* [PATCH] smp: Use release stores for csd_lock_record() state
@ 2026-06-17 21:20 Usama Arif
2026-06-18 1:44 ` Alan Stern
0 siblings, 1 reply; 7+ messages in thread
From: Usama Arif @ 2026-06-17 21:20 UTC (permalink / raw)
To: lkmm, joelagnelf, linux-kernel, marco.crivellari, paulmck,
rafael.j.wysocki, rdunlap, riel, sshegde, tglx, ulfh, usama.arif,
yury.norov, rcu
Cc: shakeel.butt, hannes, kernel-team
__csd_lock_record() publishes per-CPU CSD debug state that is read by
csd_lock_wait_toolong() on another CPU. The remote side first reads
cur_csd with smp_load_acquire() and, when non-NULL, may then read the
matching cur_csd_func and cur_csd_info fields.
Use smp_store_release() when publishing cur_csd so that the preceding
cur_csd_func and cur_csd_info stores are ordered before the pointer
that csd_lock_wait_toolong() acquires. This replaces the open-coded
smp_wmb() plus plain cur_csd store with the release operation that
matches the smp_load_acquire() in csd_lock_wait_toolong().
For the clear path, use smp_store_release(&cur_csd, NULL) so that
clearing the diagnostic state remains ordered after the preceding
callback/unlock work, without requiring a full barrier before the
store. On x86 this removes the locked full barrier from the clear
path; on weaker memory models it uses the release operation needed by
the smp_load_acquire() in csd_lock_wait_toolong().
The old code also had smp_mb() calls around cur_csd updates. Those would
only be needed if cur_csd were treated as an exact live-state marker whose
publication had to be observed before callback execution or CSD unlock.
CSD stall warnings do not currently have RCU-style stall-ended checks, so
they already allow the stall to end while diagnostics are being assembled.
The cur_csd record is therefore best-effort diagnostic context, not a
precise completion/stall boundary.
Signed-off-by: Usama Arif <usama.arif@linux.dev>
---
kernel/smp.c | 8 ++------
1 file changed, 2 insertions(+), 6 deletions(-)
diff --git a/kernel/smp.c b/kernel/smp.c
index a0bb56bd8dda..5ba4a20ba77d 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -182,16 +182,12 @@ static atomic_t csd_bug_count = ATOMIC_INIT(0);
static void __csd_lock_record(call_single_data_t *csd)
{
if (!csd) {
- smp_mb(); /* NULL cur_csd after unlock. */
- __this_cpu_write(cur_csd, NULL);
+ smp_store_release(this_cpu_ptr(&cur_csd), NULL);
return;
}
__this_cpu_write(cur_csd_func, csd->func);
__this_cpu_write(cur_csd_info, csd->info);
- smp_wmb(); /* func and info before csd. */
- __this_cpu_write(cur_csd, csd);
- smp_mb(); /* Update cur_csd before function call. */
- /* Or before unlock, as the case may be. */
+ smp_store_release(this_cpu_ptr(&cur_csd), csd);
}
static __always_inline void csd_lock_record(call_single_data_t *csd)
--
2.53.0-Meta
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: [PATCH] smp: Use release stores for csd_lock_record() state
2026-06-17 21:20 [PATCH] smp: Use release stores for csd_lock_record() state Usama Arif
@ 2026-06-18 1:44 ` Alan Stern
2026-06-18 3:30 ` Randy Dunlap
0 siblings, 1 reply; 7+ messages in thread
From: Alan Stern @ 2026-06-18 1:44 UTC (permalink / raw)
To: Usama Arif
Cc: lkmm, joelagnelf, linux-kernel, marco.crivellari, paulmck,
rafael.j.wysocki, rdunlap, riel, sshegde, tglx, ulfh, yury.norov,
rcu, shakeel.butt, hannes, kernel-team
On Wed, Jun 17, 2026 at 02:20:01PM -0700, Usama Arif wrote:
> __csd_lock_record() publishes per-CPU CSD debug state that is read by
> csd_lock_wait_toolong() on another CPU. The remote side first reads
> cur_csd with smp_load_acquire() and, when non-NULL, may then read the
> matching cur_csd_func and cur_csd_info fields.
>
> Use smp_store_release() when publishing cur_csd so that the preceding
> cur_csd_func and cur_csd_info stores are ordered before the pointer
> that csd_lock_wait_toolong() acquires. This replaces the open-coded
> smp_wmb() plus plain cur_csd store with the release operation that
> matches the smp_load_acquire() in csd_lock_wait_toolong().
>
> For the clear path, use smp_store_release(&cur_csd, NULL) so that
> clearing the diagnostic state remains ordered after the preceding
> callback/unlock work, without requiring a full barrier before the
> store. On x86 this removes the locked full barrier from the clear
> path; on weaker memory models it uses the release operation needed by
> the smp_load_acquire() in csd_lock_wait_toolong().
>
> The old code also had smp_mb() calls around cur_csd updates. Those would
> only be needed if cur_csd were treated as an exact live-state marker whose
> publication had to be observed before callback execution or CSD unlock.
> CSD stall warnings do not currently have RCU-style stall-ended checks, so
> they already allow the stall to end while diagnostics are being assembled.
> The cur_csd record is therefore best-effort diagnostic context, not a
> precise completion/stall boundary.
>
> Signed-off-by: Usama Arif <usama.arif@linux.dev>
> ---
> kernel/smp.c | 8 ++------
> 1 file changed, 2 insertions(+), 6 deletions(-)
>
> diff --git a/kernel/smp.c b/kernel/smp.c
> index a0bb56bd8dda..5ba4a20ba77d 100644
> --- a/kernel/smp.c
> +++ b/kernel/smp.c
> @@ -182,16 +182,12 @@ static atomic_t csd_bug_count = ATOMIC_INIT(0);
> static void __csd_lock_record(call_single_data_t *csd)
> {
> if (!csd) {
> - smp_mb(); /* NULL cur_csd after unlock. */
> - __this_cpu_write(cur_csd, NULL);
> + smp_store_release(this_cpu_ptr(&cur_csd), NULL);
> return;
> }
> __this_cpu_write(cur_csd_func, csd->func);
> __this_cpu_write(cur_csd_info, csd->info);
> - smp_wmb(); /* func and info before csd. */
> - __this_cpu_write(cur_csd, csd);
> - smp_mb(); /* Update cur_csd before function call. */
> - /* Or before unlock, as the case may be. */
> + smp_store_release(this_cpu_ptr(&cur_csd), csd);
Isn't there a general policy in the kernel that memory barriers should
be accompanied by a comment explaining what other memory barriers they
synchronize with? Including such comments is a good idea in any case.
Alan Stern
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] smp: Use release stores for csd_lock_record() state
2026-06-18 1:44 ` Alan Stern
@ 2026-06-18 3:30 ` Randy Dunlap
2026-06-18 4:01 ` Paul E. McKenney
2026-06-18 14:44 ` Usama Arif
0 siblings, 2 replies; 7+ messages in thread
From: Randy Dunlap @ 2026-06-18 3:30 UTC (permalink / raw)
To: Alan Stern, Usama Arif
Cc: lkmm, joelagnelf, linux-kernel, marco.crivellari, paulmck,
rafael.j.wysocki, riel, sshegde, tglx, ulfh, yury.norov, rcu,
shakeel.butt, hannes, kernel-team
On 6/17/26 6:44 PM, Alan Stern wrote:
> On Wed, Jun 17, 2026 at 02:20:01PM -0700, Usama Arif wrote:
>> __csd_lock_record() publishes per-CPU CSD debug state that is read by
>> csd_lock_wait_toolong() on another CPU. The remote side first reads
>> cur_csd with smp_load_acquire() and, when non-NULL, may then read the
>> matching cur_csd_func and cur_csd_info fields.
>>
>> Use smp_store_release() when publishing cur_csd so that the preceding
>> cur_csd_func and cur_csd_info stores are ordered before the pointer
>> that csd_lock_wait_toolong() acquires. This replaces the open-coded
>> smp_wmb() plus plain cur_csd store with the release operation that
>> matches the smp_load_acquire() in csd_lock_wait_toolong().
>>
>> For the clear path, use smp_store_release(&cur_csd, NULL) so that
>> clearing the diagnostic state remains ordered after the preceding
>> callback/unlock work, without requiring a full barrier before the
>> store. On x86 this removes the locked full barrier from the clear
>> path; on weaker memory models it uses the release operation needed by
>> the smp_load_acquire() in csd_lock_wait_toolong().
>>
>> The old code also had smp_mb() calls around cur_csd updates. Those would
>> only be needed if cur_csd were treated as an exact live-state marker whose
>> publication had to be observed before callback execution or CSD unlock.
>> CSD stall warnings do not currently have RCU-style stall-ended checks, so
>> they already allow the stall to end while diagnostics are being assembled.
>> The cur_csd record is therefore best-effort diagnostic context, not a
>> precise completion/stall boundary.
>>
>> Signed-off-by: Usama Arif <usama.arif@linux.dev>
>> ---
>> kernel/smp.c | 8 ++------
>> 1 file changed, 2 insertions(+), 6 deletions(-)
>>
>> diff --git a/kernel/smp.c b/kernel/smp.c
>> index a0bb56bd8dda..5ba4a20ba77d 100644
>> --- a/kernel/smp.c
>> +++ b/kernel/smp.c
>> @@ -182,16 +182,12 @@ static atomic_t csd_bug_count = ATOMIC_INIT(0);
>> static void __csd_lock_record(call_single_data_t *csd)
>> {
>> if (!csd) {
>> - smp_mb(); /* NULL cur_csd after unlock. */
>> - __this_cpu_write(cur_csd, NULL);
>> + smp_store_release(this_cpu_ptr(&cur_csd), NULL);
>> return;
>> }
>> __this_cpu_write(cur_csd_func, csd->func);
>> __this_cpu_write(cur_csd_info, csd->info);
>> - smp_wmb(); /* func and info before csd. */
>> - __this_cpu_write(cur_csd, csd);
>> - smp_mb(); /* Update cur_csd before function call. */
>> - /* Or before unlock, as the case may be. */
>> + smp_store_release(this_cpu_ptr(&cur_csd), csd);
>
> Isn't there a general policy in the kernel that memory barriers should
> be accompanied by a comment explaining what other memory barriers they
> synchronize with? Including such comments is a good idea in any case.
in Documentation/process/submit-checklist.rst:
3) All memory barriers {e.g., ``barrier()``, ``rmb()``, ``wmb()``} need a
comment in the source code that explains the logic of what they are doing
and why.
in Documentation/process/4.Coding.rst:
Certain things should always be commented. Uses of memory barriers should
be accompanied by a line explaining why the barrier is necessary.
but looking in the 3000+ lines of Documentation/memory-barriers.txt won't tell
anyone about that.
--
~Randy
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] smp: Use release stores for csd_lock_record() state
2026-06-18 3:30 ` Randy Dunlap
@ 2026-06-18 4:01 ` Paul E. McKenney
2026-06-18 14:44 ` Usama Arif
1 sibling, 0 replies; 7+ messages in thread
From: Paul E. McKenney @ 2026-06-18 4:01 UTC (permalink / raw)
To: Randy Dunlap
Cc: Alan Stern, Usama Arif, lkmm, joelagnelf, linux-kernel,
marco.crivellari, rafael.j.wysocki, riel, sshegde, tglx, ulfh,
yury.norov, rcu, shakeel.butt, hannes, kernel-team
On Wed, Jun 17, 2026 at 08:30:32PM -0700, Randy Dunlap wrote:
>
>
> On 6/17/26 6:44 PM, Alan Stern wrote:
> > On Wed, Jun 17, 2026 at 02:20:01PM -0700, Usama Arif wrote:
> >> __csd_lock_record() publishes per-CPU CSD debug state that is read by
> >> csd_lock_wait_toolong() on another CPU. The remote side first reads
> >> cur_csd with smp_load_acquire() and, when non-NULL, may then read the
> >> matching cur_csd_func and cur_csd_info fields.
> >>
> >> Use smp_store_release() when publishing cur_csd so that the preceding
> >> cur_csd_func and cur_csd_info stores are ordered before the pointer
> >> that csd_lock_wait_toolong() acquires. This replaces the open-coded
> >> smp_wmb() plus plain cur_csd store with the release operation that
> >> matches the smp_load_acquire() in csd_lock_wait_toolong().
> >>
> >> For the clear path, use smp_store_release(&cur_csd, NULL) so that
> >> clearing the diagnostic state remains ordered after the preceding
> >> callback/unlock work, without requiring a full barrier before the
> >> store. On x86 this removes the locked full barrier from the clear
> >> path; on weaker memory models it uses the release operation needed by
> >> the smp_load_acquire() in csd_lock_wait_toolong().
> >>
> >> The old code also had smp_mb() calls around cur_csd updates. Those would
> >> only be needed if cur_csd were treated as an exact live-state marker whose
> >> publication had to be observed before callback execution or CSD unlock.
> >> CSD stall warnings do not currently have RCU-style stall-ended checks, so
> >> they already allow the stall to end while diagnostics are being assembled.
> >> The cur_csd record is therefore best-effort diagnostic context, not a
> >> precise completion/stall boundary.
> >>
> >> Signed-off-by: Usama Arif <usama.arif@linux.dev>
> >> ---
> >> kernel/smp.c | 8 ++------
> >> 1 file changed, 2 insertions(+), 6 deletions(-)
> >>
> >> diff --git a/kernel/smp.c b/kernel/smp.c
> >> index a0bb56bd8dda..5ba4a20ba77d 100644
> >> --- a/kernel/smp.c
> >> +++ b/kernel/smp.c
> >> @@ -182,16 +182,12 @@ static atomic_t csd_bug_count = ATOMIC_INIT(0);
> >> static void __csd_lock_record(call_single_data_t *csd)
> >> {
> >> if (!csd) {
> >> - smp_mb(); /* NULL cur_csd after unlock. */
> >> - __this_cpu_write(cur_csd, NULL);
> >> + smp_store_release(this_cpu_ptr(&cur_csd), NULL);
> >> return;
> >> }
> >> __this_cpu_write(cur_csd_func, csd->func);
> >> __this_cpu_write(cur_csd_info, csd->info);
> >> - smp_wmb(); /* func and info before csd. */
> >> - __this_cpu_write(cur_csd, csd);
> >> - smp_mb(); /* Update cur_csd before function call. */
> >> - /* Or before unlock, as the case may be. */
> >> + smp_store_release(this_cpu_ptr(&cur_csd), csd);
> >
> > Isn't there a general policy in the kernel that memory barriers should
> > be accompanied by a comment explaining what other memory barriers they
> > synchronize with? Including such comments is a good idea in any case.
>
> in Documentation/process/submit-checklist.rst:
>
> 3) All memory barriers {e.g., ``barrier()``, ``rmb()``, ``wmb()``} need a
> comment in the source code that explains the logic of what they are doing
> and why.
>
> in Documentation/process/4.Coding.rst:
>
> Certain things should always be commented. Uses of memory barriers should
> be accompanied by a line explaining why the barrier is necessary.
>
> but looking in the 3000+ lines of Documentation/memory-barriers.txt won't tell
> anyone about that.
Is smp_store_release() a memory barrier? Sorry, couldn't resist... ;-)
Thanx, Paul
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] smp: Use release stores for csd_lock_record() state
2026-06-18 3:30 ` Randy Dunlap
2026-06-18 4:01 ` Paul E. McKenney
@ 2026-06-18 14:44 ` Usama Arif
2026-06-18 14:57 ` Paul E. McKenney
1 sibling, 1 reply; 7+ messages in thread
From: Usama Arif @ 2026-06-18 14:44 UTC (permalink / raw)
To: Randy Dunlap, Alan Stern, paulmck
Cc: lkmm, joelagnelf, linux-kernel, marco.crivellari, paulmck,
rafael.j.wysocki, riel, sshegde, tglx, ulfh, yury.norov, rcu,
shakeel.butt, hannes, kernel-team
On 18/06/2026 04:30, Randy Dunlap wrote:
>
>
> On 6/17/26 6:44 PM, Alan Stern wrote:
>> On Wed, Jun 17, 2026 at 02:20:01PM -0700, Usama Arif wrote:
>>> __csd_lock_record() publishes per-CPU CSD debug state that is read by
>>> csd_lock_wait_toolong() on another CPU. The remote side first reads
>>> cur_csd with smp_load_acquire() and, when non-NULL, may then read the
>>> matching cur_csd_func and cur_csd_info fields.
>>>
>>> Use smp_store_release() when publishing cur_csd so that the preceding
>>> cur_csd_func and cur_csd_info stores are ordered before the pointer
>>> that csd_lock_wait_toolong() acquires. This replaces the open-coded
>>> smp_wmb() plus plain cur_csd store with the release operation that
>>> matches the smp_load_acquire() in csd_lock_wait_toolong().
>>>
>>> For the clear path, use smp_store_release(&cur_csd, NULL) so that
>>> clearing the diagnostic state remains ordered after the preceding
>>> callback/unlock work, without requiring a full barrier before the
>>> store. On x86 this removes the locked full barrier from the clear
>>> path; on weaker memory models it uses the release operation needed by
>>> the smp_load_acquire() in csd_lock_wait_toolong().
>>>
>>> The old code also had smp_mb() calls around cur_csd updates. Those would
>>> only be needed if cur_csd were treated as an exact live-state marker whose
>>> publication had to be observed before callback execution or CSD unlock.
>>> CSD stall warnings do not currently have RCU-style stall-ended checks, so
>>> they already allow the stall to end while diagnostics are being assembled.
>>> The cur_csd record is therefore best-effort diagnostic context, not a
>>> precise completion/stall boundary.
>>>
>>> Signed-off-by: Usama Arif <usama.arif@linux.dev>
>>> ---
>>> kernel/smp.c | 8 ++------
>>> 1 file changed, 2 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/kernel/smp.c b/kernel/smp.c
>>> index a0bb56bd8dda..5ba4a20ba77d 100644
>>> --- a/kernel/smp.c
>>> +++ b/kernel/smp.c
>>> @@ -182,16 +182,12 @@ static atomic_t csd_bug_count = ATOMIC_INIT(0);
>>> static void __csd_lock_record(call_single_data_t *csd)
>>> {
>>> if (!csd) {
>>> - smp_mb(); /* NULL cur_csd after unlock. */
>>> - __this_cpu_write(cur_csd, NULL);
>>> + smp_store_release(this_cpu_ptr(&cur_csd), NULL);
>>> return;
>>> }
>>> __this_cpu_write(cur_csd_func, csd->func);
>>> __this_cpu_write(cur_csd_info, csd->info);
>>> - smp_wmb(); /* func and info before csd. */
>>> - __this_cpu_write(cur_csd, csd);
>>> - smp_mb(); /* Update cur_csd before function call. */
>>> - /* Or before unlock, as the case may be. */
>>> + smp_store_release(this_cpu_ptr(&cur_csd), csd);
>>
>> Isn't there a general policy in the kernel that memory barriers should
>> be accompanied by a comment explaining what other memory barriers they
>> synchronize with? Including such comments is a good idea in any case.
>
> in Documentation/process/submit-checklist.rst:
>
> 3) All memory barriers {e.g., ``barrier()``, ``rmb()``, ``wmb()``} need a
> comment in the source code that explains the logic of what they are doing
> and why.
>
> in Documentation/process/4.Coding.rst:
>
> Certain things should always be commented. Uses of memory barriers should
> be accompanied by a line explaining why the barrier is necessary.
>
> but looking in the 3000+ lines of Documentation/memory-barriers.txt won't tell
> anyone about that.
>
>
Thanks!
I will send a v2 with the below diff if there are no objections?
diff --git a/kernel/smp.c b/kernel/smp.c
index 5ba4a20ba77d..685829875a3e 100644
--- a/kernel/smp.c
+++ b/kernel/smp.c
@@ -182,11 +182,21 @@ static atomic_t csd_bug_count = ATOMIC_INIT(0);
static void __csd_lock_record(call_single_data_t *csd)
{
if (!csd) {
+ /*
+ * Pairs with smp_load_acquire() of cur_csd in
+ * csd_lock_wait_toolong(): orders any preceding CSD
+ * callback/unlock before a remote reader observes NULL.
+ */
smp_store_release(this_cpu_ptr(&cur_csd), NULL);
return;
}
__this_cpu_write(cur_csd_func, csd->func);
__this_cpu_write(cur_csd_info, csd->info);
+ /*
+ * Pairs with smp_load_acquire() of cur_csd in
+ * csd_lock_wait_toolong(): publishes cur_csd_func and
+ * cur_csd_info before the non-NULL pointer becomes visible.
+ */
smp_store_release(this_cpu_ptr(&cur_csd), csd);
}
^ permalink raw reply related [flat|nested] 7+ messages in thread
* Re: [PATCH] smp: Use release stores for csd_lock_record() state
2026-06-18 14:44 ` Usama Arif
@ 2026-06-18 14:57 ` Paul E. McKenney
2026-06-18 15:30 ` Usama Arif
0 siblings, 1 reply; 7+ messages in thread
From: Paul E. McKenney @ 2026-06-18 14:57 UTC (permalink / raw)
To: Usama Arif
Cc: Randy Dunlap, Alan Stern, lkmm, joelagnelf, linux-kernel,
marco.crivellari, rafael.j.wysocki, riel, sshegde, tglx, ulfh,
yury.norov, rcu, shakeel.butt, hannes, kernel-team
On Thu, Jun 18, 2026 at 03:44:36PM +0100, Usama Arif wrote:
> On 18/06/2026 04:30, Randy Dunlap wrote:
> > On 6/17/26 6:44 PM, Alan Stern wrote:
> >> On Wed, Jun 17, 2026 at 02:20:01PM -0700, Usama Arif wrote:
> >>> __csd_lock_record() publishes per-CPU CSD debug state that is read by
> >>> csd_lock_wait_toolong() on another CPU. The remote side first reads
> >>> cur_csd with smp_load_acquire() and, when non-NULL, may then read the
> >>> matching cur_csd_func and cur_csd_info fields.
> >>>
> >>> Use smp_store_release() when publishing cur_csd so that the preceding
> >>> cur_csd_func and cur_csd_info stores are ordered before the pointer
> >>> that csd_lock_wait_toolong() acquires. This replaces the open-coded
> >>> smp_wmb() plus plain cur_csd store with the release operation that
> >>> matches the smp_load_acquire() in csd_lock_wait_toolong().
> >>>
> >>> For the clear path, use smp_store_release(&cur_csd, NULL) so that
> >>> clearing the diagnostic state remains ordered after the preceding
> >>> callback/unlock work, without requiring a full barrier before the
> >>> store. On x86 this removes the locked full barrier from the clear
> >>> path; on weaker memory models it uses the release operation needed by
> >>> the smp_load_acquire() in csd_lock_wait_toolong().
> >>>
> >>> The old code also had smp_mb() calls around cur_csd updates. Those would
> >>> only be needed if cur_csd were treated as an exact live-state marker whose
> >>> publication had to be observed before callback execution or CSD unlock.
> >>> CSD stall warnings do not currently have RCU-style stall-ended checks, so
> >>> they already allow the stall to end while diagnostics are being assembled.
> >>> The cur_csd record is therefore best-effort diagnostic context, not a
> >>> precise completion/stall boundary.
> >>>
> >>> Signed-off-by: Usama Arif <usama.arif@linux.dev>
> >>> ---
> >>> kernel/smp.c | 8 ++------
> >>> 1 file changed, 2 insertions(+), 6 deletions(-)
> >>>
> >>> diff --git a/kernel/smp.c b/kernel/smp.c
> >>> index a0bb56bd8dda..5ba4a20ba77d 100644
> >>> --- a/kernel/smp.c
> >>> +++ b/kernel/smp.c
> >>> @@ -182,16 +182,12 @@ static atomic_t csd_bug_count = ATOMIC_INIT(0);
> >>> static void __csd_lock_record(call_single_data_t *csd)
> >>> {
> >>> if (!csd) {
> >>> - smp_mb(); /* NULL cur_csd after unlock. */
> >>> - __this_cpu_write(cur_csd, NULL);
> >>> + smp_store_release(this_cpu_ptr(&cur_csd), NULL);
> >>> return;
> >>> }
> >>> __this_cpu_write(cur_csd_func, csd->func);
> >>> __this_cpu_write(cur_csd_info, csd->info);
> >>> - smp_wmb(); /* func and info before csd. */
> >>> - __this_cpu_write(cur_csd, csd);
> >>> - smp_mb(); /* Update cur_csd before function call. */
> >>> - /* Or before unlock, as the case may be. */
> >>> + smp_store_release(this_cpu_ptr(&cur_csd), csd);
> >>
> >> Isn't there a general policy in the kernel that memory barriers should
> >> be accompanied by a comment explaining what other memory barriers they
> >> synchronize with? Including such comments is a good idea in any case.
> >
> > in Documentation/process/submit-checklist.rst:
> >
> > 3) All memory barriers {e.g., ``barrier()``, ``rmb()``, ``wmb()``} need a
> > comment in the source code that explains the logic of what they are doing
> > and why.
> >
> > in Documentation/process/4.Coding.rst:
> >
> > Certain things should always be commented. Uses of memory barriers should
> > be accompanied by a line explaining why the barrier is necessary.
> >
> > but looking in the 3000+ lines of Documentation/memory-barriers.txt won't tell
> > anyone about that.
> >
> >
>
> Thanks!
> I will send a v2 with the below diff if there are no objections?
>
> diff --git a/kernel/smp.c b/kernel/smp.c
> index 5ba4a20ba77d..685829875a3e 100644
> --- a/kernel/smp.c
> +++ b/kernel/smp.c
> @@ -182,11 +182,21 @@ static atomic_t csd_bug_count = ATOMIC_INIT(0);
> static void __csd_lock_record(call_single_data_t *csd)
> {
> if (!csd) {
> + /*
> + * Pairs with smp_load_acquire() of cur_csd in
> + * csd_lock_wait_toolong(): orders any preceding CSD
> + * callback/unlock before a remote reader observes NULL.
> + */
Please replace the spaces with tabs. (Probably a copy-pasta issue.)
> smp_store_release(this_cpu_ptr(&cur_csd), NULL);
smp_store_release(this_cpu_ptr(&cur_csd), NULL); /* ^^^ */
Adding the comment as show above will satisfy tools such as checkpatch.
> return;
> }
> __this_cpu_write(cur_csd_func, csd->func);
> __this_cpu_write(cur_csd_info, csd->info);
> + /*
> + * Pairs with smp_load_acquire() of cur_csd in
> + * csd_lock_wait_toolong(): publishes cur_csd_func and
> + * cur_csd_info before the non-NULL pointer becomes visible.
> + */
> smp_store_release(this_cpu_ptr(&cur_csd), csd);
> }
The comments look good to me, but I must defer to Alan and Randy.
Thanx, Paul
^ permalink raw reply [flat|nested] 7+ messages in thread
* Re: [PATCH] smp: Use release stores for csd_lock_record() state
2026-06-18 14:57 ` Paul E. McKenney
@ 2026-06-18 15:30 ` Usama Arif
0 siblings, 0 replies; 7+ messages in thread
From: Usama Arif @ 2026-06-18 15:30 UTC (permalink / raw)
To: paulmck
Cc: Randy Dunlap, Alan Stern, lkmm, joelagnelf, linux-kernel,
marco.crivellari, rafael.j.wysocki, riel, sshegde, tglx, ulfh,
yury.norov, rcu, shakeel.butt, hannes, kernel-team
On 18/06/2026 15:57, Paul E. McKenney wrote:
> On Thu, Jun 18, 2026 at 03:44:36PM +0100, Usama Arif wrote:
>> On 18/06/2026 04:30, Randy Dunlap wrote:
>>> On 6/17/26 6:44 PM, Alan Stern wrote:
>>>> On Wed, Jun 17, 2026 at 02:20:01PM -0700, Usama Arif wrote:
>>>>> __csd_lock_record() publishes per-CPU CSD debug state that is read by
>>>>> csd_lock_wait_toolong() on another CPU. The remote side first reads
>>>>> cur_csd with smp_load_acquire() and, when non-NULL, may then read the
>>>>> matching cur_csd_func and cur_csd_info fields.
>>>>>
>>>>> Use smp_store_release() when publishing cur_csd so that the preceding
>>>>> cur_csd_func and cur_csd_info stores are ordered before the pointer
>>>>> that csd_lock_wait_toolong() acquires. This replaces the open-coded
>>>>> smp_wmb() plus plain cur_csd store with the release operation that
>>>>> matches the smp_load_acquire() in csd_lock_wait_toolong().
>>>>>
>>>>> For the clear path, use smp_store_release(&cur_csd, NULL) so that
>>>>> clearing the diagnostic state remains ordered after the preceding
>>>>> callback/unlock work, without requiring a full barrier before the
>>>>> store. On x86 this removes the locked full barrier from the clear
>>>>> path; on weaker memory models it uses the release operation needed by
>>>>> the smp_load_acquire() in csd_lock_wait_toolong().
>>>>>
>>>>> The old code also had smp_mb() calls around cur_csd updates. Those would
>>>>> only be needed if cur_csd were treated as an exact live-state marker whose
>>>>> publication had to be observed before callback execution or CSD unlock.
>>>>> CSD stall warnings do not currently have RCU-style stall-ended checks, so
>>>>> they already allow the stall to end while diagnostics are being assembled.
>>>>> The cur_csd record is therefore best-effort diagnostic context, not a
>>>>> precise completion/stall boundary.
>>>>>
>>>>> Signed-off-by: Usama Arif <usama.arif@linux.dev>
>>>>> ---
>>>>> kernel/smp.c | 8 ++------
>>>>> 1 file changed, 2 insertions(+), 6 deletions(-)
>>>>>
>>>>> diff --git a/kernel/smp.c b/kernel/smp.c
>>>>> index a0bb56bd8dda..5ba4a20ba77d 100644
>>>>> --- a/kernel/smp.c
>>>>> +++ b/kernel/smp.c
>>>>> @@ -182,16 +182,12 @@ static atomic_t csd_bug_count = ATOMIC_INIT(0);
>>>>> static void __csd_lock_record(call_single_data_t *csd)
>>>>> {
>>>>> if (!csd) {
>>>>> - smp_mb(); /* NULL cur_csd after unlock. */
>>>>> - __this_cpu_write(cur_csd, NULL);
>>>>> + smp_store_release(this_cpu_ptr(&cur_csd), NULL);
>>>>> return;
>>>>> }
>>>>> __this_cpu_write(cur_csd_func, csd->func);
>>>>> __this_cpu_write(cur_csd_info, csd->info);
>>>>> - smp_wmb(); /* func and info before csd. */
>>>>> - __this_cpu_write(cur_csd, csd);
>>>>> - smp_mb(); /* Update cur_csd before function call. */
>>>>> - /* Or before unlock, as the case may be. */
>>>>> + smp_store_release(this_cpu_ptr(&cur_csd), csd);
>>>>
>>>> Isn't there a general policy in the kernel that memory barriers should
>>>> be accompanied by a comment explaining what other memory barriers they
>>>> synchronize with? Including such comments is a good idea in any case.
>>>
>>> in Documentation/process/submit-checklist.rst:
>>>
>>> 3) All memory barriers {e.g., ``barrier()``, ``rmb()``, ``wmb()``} need a
>>> comment in the source code that explains the logic of what they are doing
>>> and why.
>>>
>>> in Documentation/process/4.Coding.rst:
>>>
>>> Certain things should always be commented. Uses of memory barriers should
>>> be accompanied by a line explaining why the barrier is necessary.
>>>
>>> but looking in the 3000+ lines of Documentation/memory-barriers.txt won't tell
>>> anyone about that.
>>>
>>>
>>
>> Thanks!
>> I will send a v2 with the below diff if there are no objections?
>>
>> diff --git a/kernel/smp.c b/kernel/smp.c
>> index 5ba4a20ba77d..685829875a3e 100644
>> --- a/kernel/smp.c
>> +++ b/kernel/smp.c
>> @@ -182,11 +182,21 @@ static atomic_t csd_bug_count = ATOMIC_INIT(0);
>> static void __csd_lock_record(call_single_data_t *csd)
>> {
>> if (!csd) {
>> + /*
>> + * Pairs with smp_load_acquire() of cur_csd in
>> + * csd_lock_wait_toolong(): orders any preceding CSD
>> + * callback/unlock before a remote reader observes NULL.
>> + */
>
> Please replace the spaces with tabs. (Probably a copy-pasta issue.)
Yes, its tabs in my commit, looks like the email client messed it up.
>
>> smp_store_release(this_cpu_ptr(&cur_csd), NULL);
>
> smp_store_release(this_cpu_ptr(&cur_csd), NULL); /* ^^^ */
>
> Adding the comment as show above will satisfy tools such as checkpatch.
Will do, Thanks!
>
>> return;
>> }
>> __this_cpu_write(cur_csd_func, csd->func);
>> __this_cpu_write(cur_csd_info, csd->info);
>> + /*
>> + * Pairs with smp_load_acquire() of cur_csd in
>> + * csd_lock_wait_toolong(): publishes cur_csd_func and
>> + * cur_csd_info before the non-NULL pointer becomes visible.
>> + */
>> smp_store_release(this_cpu_ptr(&cur_csd), csd);
>> }
>
> The comments look good to me, but I must defer to Alan and Randy
Thanks!
>
> Thanx, Paul
^ permalink raw reply [flat|nested] 7+ messages in thread
end of thread, other threads:[~2026-06-18 15:30 UTC | newest]
Thread overview: 7+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-06-17 21:20 [PATCH] smp: Use release stores for csd_lock_record() state Usama Arif
2026-06-18 1:44 ` Alan Stern
2026-06-18 3:30 ` Randy Dunlap
2026-06-18 4:01 ` Paul E. McKenney
2026-06-18 14:44 ` Usama Arif
2026-06-18 14:57 ` Paul E. McKenney
2026-06-18 15:30 ` Usama Arif
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.