| History log of /src/sys/kern/subr_pcq.c |
| Revision | | Date | Author | Comments |
| 1.20 |
| 24-Feb-2023 |
riastradh | kern: Eliminate most __HAVE_ATOMIC_AS_MEMBAR conditionals.
I'm leaving in the conditional around the legacy membar_enters
(store-before-load, store-before-store) in kern_mutex.c and in
kern_lock.c because they may still matter: store-before-load barriers
tend to be the most expensive kind, so eliding them is probably
worthwhile on x86. (It also may not matter; I just don't care to do
measurements right now, and it's a single valid and potentially
justifiable use case in the whole tree.)
However, membar_release/acquire can be mere instruction barriers on
all TSO platforms including x86, so there's no need to go out of our
way with a bad API to conditionalize them. If the procedure call
overhead is measurable we just could change them to be macros on x86
that expand into __insn_barrier.
Discussed on tech-kern:
https://mail-index.netbsd.org/tech-kern/2023/02/23/msg028729.html
|
| 1.19 |
| 23-Feb-2023 |
riastradh | pcq(9): KASSERT(A && B) -> KASSERT(A); KASSERT(B)
|
| 1.18 |
| 23-Feb-2023 |
riastradh | pcq(9): Sketch correctness proof for some critical properties.
No functional change intended.
|
| 1.17 |
| 23-Feb-2023 |
riastradh | pcq(9): Explain why membar_release isn't needed in pcq_get.
No functional change intended.
|
| 1.16 |
| 23-Feb-2023 |
riastradh | pcq(9): Explain why store need not be atomic in pcq_get.
No functional change intended.
|
| 1.15 |
| 23-Feb-2023 |
riastradh | pcq(9): Fix consume operation in pcq_peek/get.
These use atomic_load_consume to match the atomic_store_release in
pcq_put for pcq->pcq_items[c].
Reading the snapshot of pcq->pcq_pc need not be ordered:
- The consumer side (pcq_peek/get) is serialized by the API contract
(single-consumer, multi-producer), so no ordering is necessary.
- The producer side updates pcq->pcq_pc first; if the consumer side
sees that before the producer side has stored pcq->pcq_items[c],
there's no problem -- it's as if the consumer had happened just a
moment earlier and the producer hadn't entered pcq_put yet.
However, it should be an atomic load, not a plain load. So use
atomic_load_relaxed, if for no other reason than to pacify thread
sanitizers.
XXX pullup-9
XXX pullup-10
|
| 1.14 |
| 23-Feb-2023 |
riastradh | pcq(9): Make pcq_put a release operation, in memory ordering.
Otherwise, for example, the following assertion could fail:
/* publisher */
nusers = foo->nusers;
pcq_put(pcq, foo);
KASSERT(nusers == 0);
/* user */
foo = pcq_get(pcq);
if (foo != NULL)
atomic_inc_uint(&foo->nusers);
XXX pullup-9
XXX pullup-10
|
| 1.13 |
| 08-Feb-2021 |
wiz | branches: 1.13.18;
Fix typo in comment.
|
| 1.12 |
| 12-Jan-2021 |
skrll | Typo in comment
|
| 1.11 |
| 16-Dec-2019 |
ad | branches: 1.11.8;
pcq_create(): fix broken assertion.
|
| 1.10 |
| 08-Feb-2018 |
dholland | branches: 1.10.4;
Typos.
|
| 1.9 |
| 08-Jan-2015 |
riastradh | Use membar_datadep_consumer where code inspection reveals it necessary.
|
| 1.8 |
| 09-Jun-2014 |
rmind | branches: 1.8.4;
Add PCQ_MAXLEN constant.
|
| 1.7 |
| 06-Feb-2014 |
riastradh | branches: 1.7.2;
__HAVE_ATOMIC_AS_MEMBAR is spelled with two leading underscores.
This underscores the need to replace this error-prone cpp API by
unconditionally defined {pre,post}atomic_membar_*.
This change should only remove unnecessary membar_producers on x86.
|
| 1.6 |
| 31-Jan-2012 |
alnsn | branches: 1.6.2; 1.6.6; 1.6.10;
Revert to more readable but non-standard use of offsetof to calculate
a size of a structure with a flexible array member.
|
| 1.5 |
| 31-Jan-2012 |
alnsn | Replace offsetof(pcq_t, pcq_items[nitems]) with sizeof(pcq_t) + sizeof(void *[nitems]).
|
| 1.4 |
| 22-Jan-2012 |
rmind | Replace pcq(9) with the implementation from ad@ and minor changes by me.
PR/40516, PR/45631.
|
| 1.3 |
| 11-Nov-2008 |
rmind | branches: 1.3.4; 1.3.6; 1.3.8; 1.3.12; 1.3.24; 1.3.28;
- Optimise by avoiding few memory barriers, when atomic operations
performs that for us. OK by <matt>.
- Add RCS ID, and a bit of KNF.
|
| 1.2 |
| 11-Nov-2008 |
snj | Fix a couple spelling mistakes in comments.
|
| 1.1 |
| 11-Nov-2008 |
matt | Add generic producer/consumer queue code.
|
| 1.3.28.1 |
| 18-Feb-2012 |
mrg | merge to -current.
|
| 1.3.24.2 |
| 22-May-2014 |
yamt | sync with head.
for a reference, the tree before this commit was tagged
as yamt-pagecache-tag8.
this commit was splitted into small chunks to avoid
a limitation of cvs. ("Protocol error: too many arguments")
|
| 1.3.24.1 |
| 17-Apr-2012 |
yamt | sync with head
|
| 1.3.12.2 |
| 04-May-2009 |
yamt | sync with head.
|
| 1.3.12.1 |
| 11-Nov-2008 |
yamt | file subr_pcq.c was added on branch yamt-nfs-mp on 2009-05-04 08:13:48 +0000
|
| 1.3.8.2 |
| 19-Jan-2009 |
skrll | Sync with HEAD.
|
| 1.3.8.1 |
| 11-Nov-2008 |
skrll | file subr_pcq.c was added on branch nick-hppapmap on 2009-01-19 13:19:39 +0000
|
| 1.3.6.2 |
| 17-Jan-2009 |
mjf | Sync with HEAD.
|
| 1.3.6.1 |
| 11-Nov-2008 |
mjf | file subr_pcq.c was added on branch mjf-devfs2 on 2009-01-17 13:29:19 +0000
|
| 1.3.4.2 |
| 13-Dec-2008 |
haad | Update haad-dm branch to haad-dm-base2.
|
| 1.3.4.1 |
| 11-Nov-2008 |
haad | file subr_pcq.c was added on branch haad-dm on 2008-12-13 01:15:08 +0000
|
| 1.6.10.1 |
| 18-May-2014 |
rmind | sync with head
|
| 1.6.6.2 |
| 03-Dec-2017 |
jdolecek | update from HEAD
|
| 1.6.6.1 |
| 20-Aug-2014 |
tls | Rebase to HEAD as of a few days ago.
|
| 1.6.2.1 |
| 27-Oct-2014 |
msaitoh | Pull up following revision(s) (requested by riastradh in ticket #1132):
sys/kern/subr_pcq.c: revision 1.7
__HAVE_ATOMIC_AS_MEMBAR is spelled with two leading underscores.
This underscores the need to replace this error-prone cpp API by
unconditionally defined {pre,post}atomic_membar_*.
This change should only remove unnecessary membar_producers on x86.
|
| 1.7.2.1 |
| 10-Aug-2014 |
tls | Rebase.
|
| 1.8.4.1 |
| 06-Apr-2015 |
skrll | Sync with HEAD
|
| 1.10.4.1 |
| 08-Apr-2020 |
martin | Merge changes from current as of 20200406
|
| 1.11.8.1 |
| 03-Apr-2021 |
thorpej | Sync with HEAD.
|
| 1.13.18.1 |
| 30-Jul-2023 |
martin | Pull up following revision(s) (requested by riastradh in ticket #263):
share/man/man9/pcq.9: revision 1.9
sys/kern/subr_pcq.c: revision 1.14
sys/kern/subr_pcq.c: revision 1.15
sys/kern/subr_pcq.c: revision 1.16
sys/kern/subr_pcq.c: revision 1.17
sys/kern/subr_pcq.c: revision 1.18
sys/kern/subr_pcq.c: revision 1.19
pcq(9): Make pcq_put a release operation, in memory ordering.
Otherwise, for example, the following assertion could fail:
/* publisher */
nusers = foo->nusers;
pcq_put(pcq, foo);
KASSERT(nusers == 0);
/* user */
foo = pcq_get(pcq);
if (foo != NULL)
atomic_inc_uint(&foo->nusers);
pcq(9): Fix consume operation in pcq_peek/get.
These use atomic_load_consume to match the atomic_store_release in
pcq_put for pcq->pcq_items[c].
Reading the snapshot of pcq->pcq_pc need not be ordered:
- The consumer side (pcq_peek/get) is serialized by the API contract
(single-consumer, multi-producer), so no ordering is necessary.
- The producer side updates pcq->pcq_pc first; if the consumer side
sees that before the producer side has stored pcq->pcq_items[c],
there's no problem -- it's as if the consumer had happened just a
moment earlier and the producer hadn't entered pcq_put yet.
However, it should be an atomic load, not a plain load. So use
atomic_load_relaxed, if for no other reason than to pacify thread
sanitizers.
pcq(9): Explain why store need not be atomic in pcq_get.
No functional change intended.
pcq(9): Explain why membar_release isn't needed in pcq_get.
No functional change intended.
pcq(9): Document memory ordering guarantees.
pcq(9): Sketch correctness proof for some critical properties.
No functional change intended.
pcq(9): KASSERT(A && B) -> KASSERT(A); KASSERT(B)
|
