perf: Fix inherit vs. context rotation bug
Commit-ID: dddd3379a619a4cb8247bfd3c94ca9ae3797aa2e Gitweb: http://git.kernel.org/tip/dddd3379a619a4cb8247bfd3c94ca9ae3797aa2e Author: Thomas Gleixner <[email protected]> AuthorDate: Wed, 24 Nov 2010 10:05:55 +0100 Committer: Ingo Molnar <[email protected]> CommitDate: Fri, 26 Nov 2010 15:00:56 +0100
perf: Fix inherit vs. context rotation bug
It was found that sometimes children of tasks with inherited events had one extra event. Eventually it turned out to be due to the list rotation no being exclusive with the list iteration in the inheritance code.
Cure this by temporarily disabling the rotation while we inherit the events.
LKML-Reference: <new-submission>
--- include/linux/perf_event.h | 1 + kernel/perf_event.c | 22 ++++++++++++++++++++-- 2 files changed, 21 insertions(+), 2 deletions(-)
include/linux/perf_event.h +1 -0
@@ -850,6 +850,7 @@ struct perf_event_context {int nr_active;int is_active;int nr_stat; +int rotate_disable;atomic_t refcount;struct task_struct *task;
kernel/perf_event.c +20 -2
@@ -1622,8 +1622,12 @@ static void rotate_ctx(struct perf_event_context *ctx) {raw_spin_lock (&ctx->lock); -/* Rotate the first entry last of non-pinned groups */ -list_rotate_left (&ctx->flexible_groups); +/* +* Rotate the first entry last of non-pinned groups. Rotation might be +* disabled by the inheritance code. +*/ +if (!ctx->rotate_disable) +list_rotate_left (&ctx->flexible_groups);raw_spin_unlock (&ctx->lock); }
@@ -6162,6 +6166,7 @@ int perf_event_init_context(struct task_struct *child, int ctxn)struct perf_event *event;struct task_struct *parent = current;int inherited_all =1 ; +unsigned long flags;int ret =0 ; child->perf_event_ctxp[ctxn] =NULL ;
@@ -6202,6 +6207,15 @@ int perf_event_init_context(struct task_struct *child, int ctxn)break ; } +/* +* We can't hold ctx->lock when iterating the ->flexible_group list due +* to allocations, but we need to prevent rotation because +* rotate_ctx() will change the list from interrupt context. +*/ +raw_spin_lock_irqsave (&parent_ctx->lock, flags); + parent_ctx->rotate_disable =1 ; +raw_spin_unlock_irqrestore (&parent_ctx->lock, flags); +list_for_each_entry (event, &parent_ctx->flexible_groups, group_entry) { ret =inherit_task_group (event, parent, parent_ctx, child, ctxn, &inherited_all);
@@ -6209,6 +6223,10 @@ int perf_event_init_context(struct task_struct *child, int ctxn)break ; } +raw_spin_lock_irqsave (&parent_ctx->lock, flags); + parent_ctx->rotate_disable =0 ; +raw_spin_unlock_irqrestore (&parent_ctx->lock, flags); + child_ctx = child->perf_event_ctxp[ctxn];if (child_ctx && inherited_all) {
Quoted 30 lines · tip-bot for Thomas Gleixner
Commit-ID: dddd3379a619a4cb8247bfd3c94ca9ae3797aa2e
Gitweb: http://git.kernel.org/tip/dddd3379a619a4cb8247bfd3c94ca9ae3797aa2e
Author: Thomas Gleixner <[email protected]>
AuthorDate: Wed, 24 Nov 2010 10:05:55 +0100
Committer: Ingo Molnar <[email protected]>
CommitDate: Fri, 26 Nov 2010 15:00:56 +0100perf: Fix inherit vs. context rotation bug
It was found that sometimes children of tasks with inherited events had one extra event. Eventually it turned out to be due to the list rotation no being exclusive with the list iteration in the inheritance code.
Cure this by temporarily disabling the rotation while we inherit the events.
Signed-off-by Thomas Gleixner <[email protected]>Signed-off-by Peter Zijlstra <[email protected]>LKML-Reference: <new-submission>
Cc <[email protected]>Signed-off-by Ingo Molnar <[email protected]>---
include/linux/perf_event.h | 1 +kernel/perf_event.c| 22 ++++++++++++++++++++--2 files changed, 21 insertions(+), 2 deletions(-)include/linux/perf_event.h +0
-0
include/linux/perf_event.h struct perf_event_context { · line 853
Quoted 4 lines
include/linux/perf_event.h +1
-0@@ -850,6 +850,7 @@ struct perf_event_context { int nr_active; int is_active; int nr_stat; + int rotate_disable;
Isn't it convincing to use atomic_t instead of int as rotate_disable type. Since, we just have to increment or decrement it inside raw_spin_lock_irqsave(), we could've easily do it using atomic_inc/atomic_dec and atomic_read while calling list_rotate_left(). It will allow us to remove raw_spin_lock_irqsave() at increment/decrement. Am I missing anyting?
thanks,
rakib
Quoted 63 lines
include/linux/perf_event.h +0
-0atomic_t refcount; struct task_struct *task;kernel/perf_event.c +20
-3@@ -1622,8 +1622,12 @@ static void rotate_ctx(struct perf_event_context *ctx) { raw_spin_lock(&ctx->lock); - /* Rotate the first entry last of non-pinned groups */ - list_rotate_left(&ctx->flexible_groups); + /* + * Rotate the first entry last of non-pinned groups. Rotation might be + * disabled by the inheritance code. + */ + if (!ctx->rotate_disable) + list_rotate_left(&ctx->flexible_groups); raw_spin_unlock(&ctx->lock); }@@ -6162,6 +6166,7 @@ int perf_event_init_context(struct task_struct *child, int ctxn) struct perf_event *event; struct task_struct *parent = current; int inherited_all = 1; + unsigned long flags; int ret = 0; child->perf_event_ctxp[ctxn] = NULL;@@ -6202,6 +6207,15 @@ int perf_event_init_context(struct task_struct *child, int ctxn) break; } + /* + * We can't hold ctx->lock when iterating the ->flexible_group list due + * to allocations, but we need to prevent rotation because + * rotate_ctx() will change the list from interrupt context. + */ + raw_spin_lock_irqsave(&parent_ctx->lock, flags); + parent_ctx->rotate_disable = 1; + raw_spin_unlock_irqrestore(&parent_ctx->lock, flags); + list_for_each_entry(event, &parent_ctx->flexible_groups, group_entry) { ret = inherit_task_group(event, parent, parent_ctx, child, ctxn, &inherited_all);@@ -6209,6 +6223,10 @@ int perf_event_init_context(struct task_struct *child, int ctxn) break; } + raw_spin_lock_irqsave(&parent_ctx->lock, flags); + parent_ctx->rotate_disable = 0; + raw_spin_unlock_irqrestore(&parent_ctx->lock, flags); + child_ctx = child->perf_event_ctxp[ctxn]; if (child_ctx && inherited_all) { --To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/