From f4166c54bfe04f64603974058e44fbd7cfef0ccc Mon Sep 17 00:00:00 2001 From: Markus Metzger Date: Sun, 9 Nov 2008 14:29:21 +0100 Subject: x86, bts: DS and BTS initialization Impact: widen BTS/PEBS ptrace enablement to more CPU models Move BTS initialisation out of an #ifdef CONFIG_X86_64 guard. Assume core2 BTS and DS layout for future models of family 6 processors. Signed-off-by: Markus Metzger Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/intel.c | 3 +-- arch/x86/kernel/ds.c | 9 ++++----- arch/x86/kernel/ptrace.c | 9 ++++----- 3 files changed, 9 insertions(+), 12 deletions(-) (limited to 'arch/x86/kernel') diff --git a/arch/x86/kernel/cpu/intel.c b/arch/x86/kernel/cpu/intel.c index cce0b6118d55..816f27f289b1 100644 --- a/arch/x86/kernel/cpu/intel.c +++ b/arch/x86/kernel/cpu/intel.c @@ -307,12 +307,11 @@ static void __cpuinit init_intel(struct cpuinfo_x86 *c) set_cpu_cap(c, X86_FEATURE_P4); if (c->x86 == 6) set_cpu_cap(c, X86_FEATURE_P3); +#endif if (cpu_has_bts) ptrace_bts_init_intel(c); -#endif - detect_extended_topology(c); if (!cpu_has(c, X86_FEATURE_XTOPOLOGY)) { /* diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c index 2b69994fd3a8..c570252905a1 100644 --- a/arch/x86/kernel/ds.c +++ b/arch/x86/kernel/ds.c @@ -821,17 +821,16 @@ void __cpuinit ds_init_intel(struct cpuinfo_x86 *c) switch (c->x86) { case 0x6: switch (c->x86_model) { + case 0 ... 0xC: + /* sorry, don't know about them */ + break; case 0xD: case 0xE: /* Pentium M */ ds_configure(&ds_cfg_var); break; - case 0xF: /* Core2 */ - case 0x1C: /* Atom */ + default: /* Core2, Atom, ... */ ds_configure(&ds_cfg_64); break; - default: - /* sorry, don't know about them */ - break; } break; case 0xF: diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c index 0a6d8c12e10d..06180dff5b2e 100644 --- a/arch/x86/kernel/ptrace.c +++ b/arch/x86/kernel/ptrace.c @@ -929,17 +929,16 @@ void __cpuinit ptrace_bts_init_intel(struct cpuinfo_x86 *c) switch (c->x86) { case 0x6: switch (c->x86_model) { + case 0 ... 0xC: + /* sorry, don't know about them */ + break; case 0xD: case 0xE: /* Pentium M */ bts_configure(&bts_cfg_pentium_m); break; - case 0xF: /* Core2 */ - case 0x1C: /* Atom */ + default: /* Core2, Atom, ... */ bts_configure(&bts_cfg_core2); break; - default: - /* sorry, don't know about them */ - break; } break; case 0xF: -- cgit v1.2.2 From 9bc646f163b136684390081262fab0fd8f5343ca Mon Sep 17 00:00:00 2001 From: Rakib Mullick Date: Thu, 20 Nov 2008 19:08:45 +0600 Subject: x86: fix __cpuinit/__init tangle in init_thread_xstate() Impact: fix incorrect __init annotation This patch removes the following section mismatch warning. A patch set was send previously (http://lkml.org/lkml/2008/11/10/407). But introduce some other problem, reported by Rufus (http://lkml.org/lkml/2008/11/11/46). Then Ingo Molnar suggest that, it's best to remove __init from xsave_cntxt_init(void). Which is the second patch in this series. Now, this one removes the following warning. WARNING: arch/x86/kernel/built-in.o(.cpuinit.text+0x2237): Section mismatch in reference from the function cpu_init() to the function .init.text:init_thread_xstate() The function __cpuinit cpu_init() references a function __init init_thread_xstate(). If init_thread_xstate is only used by cpu_init then annotate init_thread_xstate with a matching annotation. Signed-off-by: Rakib Mullick Signed-off-by: Ingo Molnar --- arch/x86/kernel/i387.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'arch/x86/kernel') diff --git a/arch/x86/kernel/i387.c b/arch/x86/kernel/i387.c index 1f20608d4ca8..b0f61f0dcd0a 100644 --- a/arch/x86/kernel/i387.c +++ b/arch/x86/kernel/i387.c @@ -58,7 +58,7 @@ void __cpuinit mxcsr_feature_mask_init(void) stts(); } -void __init init_thread_xstate(void) +void __cpuinit init_thread_xstate(void) { if (!HAVE_HWFP) { xstate_size = sizeof(struct i387_soft_struct); -- cgit v1.2.2 From bfe085f62f98a49e1b864e4950389c7205174e4f Mon Sep 17 00:00:00 2001 From: Rakib Mullick Date: Thu, 20 Nov 2008 19:12:50 +0600 Subject: x86: fixing __cpuinit/__init tangle, xsave_cntxt_init() Annotate xsave_cntxt_init() as "can be called outside of __init". Signed-off-by: Rakib Mullick Signed-off-by: Ingo Molnar --- arch/x86/kernel/xsave.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'arch/x86/kernel') diff --git a/arch/x86/kernel/xsave.c b/arch/x86/kernel/xsave.c index b13acb75e822..15c3e6999182 100644 --- a/arch/x86/kernel/xsave.c +++ b/arch/x86/kernel/xsave.c @@ -310,7 +310,7 @@ static void __init setup_xstate_init(void) /* * Enable and initialize the xsave feature. */ -void __init xsave_cntxt_init(void) +void __ref xsave_cntxt_init(void) { unsigned int eax, ebx, ecx, edx; -- cgit v1.2.2 From a1967d64414dab500e86cbbddf8eae6ad2047903 Mon Sep 17 00:00:00 2001 From: Thomas Gleixner Date: Fri, 21 Nov 2008 11:16:48 -0800 Subject: x86: revert irq number limitation Impact: fix MSIx not enough irq numbers available regression The manual revert of the sparse_irq patches missed to bring the number of possible irqs back to the .27 status. This resulted in a regression when two multichannel network cards were placed in a system with only one IO_APIC - causing the networking driver to not have the right IRQ and the device not coming up. Remove the dynamic allocation logic leftovers and simply return NR_IRQS in probe_nr_irqs() for now. Fixes: http://lkml.org/lkml/2008/11/19/354 Reported-by: Jesper Dangaard Brouer Signed-off-by: Thomas Gleixner Tested-by: Jesper Dangaard Brouer Acked-by: Yinghai Lu Signed-off-by: Ingo Molnar --- arch/x86/kernel/io_apic.c | 22 +--------------------- 1 file changed, 1 insertion(+), 21 deletions(-) (limited to 'arch/x86/kernel') diff --git a/arch/x86/kernel/io_apic.c b/arch/x86/kernel/io_apic.c index c9513e1ff28d..1fec0f9b1508 100644 --- a/arch/x86/kernel/io_apic.c +++ b/arch/x86/kernel/io_apic.c @@ -3608,27 +3608,7 @@ int __init io_apic_get_redir_entries (int ioapic) int __init probe_nr_irqs(void) { - int idx; - int nr = 0; -#ifndef CONFIG_XEN - int nr_min = 32; -#else - int nr_min = NR_IRQS; -#endif - - for (idx = 0; idx < nr_ioapics; idx++) - nr += io_apic_get_redir_entries(idx) + 1; - - /* double it for hotplug and msi and nmi */ - nr <<= 1; - - /* something wrong ? */ - if (nr < nr_min) - nr = nr_min; - if (WARN_ON(nr > NR_IRQS)) - nr = NR_IRQS; - - return nr; + return NR_IRQS; } /* -------------------------------------------------------------------------- -- cgit v1.2.2 From eff79aee91dd07e944df65fa448c8baeee7709d8 Mon Sep 17 00:00:00 2001 From: Julia Lawall Date: Tue, 25 Nov 2008 14:13:03 +0100 Subject: arch/x86/kernel/pci-calgary_64.c: change simple_strtol to simple_strtoul Impact: fix theoretical option string parsing overflow Since bridge is unsigned, it would seem better to use simple_strtoul that simple_strtol. A simplified version of the semantic patch that makes this change is as follows: (http://www.emn.fr/x-info/coccinelle/) // @r2@ long e; position p; @@ e = simple_strtol@p(...) @@ position p != r2.p; type T; T e; @@ e = - simple_strtol@p + simple_strtoul (...) // Signed-off-by: Julia Lawall Cc: muli@il.ibm.com Cc: jdmason@kudzu.us Cc: discuss@x86-64.org Signed-off-by: Ingo Molnar --- arch/x86/kernel/pci-calgary_64.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) (limited to 'arch/x86/kernel') diff --git a/arch/x86/kernel/pci-calgary_64.c b/arch/x86/kernel/pci-calgary_64.c index e1e731d78f38..d28bbdc35e4e 100644 --- a/arch/x86/kernel/pci-calgary_64.c +++ b/arch/x86/kernel/pci-calgary_64.c @@ -1567,7 +1567,7 @@ static int __init calgary_parse_options(char *p) ++p; if (*p == '\0') break; - bridge = simple_strtol(p, &endp, 0); + bridge = simple_strtoul(p, &endp, 0); if (p == endp) break; -- cgit v1.2.2 From 292c669cd7087a090d6420e223eb1072f3e3c50b Mon Sep 17 00:00:00 2001 From: Markus Metzger Date: Tue, 25 Nov 2008 08:45:13 +0100 Subject: x86, bts: exclude ds.c from build when disabled Impact: cleanup Move the CONFIG guard from the .c file into the makefile. Reported-by: Andi Kleen Signed-off-by: Markus Metzger Signed-off-by: Ingo Molnar --- arch/x86/kernel/Makefile | 2 +- arch/x86/kernel/ds.c | 3 --- 2 files changed, 1 insertion(+), 4 deletions(-) (limited to 'arch/x86/kernel') diff --git a/arch/x86/kernel/Makefile b/arch/x86/kernel/Makefile index e489ff9cb3e2..b62a7667828e 100644 --- a/arch/x86/kernel/Makefile +++ b/arch/x86/kernel/Makefile @@ -41,7 +41,7 @@ obj-$(CONFIG_X86_TRAMPOLINE) += trampoline.o obj-y += process.o obj-y += i387.o xsave.o obj-y += ptrace.o -obj-y += ds.o +obj-$(CONFIG_X86_DS) += ds.o obj-$(CONFIG_X86_32) += tls.o obj-$(CONFIG_IA32_EMULATION) += tls.o obj-y += step.o diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c index d1a121443bde..4c8d57ec9663 100644 --- a/arch/x86/kernel/ds.c +++ b/arch/x86/kernel/ds.c @@ -21,8 +21,6 @@ */ -#ifdef CONFIG_X86_DS - #include #include @@ -878,4 +876,3 @@ void ds_free(struct ds_context *context) while (leftovers--) ds_put_context(context); } -#endif /* CONFIG_X86_DS */ -- cgit v1.2.2 From c4858ffc8f2dc850cb1f609c679b1ac1ad36ef0c Mon Sep 17 00:00:00 2001 From: Markus Metzger Date: Tue, 25 Nov 2008 08:49:06 +0100 Subject: x86, pebs: fix PEBS record size configuration Impact: fix DS hw enablement on 64-bit x86 Fix the PEBS record size in the DS configuration. Reported-by: Stephane Eranian Signed-off-by: Markus Metzger Signed-off-by: Ingo Molnar --- arch/x86/kernel/ds.c | 8 ++++++++ 1 file changed, 8 insertions(+) (limited to 'arch/x86/kernel') diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c index 4c8d57ec9663..04e38ef646af 100644 --- a/arch/x86/kernel/ds.c +++ b/arch/x86/kernel/ds.c @@ -816,13 +816,21 @@ static const struct ds_configuration ds_cfg_var = { .sizeof_ds = sizeof(long) * 12, .sizeof_field = sizeof(long), .sizeof_rec[ds_bts] = sizeof(long) * 3, +#ifdef __i386__ .sizeof_rec[ds_pebs] = sizeof(long) * 10 +#else + .sizeof_rec[ds_pebs] = sizeof(long) * 18 +#endif }; static const struct ds_configuration ds_cfg_64 = { .sizeof_ds = 8 * 12, .sizeof_field = 8, .sizeof_rec[ds_bts] = 8 * 3, +#ifdef __i386__ .sizeof_rec[ds_pebs] = 8 * 10 +#else + .sizeof_rec[ds_pebs] = 8 * 18 +#endif }; static inline void -- cgit v1.2.2 From de90add30e79261c3b5be68bb0f22d2ef98e8113 Mon Sep 17 00:00:00 2001 From: Markus Metzger Date: Tue, 25 Nov 2008 08:52:56 +0100 Subject: x86, bts: fix wrmsr and spinlock over kmalloc Impact: fix sleeping-with-spinlock-held bugs/crashes - Turn a wrmsr to write the DS_AREA MSR into a wrmsrl. - Use irqsave variants of spinlocks. - Do not allocate memory while holding spinlocks. Reported-by: Stephane Eranian Reported-by: Ingo Molnar Signed-off-by: Markus Metzger Signed-off-by: Ingo Molnar --- arch/x86/kernel/ds.c | 77 +++++++++++++++++++++++++++------------------------- 1 file changed, 40 insertions(+), 37 deletions(-) (limited to 'arch/x86/kernel') diff --git a/arch/x86/kernel/ds.c b/arch/x86/kernel/ds.c index 04e38ef646af..a2d1176c38ee 100644 --- a/arch/x86/kernel/ds.c +++ b/arch/x86/kernel/ds.c @@ -209,14 +209,15 @@ static DEFINE_PER_CPU(struct ds_context *, system_context); static inline struct ds_context *ds_get_context(struct task_struct *task) { struct ds_context *context; + unsigned long irq; - spin_lock(&ds_lock); + spin_lock_irqsave(&ds_lock, irq); context = (task ? task->thread.ds_ctx : this_system_context); if (context) context->count++; - spin_unlock(&ds_lock); + spin_unlock_irqrestore(&ds_lock, irq); return context; } @@ -224,55 +225,46 @@ static inline struct ds_context *ds_get_context(struct task_struct *task) /* * Same as ds_get_context, but allocates the context and it's DS * structure, if necessary; returns NULL; if out of memory. - * - * pre: requires ds_lock to be held */ static inline struct ds_context *ds_alloc_context(struct task_struct *task) { struct ds_context **p_context = (task ? &task->thread.ds_ctx : &this_system_context); struct ds_context *context = *p_context; + unsigned long irq; if (!context) { - spin_unlock(&ds_lock); - context = kzalloc(sizeof(*context), GFP_KERNEL); - - if (!context) { - spin_lock(&ds_lock); + if (!context) return NULL; - } context->ds = kzalloc(ds_cfg.sizeof_ds, GFP_KERNEL); if (!context->ds) { kfree(context); - spin_lock(&ds_lock); return NULL; } - spin_lock(&ds_lock); - /* - * Check for race - another CPU could have allocated - * it meanwhile: - */ + spin_lock_irqsave(&ds_lock, irq); + if (*p_context) { kfree(context->ds); kfree(context); - return *p_context; - } - - *p_context = context; - context->this = p_context; - context->task = task; + context = *p_context; + } else { + *p_context = context; - if (task) - set_tsk_thread_flag(task, TIF_DS_AREA_MSR); + context->this = p_context; + context->task = task; - if (!task || (task == current)) - wrmsr(MSR_IA32_DS_AREA, (unsigned long)context->ds, 0); + if (task) + set_tsk_thread_flag(task, TIF_DS_AREA_MSR); - get_tracer(task); + if (!task || (task == current)) + wrmsrl(MSR_IA32_DS_AREA, + (unsigned long)context->ds); + } + spin_unlock_irqrestore(&ds_lock, irq); } context->count++; @@ -286,10 +278,12 @@ static inline struct ds_context *ds_alloc_context(struct task_struct *task) */ static inline void ds_put_context(struct ds_context *context) { + unsigned long irq; + if (!context) return; - spin_lock(&ds_lock); + spin_lock_irqsave(&ds_lock, irq); if (--context->count) goto out; @@ -311,7 +305,7 @@ static inline void ds_put_context(struct ds_context *context) kfree(context->ds); kfree(context); out: - spin_unlock(&ds_lock); + spin_unlock_irqrestore(&ds_lock, irq); } @@ -382,6 +376,7 @@ static int ds_request(struct task_struct *task, void *base, size_t size, struct ds_context *context; unsigned long buffer, adj; const unsigned long alignment = (1 << 3); + unsigned long irq; int error = 0; if (!ds_cfg.sizeof_ds) @@ -396,26 +391,27 @@ static int ds_request(struct task_struct *task, void *base, size_t size, return -EOPNOTSUPP; - spin_lock(&ds_lock); - - error = -ENOMEM; context = ds_alloc_context(task); if (!context) - goto out_unlock; + return -ENOMEM; + + spin_lock_irqsave(&ds_lock, irq); error = -EPERM; if (!check_tracer(task)) goto out_unlock; + get_tracer(task); + error = -EALREADY; if (context->owner[qual] == current) - goto out_unlock; + goto out_put_tracer; error = -EPERM; if (context->owner[qual] != NULL) - goto out_unlock; + goto out_put_tracer; context->owner[qual] = current; - spin_unlock(&ds_lock); + spin_unlock_irqrestore(&ds_lock, irq); error = -ENOMEM; @@ -463,10 +459,17 @@ static int ds_request(struct task_struct *task, void *base, size_t size, out_release: context->owner[qual] = NULL; ds_put_context(context); + put_tracer(task); + return error; + + out_put_tracer: + spin_unlock_irqrestore(&ds_lock, irq); + ds_put_context(context); + put_tracer(task); return error; out_unlock: - spin_unlock(&ds_lock); + spin_unlock_irqrestore(&ds_lock, irq); ds_put_context(context); return error; } -- cgit v1.2.2