From 10617bbe84628eb18ab5f723d3ba35005adde143 Mon Sep 17 00:00:00 2001 From: Tony Luck Date: Tue, 12 Aug 2008 10:34:20 -0700 Subject: [IA64] Ensure cpu0 can access per-cpu variables in early boot code ia64 handles per-cpu variables a litle differently from other architectures in that it maps the physical memory allocated for each cpu at a constant virtual address (0xffffffffffff0000). This mapping is not enabled until the architecture specific cpu_init() function is run, which causes problems since some generic code is run before this point. In particular when CONFIG_PRINTK_TIME is enabled, the boot cpu will trap on the access to per-cpu memory at the first printk() call so the boot will fail without the kernel printing anything to the console. Fix this by allocating percpu memory for cpu0 in the kernel data section and doing all initialization to enable percpu access in head.S before calling any generic code. Other cpus must take care not to access per-cpu variables too early, but their code path from start_secondary() to cpu_init() is all in arch/ia64 Signed-off-by: Tony Luck --- arch/ia64/kernel/head.S | 26 +++++++++++++++++++++++++- arch/ia64/kernel/setup.c | 18 ++++++++++-------- arch/ia64/kernel/smpboot.c | 2 ++ arch/ia64/kernel/vmlinux.lds.S | 3 +++ 4 files changed, 40 insertions(+), 9 deletions(-) (limited to 'arch/ia64/kernel') diff --git a/arch/ia64/kernel/head.S b/arch/ia64/kernel/head.S index 41c712917ff..8bdea8eb62e 100644 --- a/arch/ia64/kernel/head.S +++ b/arch/ia64/kernel/head.S @@ -359,7 +359,31 @@ start_ap: mov ar.rsc=0 // place RSE in enforced lazy mode ;; loadrs // clear the dirty partition - mov IA64_KR(PER_CPU_DATA)=r0 // clear physical per-CPU base + movl r19=__phys_per_cpu_start + mov r18=PERCPU_PAGE_SIZE + ;; +#ifndef CONFIG_SMP + add r19=r19,r18 + ;; +#else +(isAP) br.few 2f + mov r20=r19 + sub r19=r19,r18 + ;; + shr.u r18=r18,3 +1: + ld8 r21=[r20],8;; + st8[r19]=r21,8 + adds r18=-1,r18;; + cmp4.lt p7,p6=0,r18 +(p7) br.cond.dptk.few 1b +2: +#endif + tpa r19=r19 + ;; + .pred.rel.mutex isBP,isAP +(isBP) mov IA64_KR(PER_CPU_DATA)=r19 // per-CPU base for cpu0 +(isAP) mov IA64_KR(PER_CPU_DATA)=r0 // clear physical per-CPU base ;; mov ar.bspstore=r2 // establish the new RSE stack ;; diff --git a/arch/ia64/kernel/setup.c b/arch/ia64/kernel/setup.c index 593279f33e9..c27d5b2c182 100644 --- a/arch/ia64/kernel/setup.c +++ b/arch/ia64/kernel/setup.c @@ -927,17 +927,19 @@ cpu_init (void) if (smp_processor_id() == 0) { cpu_set(0, per_cpu(cpu_sibling_map, 0)); cpu_set(0, cpu_core_map[0]); + } else { + /* + * Set ar.k3 so that assembly code in MCA handler can compute + * physical addresses of per cpu variables with a simple: + * phys = ar.k3 + &per_cpu_var + * and the alt-dtlb-miss handler can set per-cpu mapping into + * the TLB when needed. head.S already did this for cpu0. + */ + ia64_set_kr(IA64_KR_PER_CPU_DATA, + ia64_tpa(cpu_data) - (long) __per_cpu_start); } #endif - /* - * We set ar.k3 so that assembly code in MCA handler can compute - * physical addresses of per cpu variables with a simple: - * phys = ar.k3 + &per_cpu_var - */ - ia64_set_kr(IA64_KR_PER_CPU_DATA, - ia64_tpa(cpu_data) - (long) __per_cpu_start); - get_max_cacheline_size(); /* diff --git a/arch/ia64/kernel/smpboot.c b/arch/ia64/kernel/smpboot.c index 03f1a9908af..b39853a292d 100644 --- a/arch/ia64/kernel/smpboot.c +++ b/arch/ia64/kernel/smpboot.c @@ -467,7 +467,9 @@ start_secondary (void *unused) { /* Early console may use I/O ports */ ia64_set_kr(IA64_KR_IO_BASE, __pa(ia64_iobase)); +#ifndef CONFIG_PRINTK_TIME Dprintk("start_secondary: starting CPU 0x%x\n", hard_smp_processor_id()); +#endif efi_map_pal_code(); cpu_init(); preempt_disable(); diff --git a/arch/ia64/kernel/vmlinux.lds.S b/arch/ia64/kernel/vmlinux.lds.S index 5a77206c249..de71da811cd 100644 --- a/arch/ia64/kernel/vmlinux.lds.S +++ b/arch/ia64/kernel/vmlinux.lds.S @@ -215,6 +215,9 @@ SECTIONS /* Per-cpu data: */ percpu : { } :percpu . = ALIGN(PERCPU_PAGE_SIZE); +#ifdef CONFIG_SMP + . = . + PERCPU_PAGE_SIZE; /* cpu0 per-cpu space */ +#endif __phys_per_cpu_start = .; .data.percpu PERCPU_ADDR : AT(__phys_per_cpu_start - LOAD_OFFSET) { -- cgit v1.2.3 From 97653f92c004d322ef27edb168a1b5f31142c6dc Mon Sep 17 00:00:00 2001 From: Robin Holt Date: Fri, 15 Aug 2008 09:21:24 -0500 Subject: [IA64] Shrink shadow_flush_counts to a short array to save 8k of per_cpu area. Making allmodconfig will break the current build. This patch shrinks the per_cpu__shadow_flush_counts from 16k to 8k which frees enough space to allow allmodconfig to successfully complete. Fixes http://bugzilla.kernel.org/show_bug.cgi?id=11338 Signed-off-by: Robin Holt Acked-by: Jack Steiner Signed-off-by: Tony Luck --- arch/ia64/kernel/smp.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) (limited to 'arch/ia64/kernel') diff --git a/arch/ia64/kernel/smp.c b/arch/ia64/kernel/smp.c index 3676468612b..da8f020d82c 100644 --- a/arch/ia64/kernel/smp.c +++ b/arch/ia64/kernel/smp.c @@ -58,7 +58,7 @@ static struct local_tlb_flush_counts { unsigned int count; } __attribute__((__aligned__(32))) local_tlb_flush_counts[NR_CPUS]; -static DEFINE_PER_CPU(unsigned int, shadow_flush_counts[NR_CPUS]) ____cacheline_aligned; +static DEFINE_PER_CPU(unsigned short, shadow_flush_counts[NR_CPUS]) ____cacheline_aligned; #define IPI_CALL_FUNC 0 #define IPI_CPU_STOP 1 @@ -254,7 +254,7 @@ smp_local_flush_tlb(void) void smp_flush_tlb_cpumask(cpumask_t xcpumask) { - unsigned int *counts = __ia64_per_cpu_var(shadow_flush_counts); + unsigned short *counts = __ia64_per_cpu_var(shadow_flush_counts); cpumask_t cpumask = xcpumask; int mycpu, cpu, flush_mycpu = 0; @@ -262,7 +262,7 @@ smp_flush_tlb_cpumask(cpumask_t xcpumask) mycpu = smp_processor_id(); for_each_cpu_mask(cpu, cpumask) - counts[cpu] = local_tlb_flush_counts[cpu].count; + counts[cpu] = local_tlb_flush_counts[cpu].count & 0xffff; mb(); for_each_cpu_mask(cpu, cpumask) { @@ -276,7 +276,7 @@ smp_flush_tlb_cpumask(cpumask_t xcpumask) smp_local_flush_tlb(); for_each_cpu_mask(cpu, cpumask) - while(counts[cpu] == local_tlb_flush_counts[cpu].count) + while(counts[cpu] == (local_tlb_flush_counts[cpu].count & 0xffff)) udelay(FLUSH_DELAY); preempt_enable(); -- cgit v1.2.3 From 42aca483ddee8888926d913b660e1dd38d748ba9 Mon Sep 17 00:00:00 2001 From: Robin Holt Date: Mon, 28 Jul 2008 20:36:50 -0500 Subject: [IA64] Fix ia64 build failure when CONFIG_SFC=m CONFIG_SFC=m uses topology_core_siblings() which, for ia64, expects cpu_core_map to be exported. It is not. This patch exports the needed symbol. Maintainers note: This really looks like the wrong thing to do ... it would be much better for the kernel to export an API to provide drivers like this with data they need (which in the case of this driver seems to be an estimate of the effective parallelism available on the platform). But x86 has exported this forever ... so go with the flow until such an API is defined. Signed-off-by: Ben Hutchings Signed-off-by: Matthew Wilcox Signed-off-by: Robin Holt Signed-off-by: Tony Luck --- arch/ia64/kernel/smpboot.c | 1 + 1 file changed, 1 insertion(+) (limited to 'arch/ia64/kernel') diff --git a/arch/ia64/kernel/smpboot.c b/arch/ia64/kernel/smpboot.c index b39853a292d..bcea81e432f 100644 --- a/arch/ia64/kernel/smpboot.c +++ b/arch/ia64/kernel/smpboot.c @@ -138,6 +138,7 @@ cpumask_t cpu_possible_map = CPU_MASK_NONE; EXPORT_SYMBOL(cpu_possible_map); cpumask_t cpu_core_map[NR_CPUS] __cacheline_aligned; +EXPORT_SYMBOL(cpu_core_map); DEFINE_PER_CPU_SHARED_ALIGNED(cpumask_t, cpu_sibling_map); EXPORT_PER_CPU_SYMBOL(cpu_sibling_map); -- cgit v1.2.3 From deac93df26b20cf8438339b5935b5f5643bc30c9 Mon Sep 17 00:00:00 2001 From: James Bottomley Date: Wed, 3 Sep 2008 20:43:36 -0500 Subject: lib: Correct printk %pF to work on all architectures It was introduced by "vsprintf: add support for '%pS' and '%pF' pointer formats" in commit 0fe1ef24f7bd0020f29ffe287dfdb9ead33ca0b2. However, the current way its coded doesn't work on parisc64. For two reasons: 1) parisc isn't in the #ifdef and 2) parisc has a different format for function descriptors Make dereference_function_descriptor() more accommodating by allowing architecture overrides. I put the three overrides (for parisc64, ppc64 and ia64) in arch/kernel/module.c because that's where the kernel internal linker which knows how to deal with function descriptors sits. Signed-off-by: James Bottomley Acked-by: Benjamin Herrenschmidt Acked-by: Tony Luck Acked-by: Kyle McMartin Signed-off-by: Linus Torvalds --- arch/ia64/kernel/module.c | 12 ++++++++++++ 1 file changed, 12 insertions(+) (limited to 'arch/ia64/kernel') diff --git a/arch/ia64/kernel/module.c b/arch/ia64/kernel/module.c index 29aad349e0c..545626f66a4 100644 --- a/arch/ia64/kernel/module.c +++ b/arch/ia64/kernel/module.c @@ -31,9 +31,11 @@ #include #include #include +#include #include #include +#include #include #define ARCH_MODULE_DEBUG 0 @@ -941,3 +943,13 @@ module_arch_cleanup (struct module *mod) if (mod->arch.core_unw_table) unw_remove_unwind_table(mod->arch.core_unw_table); } + +void *dereference_function_descriptor(void *ptr) +{ + struct fdesc *desc = ptr; + void *p; + + if (!probe_kernel_address(&desc->ip, p)) + ptr = p; + return ptr; +} -- cgit v1.2.3