Re: [PATCH v2 2/2] perf/x86/lbr: Move cpuc->lbr_xsave allocation out of sleeping region

From: Like Xu
Date: Tue May 18 2021 - 04:48:54 EST


Hi Peter, do we have any comments on these two patches ?

On 2021/5/10 10:10, Like Xu wrote:
Hi Peter, do you have any comments on this version ?

On 2021/4/30 13:22, Like Xu wrote:
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 intel_pmu_hw_config(). The LBR
xsave buffer is a per-CPU buffer, not a per-event buffer. This buffer is
allocated once when initializing the LBR 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")
Tested-by: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
Signed-off-by: Like Xu <like.xu@xxxxxxxxxxxxxxx>
---
v1->v2 Changelog:
- Frob reserve_lbr_buffers() in intel_pmu_hw_config().

  arch/x86/events/intel/core.c |  2 ++
  arch/x86/events/intel/lbr.c  | 25 +++++++++++++++++++------
  arch/x86/events/perf_event.h |  6 ++++++
  3 files changed, 27 insertions(+), 6 deletions(-)

diff --git a/arch/x86/events/intel/core.c b/arch/x86/events/intel/core.c
index 456aa6ffd9a1..19027aa01f82 100644
--- a/arch/x86/events/intel/core.c
+++ b/arch/x86/events/intel/core.c
@@ -3745,6 +3745,8 @@ static int intel_pmu_hw_config(struct perf_event *event)
              event->destroy = hw_perf_lbr_event_destroy;
          }
+
+        reserve_lbr_buffers(event);
      }
      if (event->attr.aux_output) {
diff --git a/arch/x86/events/intel/lbr.c b/arch/x86/events/intel/lbr.c
index bb4486c4155a..1c39155c9f67 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)
@@ -722,6 +716,25 @@ void release_lbr_buffers(void)
      }
  }
+void reserve_lbr_buffers(struct perf_event *event)
+{
+    struct kmem_cache *kmem_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);
+        kmem_cache = x86_get_pmu(cpu)->task_ctx_cache;
+        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 27fa85e7d4fd..727e5490e78c 100644
--- a/arch/x86/events/perf_event.h
+++ b/arch/x86/events/perf_event.h
@@ -1244,6 +1244,8 @@ 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;
@@ -1393,6 +1395,10 @@ 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;