From 88d4f0db7fa8785859c1d637f9aac210932b6216 Mon Sep 17 00:00:00 2001 From: Eric Dumazet Date: Tue, 25 Jan 2011 19:40:51 +0100 Subject: [PATCH 1/6] perf: Fix alloc_callchain_buffers() Commit 927c7a9e92c4 ("perf: Fix race in callchains") introduced a mismatch in the sizing of struct callchain_cpus_entries. nr_cpu_ids must be used instead of num_possible_cpus(), or we might get out of bound memory accesses on some machines. Signed-off-by: Eric Dumazet Cc: Frederic Weisbecker Cc: Peter Zijlstra Cc: Arnaldo Carvalho de Melo Cc: David Miller Cc: Stephane Eranian CC: stable@kernel.org LKML-Reference: <1295980851.3588.351.camel@edumazet-laptop> Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-) diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 126a302c481c..852ae8c66502 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1999,8 +1999,7 @@ static int alloc_callchain_buffers(void) * accessed from NMI. Use a temporary manual per cpu allocation * until that gets sorted out. */ - size = sizeof(*entries) + sizeof(struct perf_callchain_entry *) * - num_possible_cpus(); + size = offsetof(struct callchain_cpus_entries, cpu_entries[nr_cpu_ids]); entries = kzalloc(size, GFP_KERNEL); if (!entries) From d038b12c6d773a4b9f69ca5243773bf6314f7ee9 Mon Sep 17 00:00:00 2001 From: Stephane Eranian Date: Tue, 25 Jan 2011 17:32:01 +0200 Subject: [PATCH 2/6] perf: Fix Pentium4 raw event validation This patch fixes some issues with raw event validation on Pentium 4 (Netburst) based processors. As I was testing libpfm4 Netburst support, I ran into two problems in the p4_validate_raw_event() function: - the shared field must be checked ONLY when HT is on - the binding to ESCR register was missing The second item was causing raw events to not be encoded correctly compared to generic PMU events. With this patch, I can now pass Netburst events to libpfm4 examples and get meaningful results: $ task -e global_power_events:running:u noploop 1 noploop for 1 seconds 3,206,304,898 global_power_events:running Signed-off-by: Stephane Eranian Acked-by: Cyrill Gorcunov Cc: peterz@infradead.org Cc: paulus@samba.org Cc: davem@davemloft.net Cc: fweisbec@gmail.com Cc: perfmon2-devel@lists.sf.net Cc: eranian@gmail.com Cc: robert.richter@amd.com Cc: acme@redhat.com Cc: gorcunov@gmail.com Cc: ming.m.lin@intel.com LKML-Reference: <4d3efb2f.1252d80a.1a80.ffffc83f@mx.google.com> Signed-off-by: Ingo Molnar --- arch/x86/kernel/cpu/perf_event_p4.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) diff --git a/arch/x86/kernel/cpu/perf_event_p4.c b/arch/x86/kernel/cpu/perf_event_p4.c index e56b9bfbabd1..f7a0993c1e7c 100644 --- a/arch/x86/kernel/cpu/perf_event_p4.c +++ b/arch/x86/kernel/cpu/perf_event_p4.c @@ -682,7 +682,7 @@ static int p4_validate_raw_event(struct perf_event *event) * if an event is shared accross the logical threads * the user needs special permissions to be able to use it */ - if (p4_event_bind_map[v].shared) { + if (p4_ht_active() && p4_event_bind_map[v].shared) { if (perf_paranoid_cpu() && !capable(CAP_SYS_ADMIN)) return -EACCES; } @@ -727,7 +727,8 @@ static int p4_hw_config(struct perf_event *event) event->hw.config = p4_set_ht_bit(event->hw.config); if (event->attr.type == PERF_TYPE_RAW) { - + struct p4_event_bind *bind; + unsigned int esel; /* * Clear bits we reserve to be managed by kernel itself * and never allowed from a user space @@ -743,6 +744,13 @@ static int p4_hw_config(struct perf_event *event) * bits since we keep additional info here (for cache events and etc) */ event->hw.config |= event->attr.config; + bind = p4_config_get_bind(event->attr.config); + if (!bind) { + rc = -EINVAL; + goto out; + } + esel = P4_OPCODE_ESEL(bind->opcode); + event->hw.config |= p4_config_pack_cccr(P4_CCCR_ESEL(esel)); } rc = x86_setup_perfctr(event); From 4135038a582c20ffdadfcf6564852e0b72a20968 Mon Sep 17 00:00:00 2001 From: Marcin Slusarz Date: Fri, 28 Jan 2011 11:00:31 -0500 Subject: [PATCH 3/6] watchdog: Fix broken nowatchdog logic Passing nowatchdog to kernel disables 2 things: creation of watchdog threads AND initialization of percpu watchdog_hrtimer. As hrtimers are initialized only at boot it's not possible to enable watchdog later - for me all watchdog threads started to eat 100% of CPU time, but they could just crash. Additionally, even if these threads would start properly, watchdog_disable_all_cpus was guarded by no_watchdog check, so you couldn't disable watchdog. To fix this, remove no_watchdog variable and use already existing watchdog_enabled variable. Signed-off-by: Marcin Slusarz [ removed another no_watchdog instance ] Signed-off-by: Don Zickus Cc: Stephane Eranian Cc: Peter Zijlstra Cc: Frederic Weisbecker Cc: LKML-Reference: <1296230433-6261-1-git-send-email-dzickus@redhat.com> Signed-off-by: Ingo Molnar --- kernel/watchdog.c | 20 ++++++-------------- 1 file changed, 6 insertions(+), 14 deletions(-) diff --git a/kernel/watchdog.c b/kernel/watchdog.c index d7ebdf4cea98..d9961ea1c3f4 100644 --- a/kernel/watchdog.c +++ b/kernel/watchdog.c @@ -27,7 +27,7 @@ #include #include -int watchdog_enabled; +int watchdog_enabled = 1; int __read_mostly softlockup_thresh = 60; static DEFINE_PER_CPU(unsigned long, watchdog_touch_ts); @@ -43,9 +43,6 @@ static DEFINE_PER_CPU(unsigned long, hrtimer_interrupts_saved); static DEFINE_PER_CPU(struct perf_event *, watchdog_ev); #endif -static int no_watchdog; - - /* boot commands */ /* * Should we panic when a soft-lockup or hard-lockup occurs: @@ -58,7 +55,7 @@ static int __init hardlockup_panic_setup(char *str) if (!strncmp(str, "panic", 5)) hardlockup_panic = 1; else if (!strncmp(str, "0", 1)) - no_watchdog = 1; + watchdog_enabled = 0; return 1; } __setup("nmi_watchdog=", hardlockup_panic_setup); @@ -77,7 +74,7 @@ __setup("softlockup_panic=", softlockup_panic_setup); static int __init nowatchdog_setup(char *str) { - no_watchdog = 1; + watchdog_enabled = 0; return 1; } __setup("nowatchdog", nowatchdog_setup); @@ -85,7 +82,7 @@ __setup("nowatchdog", nowatchdog_setup); /* deprecated */ static int __init nosoftlockup_setup(char *str) { - no_watchdog = 1; + watchdog_enabled = 0; return 1; } __setup("nosoftlockup", nosoftlockup_setup); @@ -476,9 +473,6 @@ static void watchdog_disable_all_cpus(void) { int cpu; - if (no_watchdog) - return; - for_each_online_cpu(cpu) watchdog_disable(cpu); @@ -530,7 +524,8 @@ cpu_callback(struct notifier_block *nfb, unsigned long action, void *hcpu) break; case CPU_ONLINE: case CPU_ONLINE_FROZEN: - err = watchdog_enable(hotcpu); + if (watchdog_enabled) + err = watchdog_enable(hotcpu); break; #ifdef CONFIG_HOTPLUG_CPU case CPU_UP_CANCELED: @@ -555,9 +550,6 @@ void __init lockup_detector_init(void) void *cpu = (void *)(long)smp_processor_id(); int err; - if (no_watchdog) - return; - err = cpu_callback(&cpu_nfb, CPU_UP_PREPARE, cpu); WARN_ON(notifier_to_errno(err)); From 397357666de6b5b6adb5fa99f9758ec8cf30ac34 Mon Sep 17 00:00:00 2001 From: Marcin Slusarz Date: Fri, 28 Jan 2011 11:00:32 -0500 Subject: [PATCH 4/6] watchdog: Fix sysctl consistency If it was not possible to enable watchdog for any cpu, switch watchdog_enabled back to 0, because it's visible via kernel.watchdog sysctl. Signed-off-by: Marcin Slusarz Signed-off-by: Don Zickus Cc: Peter Zijlstra Cc: Frederic Weisbecker Cc: LKML-Reference: <1296230433-6261-2-git-send-email-dzickus@redhat.com> Signed-off-by: Ingo Molnar --- kernel/watchdog.c | 13 +++++++------ 1 file changed, 7 insertions(+), 6 deletions(-) diff --git a/kernel/watchdog.c b/kernel/watchdog.c index d9961ea1c3f4..c7e0049344bd 100644 --- a/kernel/watchdog.c +++ b/kernel/watchdog.c @@ -429,9 +429,6 @@ static int watchdog_enable(int cpu) wake_up_process(p); } - /* if any cpu succeeds, watchdog is considered enabled for the system */ - watchdog_enabled = 1; - return 0; } @@ -459,12 +456,16 @@ static void watchdog_disable(int cpu) static void watchdog_enable_all_cpus(void) { int cpu; - int result = 0; + + watchdog_enabled = 0; for_each_online_cpu(cpu) - result += watchdog_enable(cpu); + if (!watchdog_enable(cpu)) + /* if any cpu succeeds, watchdog is considered + enabled for the system */ + watchdog_enabled = 1; - if (result) + if (!watchdog_enabled) printk(KERN_ERR "watchdog: failed to be enabled on some cpus\n"); } From 9ffdc6c37df131f89d52001e0ef03091b158826f Mon Sep 17 00:00:00 2001 From: Marcin Slusarz Date: Fri, 28 Jan 2011 11:00:33 -0500 Subject: [PATCH 5/6] watchdog: Don't change watchdog state on read of sysctl Signed-off-by: Marcin Slusarz [ add {}'s to fix a warning ] Signed-off-by: Don Zickus Cc: Peter Zijlstra Cc: Frederic Weisbecker Cc: LKML-Reference: <1296230433-6261-3-git-send-email-dzickus@redhat.com> Signed-off-by: Ingo Molnar --- kernel/watchdog.c | 10 ++++++---- 1 file changed, 6 insertions(+), 4 deletions(-) diff --git a/kernel/watchdog.c b/kernel/watchdog.c index c7e0049344bd..f37f974aa81b 100644 --- a/kernel/watchdog.c +++ b/kernel/watchdog.c @@ -493,10 +493,12 @@ int proc_dowatchdog_enabled(struct ctl_table *table, int write, { proc_dointvec(table, write, buffer, length, ppos); - if (watchdog_enabled) - watchdog_enable_all_cpus(); - else - watchdog_disable_all_cpus(); + if (write) { + if (watchdog_enabled) + watchdog_enable_all_cpus(); + else + watchdog_disable_all_cpus(); + } return 0; } From 542e72fc90f5ed9eecb574f80f70868c7f296093 Mon Sep 17 00:00:00 2001 From: Peter Zijlstra Date: Wed, 26 Jan 2011 15:38:35 +0100 Subject: [PATCH 6/6] perf: Fix reading in perf_event_read() It is quite possible for the event to have been disabled between perf_event_read() sending the IPI and the CPU servicing the IPI and calling __perf_event_read(), hence revalidate the state. Reported-by: Stephane Eranian Signed-off-by: Peter Zijlstra LKML-Reference: Signed-off-by: Ingo Molnar --- kernel/perf_event.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/kernel/perf_event.c b/kernel/perf_event.c index 852ae8c66502..999835b6112b 100644 --- a/kernel/perf_event.c +++ b/kernel/perf_event.c @@ -1901,11 +1901,12 @@ static void __perf_event_read(void *info) return; raw_spin_lock(&ctx->lock); - update_context_time(ctx); + if (ctx->is_active) + update_context_time(ctx); update_event_times(event); + if (event->state == PERF_EVENT_STATE_ACTIVE) + event->pmu->read(event); raw_spin_unlock(&ctx->lock); - - event->pmu->read(event); } static inline u64 perf_event_count(struct perf_event *event)