2009-06-05 02:36:44

by Paul Mackerras

Subject: [PATCH] perf_counter: Fix lockup with interrupting counters

Commit 8e3747c1 ("perf_counter: Change data head from u32 to u64")
changed the type of 'head' in struct perf_mmap_data from atomic_t
to atomic_long_t, but missed converting one use of atomic_read on
it to atomic_long_read. The effect of using atomic_read rather than
atomic_long_read on powerpc (and other big-endian architectures) is
that we get the high half of the 64-bit quantity, resulting in the
cmpxchg retry loop in perf_output_begin spinning forever as soon as
data->head becomes non-zero. On little-endian architectures such as
x86 we would get the low half, resulting in a lockup once data->head
becomes greater than 4G.

This fixes it by using atomic_long_read rather than atomic_read.

Signed-off-by: Paul Mackerras <[email protected]>
---
kernel/perf_counter.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index 195712e..a5d3e2a 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -2234,7 +2234,7 @@ static int perf_output_begin(struct perf_output_handle *handle,
perf_output_lock(handle);

do {
- offset = head = atomic_read(&data->head);
+ offset = head = atomic_long_read(&data->head);
head += size;
} while (atomic_long_cmpxchg(&data->head, offset, head) != offset);

--
1.6.0.4

2009-06-05 06:26:32

by Paul Mackerras

Subject: [tip:perfcounters/core] perf_counter: Fix lockup with interrupting counters

Commit-ID: 6dc5f2a41759987e35e757ef00192e7b424563bb
Gitweb: http://git.kernel.org/tip/6dc5f2a41759987e35e757ef00192e7b424563bb
Author: Paul Mackerras <[email protected]>
AuthorDate: Fri, 5 Jun 2009 12:36:28 +1000
Committer: Ingo Molnar <[email protected]>
CommitDate: Fri, 5 Jun 2009 08:22:26 +0200

perf_counter: Fix lockup with interrupting counters

Commit 8e3747c1 ("perf_counter: Change data head from u32 to u64")
changed the type of 'head' in struct perf_mmap_data from atomic_t
to atomic_long_t, but missed converting one use of atomic_read on
it to atomic_long_read. The effect of using atomic_read rather than
atomic_long_read on powerpc (and other big-endian architectures) is
that we get the high half of the 64-bit quantity, resulting in the
cmpxchg retry loop in perf_output_begin spinning forever as soon as
data->head becomes non-zero. On little-endian architectures such as
x86 we would get the low half, resulting in a lockup once data->head
becomes greater than 4G.

This fixes it by using atomic_long_read rather than atomic_read.

[ Impact: fix perfcounter lockup on PowerPC / big-endian systems ]

Signed-off-by: Paul Mackerras <[email protected]>
Cc: Peter Zijlstra <[email protected]>
LKML-Reference: <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>


---
kernel/perf_counter.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index 195712e..a5d3e2a 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -2234,7 +2234,7 @@ static int perf_output_begin(struct perf_output_handle *handle,
perf_output_lock(handle);

do {
- offset = head = atomic_read(&data->head);
+ offset = head = atomic_long_read(&data->head);
head += size;
} while (atomic_long_cmpxchg(&data->head, offset, head) != offset);