From: Like Xu <like.xu@linux.intel.com>
To: Peter Zijlstra <peterz@infradead.org>,
Namhyung Kim <namhyung@kernel.org>
Cc: pbonzini@redhat.com, Ingo Molnar <mingo@redhat.com>,
Arnaldo Carvalho de Melo <acme@kernel.org>,
Mark Rutland <mark.rutland@arm.com>,
Alexander Shishkin <alexander.shishkin@linux.intel.com>,
Jiri Olsa <jolsa@redhat.com>,
Kan Liang <kan.liang@linux.intel.com>,
wei.w.wang@intel.com, x86@kernel.org,
linux-kernel@vger.kernel.org, Like Xu <like.xu@linux.intel.com>
Subject: [PATCH v5 4/5] perf/x86/lbr: Move cpuc->lbr_xsave allocation out of sleeping region
Date: Fri, 26 Mar 2021 09:19:17 +0800 [thread overview]
Message-ID: <20210326011918.183685-5-like.xu@linux.intel.com> (raw)
In-Reply-To: <20210326011918.183685-1-like.xu@linux.intel.com>
If the kernel is compiled with the CONFIG_LOCKDEP option, the conditional
might_sleep_if() deep in kmem_cache_alloc() will generate the following
trace, and potentially cause a deadlock when another LBR event is added:
[ 243.115549] BUG: sleeping function called from invalid context at include/linux/sched/mm.h:196
[ 243.117576] in_atomic(): 1, irqs_disabled(): 1, non_block: 0, pid: 839, name: perf
[ 243.119326] INFO: lockdep is turned off.
[ 243.120249] irq event stamp: 0
[ 243.120967] hardirqs last enabled at (0): [<0000000000000000>] 0x0
[ 243.122415] hardirqs last disabled at (0): [<ffffffff810d9bf5>] copy_process+0xa45/0x1dc0
[ 243.124302] softirqs last enabled at (0): [<ffffffff810d9bf5>] copy_process+0xa45/0x1dc0
[ 243.126255] softirqs last disabled at (0): [<0000000000000000>] 0x0
[ 243.128119] CPU: 0 PID: 839 Comm: perf Not tainted 5.11.0-rc4-guest+ #8
[ 243.129654] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 0.0.0 02/06/2015
[ 243.131520] Call Trace:
[ 243.132112] dump_stack+0x8d/0xb5
[ 243.132896] ___might_sleep.cold.106+0xb3/0xc3
[ 243.133984] slab_pre_alloc_hook.constprop.85+0x96/0xd0
[ 243.135208] ? intel_pmu_lbr_add+0x152/0x170
[ 243.136207] kmem_cache_alloc+0x36/0x250
[ 243.137126] intel_pmu_lbr_add+0x152/0x170
[ 243.138088] x86_pmu_add+0x83/0xd0
[ 243.138889] ? lock_acquire+0x158/0x350
[ 243.139791] ? lock_acquire+0x158/0x350
[ 243.140694] ? lock_acquire+0x158/0x350
[ 243.141625] ? lock_acquired+0x1e3/0x360
[ 243.142544] ? lock_release+0x1bf/0x340
[ 243.143726] ? trace_hardirqs_on+0x1a/0xd0
[ 243.144823] ? lock_acquired+0x1e3/0x360
[ 243.145742] ? lock_release+0x1bf/0x340
[ 243.147107] ? __slab_free+0x49/0x540
[ 243.147966] ? trace_hardirqs_on+0x1a/0xd0
[ 243.148924] event_sched_in.isra.129+0xf8/0x2a0
[ 243.149989] merge_sched_in+0x261/0x3e0
[ 243.150889] ? trace_hardirqs_on+0x1a/0xd0
[ 243.151869] visit_groups_merge.constprop.135+0x130/0x4a0
[ 243.153122] ? sched_clock_cpu+0xc/0xb0
[ 243.154023] ctx_sched_in+0x101/0x210
[ 243.154884] ctx_resched+0x6f/0xc0
[ 243.155686] perf_event_exec+0x21e/0x2e0
[ 243.156641] begin_new_exec+0x5e5/0xbd0
[ 243.157540] load_elf_binary+0x6af/0x1770
[ 243.158478] ? __kernel_read+0x19d/0x2b0
[ 243.159977] ? lock_acquire+0x158/0x350
[ 243.160876] ? __kernel_read+0x19d/0x2b0
[ 243.161796] bprm_execve+0x3c8/0x840
[ 243.162638] do_execveat_common.isra.38+0x1a5/0x1c0
[ 243.163776] __x64_sys_execve+0x32/0x40
[ 243.164676] do_syscall_64+0x33/0x40
[ 243.165514] entry_SYSCALL_64_after_hwframe+0x44/0xa9
[ 243.166746] RIP: 0033:0x7f6180a26feb
[ 243.167590] Code: Unable to access opcode bytes at RIP 0x7f6180a26fc1.
[ 243.169097] RSP: 002b:00007ffc6558ce18 EFLAGS: 00000202 ORIG_RAX: 000000000000003b
[ 243.170844] RAX: ffffffffffffffda RBX: 00007ffc65592d30 RCX: 00007f6180a26feb
[ 243.172514] RDX: 000055657f408dc0 RSI: 00007ffc65592410 RDI: 00007ffc65592d30
[ 243.174162] RBP: 00007ffc6558ce80 R08: 00007ffc6558cde0 R09: 0000000000000000
[ 243.176042] R10: 0000000000000008 R11: 0000000000000202 R12: 00007ffc65592410
[ 243.177696] R13: 000055657f408dc0 R14: 0000000000000001 R15: 00007ffc65592410
One of the solution is to use GFP_ATOMIC, but it will make the code less
reliable under memory pressue. Let's move the memory allocation out of
the sleeping region and put it into the x86_reserve_hardware(). The LBR
xsave buffer is a per-CPU buffer, not a per-event buffer. This buffer is
allocated once when initializing the first event.
The disadvantage of this fix is that the cpuc->lbr_xsave memory
will be allocated for each cpu like the legacy ds_buffer.
Fixes: c085fb8774 ("perf/x86/intel/lbr: Support XSAVES for arch LBR read")
Suggested-by: Kan Liang <kan.liang@linux.intel.com>
Tested-by: Kan Liang <kan.liang@linux.intel.com>
Signed-off-by: Like Xu <like.xu@linux.intel.com>
---
arch/x86/events/core.c | 8 +++++---
arch/x86/events/intel/bts.c | 2 +-
arch/x86/events/intel/lbr.c | 23 +++++++++++++++++------
arch/x86/events/perf_event.h | 8 +++++++-
4 files changed, 30 insertions(+), 11 deletions(-)
diff --git a/arch/x86/events/core.c b/arch/x86/events/core.c
index 18df17129695..a4ce669cc78d 100644
--- a/arch/x86/events/core.c
+++ b/arch/x86/events/core.c
@@ -373,7 +373,7 @@ set_ext_hw_attr(struct hw_perf_event *hwc, struct perf_event *event)
return x86_pmu_extra_regs(val, event);
}
-int x86_reserve_hardware(void)
+int x86_reserve_hardware(struct perf_event *event)
{
int err = 0;
@@ -382,8 +382,10 @@ int x86_reserve_hardware(void)
if (atomic_read(&pmc_refcount) == 0) {
if (!reserve_pmc_hardware())
err = -EBUSY;
- else
+ else {
reserve_ds_buffers();
+ reserve_lbr_buffers(event);
+ }
}
if (!err)
atomic_inc(&pmc_refcount);
@@ -634,7 +636,7 @@ static int __x86_pmu_event_init(struct perf_event *event)
if (!x86_pmu_initialized())
return -ENODEV;
- err = x86_reserve_hardware();
+ err = x86_reserve_hardware(event);
if (err)
return err;
diff --git a/arch/x86/events/intel/bts.c b/arch/x86/events/intel/bts.c
index 731dd8d0dbb1..057bb2f761a9 100644
--- a/arch/x86/events/intel/bts.c
+++ b/arch/x86/events/intel/bts.c
@@ -564,7 +564,7 @@ static int bts_event_init(struct perf_event *event)
if (x86_add_exclusive(x86_lbr_exclusive_bts))
return -EBUSY;
- ret = x86_reserve_hardware();
+ ret = x86_reserve_hardware(event);
if (ret) {
x86_del_exclusive(x86_lbr_exclusive_bts);
return ret;
diff --git a/arch/x86/events/intel/lbr.c b/arch/x86/events/intel/lbr.c
index 355ea70f1879..6df9a802613f 100644
--- a/arch/x86/events/intel/lbr.c
+++ b/arch/x86/events/intel/lbr.c
@@ -658,7 +658,6 @@ static inline bool branch_user_callstack(unsigned br_sel)
void intel_pmu_lbr_add(struct perf_event *event)
{
- struct kmem_cache *kmem_cache = event->pmu->task_ctx_cache;
struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
if (!x86_pmu.lbr_nr)
@@ -696,11 +695,6 @@ void intel_pmu_lbr_add(struct perf_event *event)
perf_sched_cb_inc(event->ctx->pmu);
if (!cpuc->lbr_users++ && !event->total_time_running)
intel_pmu_lbr_reset();
-
- if (static_cpu_has(X86_FEATURE_ARCH_LBR) &&
- kmem_cache && !cpuc->lbr_xsave &&
- (cpuc->lbr_users != cpuc->lbr_pebs_users))
- cpuc->lbr_xsave = kmem_cache_alloc(kmem_cache, GFP_KERNEL);
}
void release_lbr_buffers(void)
@@ -721,6 +715,23 @@ void release_lbr_buffers(void)
}
}
+void reserve_lbr_buffers(struct perf_event *event)
+{
+ struct kmem_cache *kmem_cache = x86_get_pmu()->task_ctx_cache;
+ struct cpu_hw_events *cpuc;
+ int cpu;
+
+ if (!static_cpu_has(X86_FEATURE_ARCH_LBR))
+ return;
+
+ for_each_possible_cpu(cpu) {
+ cpuc = per_cpu_ptr(&cpu_hw_events, cpu);
+ if (kmem_cache && !cpuc->lbr_xsave && !event->attr.precise_ip)
+ cpuc->lbr_xsave = kmem_cache_alloc_node(kmem_cache, GFP_KERNEL,
+ cpu_to_node(cpu));
+ }
+}
+
void intel_pmu_lbr_del(struct perf_event *event)
{
struct cpu_hw_events *cpuc = this_cpu_ptr(&cpu_hw_events);
diff --git a/arch/x86/events/perf_event.h b/arch/x86/events/perf_event.h
index 53b2b5fc23bc..2fe77d3a98d6 100644
--- a/arch/x86/events/perf_event.h
+++ b/arch/x86/events/perf_event.h
@@ -968,7 +968,7 @@ int x86_add_exclusive(unsigned int what);
void x86_del_exclusive(unsigned int what);
-int x86_reserve_hardware(void);
+int x86_reserve_hardware(struct perf_event *event);
void x86_release_hardware(void);
@@ -1135,6 +1135,8 @@ void reserve_ds_buffers(void);
void release_lbr_buffers(void);
+void reserve_lbr_buffers(struct perf_event *event);
+
extern struct event_constraint bts_constraint;
extern struct event_constraint vlbr_constraint;
@@ -1282,6 +1284,10 @@ static inline void release_lbr_buffers(void)
{
}
+static inline void reserve_lbr_buffers(struct perf_event *event)
+{
+}
+
static inline int intel_pmu_init(void)
{
return 0;
--
2.29.2
next prev parent reply other threads:[~2021-03-26 1:28 UTC|newest]
Thread overview: 8+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-03-26 1:19 [PATCH v5 0/5] perf/x86: Some minor changes to support guest Arch LBR Like Xu
2021-03-26 1:19 ` [PATCH v5 1/5] perf/x86/intel: Fix the comment about guest LBR support on KVM Like Xu
2021-03-26 1:19 ` [PATCH v5 2/5] perf/x86/lbr: Simplify the exposure check for the LBR_INFO registers Like Xu
2021-03-26 1:19 ` [PATCH v5 3/5] perf/x86: Skip checking MSR for MSR 0x000 Like Xu
2021-03-26 1:19 ` Like Xu [this message]
2021-03-26 1:19 ` [PATCH v5 5/5] perf/x86: Move ARCH_LBR_CTL_MASK definition to include/asm/msr-index.h Like Xu
2021-04-06 3:20 ` [PATCH v5 0/5] perf/x86: Some minor changes to support guest Arch LBR Like Xu
2021-04-09 8:51 ` Like Xu
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20210326011918.183685-5-like.xu@linux.intel.com \
--to=like.xu@linux.intel.com \
--cc=acme@kernel.org \
--cc=alexander.shishkin@linux.intel.com \
--cc=jolsa@redhat.com \
--cc=kan.liang@linux.intel.com \
--cc=linux-kernel@vger.kernel.org \
--cc=mark.rutland@arm.com \
--cc=mingo@redhat.com \
--cc=namhyung@kernel.org \
--cc=pbonzini@redhat.com \
--cc=peterz@infradead.org \
--cc=wei.w.wang@intel.com \
--cc=x86@kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).