Re: [PATCH v2] smp: Use release stores for csd_lock_record() state

2026=E5=B9=B46=E6=9C=8824=E6=97=A5 23:27, "Usama Arif" <usama.arif@linux.=
dev mailto:[email protected]?to=3D%22Usama%20Arif%22%20%3Cusama.arif%4=
0linux.dev%3E > =E5=86=99=E5=88=B0:


>=20
>=20On 24/06/2026 15:15, Kunwu Chan wrote:
>=20
>=20>=20
>=20> On 6/23/26 00:38, Usama Arif wrote:
> >=20
>=20> >=20
>=20> > __csd_lock_record() publishes per-CPU CSD debug state that is rea=
d by
> > >  csd_lock_wait_toolong() on another CPU. The remote side first read=
s
> > >  cur_csd with smp_load_acquire() and, when non-NULL, may then read =
the
> > >  matching cur_csd_func and cur_csd_info fields.
> > >=20
>=20> >  Use smp_store_release() when publishing cur_csd so that the prec=
eding
> > >  cur_csd_func and cur_csd_info stores are ordered before the pointe=
r
> > >  that csd_lock_wait_toolong() acquires. This replaces the open-code=
d
> > >  smp_wmb() plus plain cur_csd store with the release operation that
> > >  matches the smp_load_acquire() in csd_lock_wait_toolong().
> > >=20
>=20> >  For the clear path, use smp_store_release(&cur_csd, NULL) so tha=
t
> > >  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().
> > >=20
>=20> >  The old code also had smp_mb() calls around cur_csd updates. Tho=
se would
> > >  only be needed if cur_csd were treated as an exact live-state mark=
er whose
> > >  publication had to be observed before callback execution or CSD un=
lock.
> > >=20
>=20>  =C2=A0The original comments around those updates seem to suggest a=
 stronger
> >  relationship:
> >  =C2=A0 /* Update cur_csd before function call. */
> >  =C2=A0 /* NULL cur_csd after unlock. */
> >=20=20
>=20>  The changelog characterizes cur_csd as best-effort diagnostic cont=
ext.=C2=A0
> >=20=20
>=20>  Is there prior consensus that cur_csd carries no ordering relation=
ship=C2=A0
> >  to callback execution / unlock state,=C2=A0
> >=20=20
>=20>  or is this patch effectively relaxing that historical assumption?
> >=20=20
>=20>=20=20
>=20>  Thanks,=C2=A0 Kunwu
> >=20
>=20Thanks for the review Kunwu!
>=20
>=20The patch preserves the ordering that reader actually needs:=20
>=20csd_lock_wait_toolong() observes non-NULL cur_csd, the matching func/=
info have
> been published. If it observes NULL, the prior callback/unlock work is =
ordered
> before the clear.
>=20
>=20So yes, this relaxes the historical implementation ordering around th=
e
> diagnostic marker, but not the CSD execution/reuse ordering.
>=20


Thanks=20for the v3 update. The explanation of reader-side tolerance for=
=20
snapshot-vs-completion=20races addresses my concern.


> >=20
>=20> >=20
>=20> > CSD stall warnings do not currently have RCU-style stall-ended ch=
ecks, so
> > >  they already allow the stall to end while diagnostics are being as=
sembled.
> > >  The cur_csd record is therefore best-effort diagnostic context, no=
t a
> > >  precise completion/stall boundary.
> > >=20
>=20> >  Signed-off-by: Usama Arif <[email protected]>
> > >  ---
> > >  v1 -> v2: https://lore.kernel.org/all/01437928-ff79-4d8e-823b-7f20=
[email protected]/
> > >  - Document where the smp_store_release() synchronizes with (Alan S=
tern,
> > >  Randy Dunlap and Paul McKenney).
> > >  ---
> > >  kernel/smp.c | 18 ++++++++++++------
> > >  1 file changed, 12 insertions(+), 6 deletions(-)
> > >=20
>=20> >  diff --git a/kernel/smp.c b/kernel/smp.c
> > >  index a0bb56bd8dda..685829875a3e 100644
> > >  --- a/kernel/smp.c
> > >  +++ b/kernel/smp.c
> > >  @@ -182,16 +182,22 @@ static atomic_t csd_bug_count =3D ATOMIC_INI=
T(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);
> > >  + /*
> > >  + * 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);
> > >  - 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. */
> > >  + /*
> > >  + * 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);
> > >  }
> > >=20=20
>=20> >  static __always_inline void csd_lock_record(call_single_data_t *=
csd)
> > >
> >
>