem->fd,
+				       mmap_arg.offset);
+		if (bo_gem->virtual == MAP_FAILED) {
+			fprintf(stderr,
+				"%s:%d: Error mapping buffer %d (%s): %s .\n",
+				__FILE__, __LINE__,
+				bo_gem->gem_handle, bo_gem->name,
+				strerror(errno));
+			pthread_mutex_unlock(&bufmgr_gem->lock);
+			return errno;
+		}
+	}
+
+	bo->virtual = bo_gem->virtual;
+	bo_gem->mapped = 1;
+	DBG("bo_map: %d (%s) -> %p\n", bo_gem->gem_handle, bo_gem->name,
+	    bo_gem->virtual);
+    }
+
+    /* Now move it to the GTT domain so that the CPU caches are flushed */
+    set_domain.handle = bo_gem->gem_handle;
+    set_domain.read_domains = I915_GEM_DOMAIN_GTT;
+    set_domain.write_domain = I915_GEM_DOMAIN_GTT;
+    do {
+	    ret = ioctl(bufmgr_gem->fd, DRM_IOCTL_I915_GEM_SET_DOMAIN,
+			&set_domain);
+    } while (ret == -1 && errno == EINTR);
+
+    if (ret != 0) {
+	    fprintf (stderr, "%s:%d: Error setting swrast %d: %s\n",
+		     __FILE__, __LINE__, bo_gem->gem_handle, strerror (errno));
+    }
+
+    pthread_mutex_unlock(&bufmgr_gem->lock);
+
+    return 0;
+}
+
 static int
 drm_intel_gem_bo_unmap(drm_intel_bo *bo)
 {
diff --git a/shared-core/drm.h b/shared-core/drm.h
index bb332829..05fc91d8 100644
--- a/shared-core/drm.h
+++ b/shared-core/drm.h
@@ -236,7 +236,7 @@ enum drm_map_type {
 	_DRM_AGP = 3,		  /**< AGP/GART */
 	_DRM_SCATTER_GATHER = 4,  /**< Scatter/gather memory for PCI DMA */
 	_DRM_CONSISTENT = 5,	  /**< Consistent memory for PCI DMA */
-	_DRM_TTM = 6
+	_DRM_GEM = 6
 };
 
 /**
diff --git a/shared-core/i915_drm.h b/shared-core/i915_drm.h
index 9211b390..976ff187 100644
--- a/shared-core/i915_drm.h
+++ b/shared-core/i915_drm.h
@@ -193,6 +193,7 @@ typedef struct drm_i915_sarea {
 #define DRM_I915_GEM_SET_TILING	0x21
 #define DRM_I915_GEM_GET_TILING	0x22
 #define DRM_I915_GEM_GET_APERTURE 0x23
+#define DRM_I915_GEM_MMAP_GTT	0x24
 
 #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)
@@ -224,6 +225,7 @@ typedef struct drm_i915_sarea {
 #define DRM_IOCTL_I915_GEM_PREAD	DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PREAD, struct drm_i915_gem_pread)
 #define DRM_IOCTL_I915_GEM_PWRITE	DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_PWRITE, struct drm_i915_gem_pwrite)
 #define DRM_IOCTL_I915_GEM_MMAP		DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP, struct drm_i915_gem_mmap)
+#define DRM_IOCTL_I915_GEM_MMAP_GTT	DRM_IOWR(DRM_COMMAND_BASE + DRM_I915_GEM_MMAP_GTT, struct drm_i915_gem_mmap_gtt)
 #define DRM_IOCTL_I915_GEM_SET_DOMAIN	DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SET_DOMAIN, struct drm_i915_gem_set_domain)
 #define DRM_IOCTL_I915_GEM_SW_FINISH	DRM_IOW (DRM_COMMAND_BASE + DRM_I915_GEM_SW_FINISH, struct drm_i915_gem_sw_finish)
 #define DRM_IOCTL_I915_GEM_SET_TILING	DRM_IOWR (DRM_COMMAND_BASE + DRM_I915_GEM_SET_TILING, struct drm_i915_gem_set_tiling)
@@ -499,6 +501,18 @@ struct drm_i915_gem_mmap {
 	uint64_t addr_ptr;	/* void *, but pointers are not 32/64 compatible */
 };
 
+struct drm_i915_gem_mmap_gtt {
+	/** Handle for the object being mapped. */
+	uint32_t handle;
+	uint32_t pad;
+	/**
+	 * Fake offset to use for subsequent mmap call
+	 *
+	 * This is a fixed-size type for 32/64 compatibility.
+	 */
+	uint64_t offset;
+};
+
 struct drm_i915_gem_set_domain {
 	/** Handle for the object */
 	uint32_t handle;