diff options
| author | Diogo Ferreira <diogo@underdev.org> | 2016-11-23 14:46:50 +0000 |
|---|---|---|
| committer | Mister Oyster <oysterized@gmail.com> | 2016-12-11 11:53:50 +0100 |
| commit | 3d9f3ece074ad2d49551c011b05ee0d439921439 (patch) | |
| tree | 4e6794c58158c2c13ae89a7d65016b86b15f145e /block | |
| parent | b0be10a223b90c38309e8a6d3cd2ee12e45727cc (diff) | |
bfq-sched: Forcefully lookup entities when the cache is inconsistent
bfq maintains a 'next-in-service' cache to prevent expensive lookups in
the hot path. However, the cache sometimes becomes inconsistent and
triggers a BUG:
[44042.622839] -(3)[154:mmcqd/0]BUG: failure at ../../../../../../kernel/cyanogen/mt6735/block/bfq-sched.c:72/bfq_check_next_in_service()!
[44042.622858] -(3)[154:mmcqd/0]Unable to handle kernel paging request at virtual address 0000dead
[44042.622866] -(3)[154:mmcqd/0]pgd = ffffffc001361000
[44042.622872] [0000dead] *pgd=000000007d816003, *pud=000000007d816003, *pmd=000000007d817003, *pte=0000000000000000
[44042.622890] -(3)[154:mmcqd/0]Internal error: Oops: 96000045 [#1] PREEMPT SMP
[44042.622907] -(3)[154:mmcqd/0]CPU: 3 PID: 154 Comm: mmcqd/0 Tainted:
[44042.622915] -(3)[154:mmcqd/0]Hardware name: MT6735 (DT)
[44042.622922] -(3)[154:mmcqd/0]task: ffffffc0378a6000 ti: ffffffc0378c4000
[44042.622936] -(3)[154:mmcqd/0]PC is at bfq_dispatch_requests+0x6c4/0x9bc
[44042.622944] -(3)[154:mmcqd/0]LR is at bfq_dispatch_requests+0x6bc/0x9bc
[44042.622952] -(3)[154:mmcqd/0]pc : [<ffffffc000306a68>] lr : [<ffffffc000306a60>] pstate: 800001c5
[44042.622958] -(3)[154:mmcqd/0]sp : ffffffc0378c7d30
[44042.622962] x29: ffffffc0378c7d30 x28: 0000000000000000
[44042.622972] x27: 0000000000000000 x26: ffffffc006c58810
[44042.622981] x25: ffffffc037f89820 x24: ffffffc000f14000
[44042.622990] x23: ffffffc036adb088 x22: ffffffc0369b2800
[44042.623000] x21: ffffffc036adb098 x20: ffffffc01d6a3b60
[44042.623009] x19: ffffffc036adb0c8 x18: 0000007f8cfa1500
[44042.623018] x17: 0000007f8db44f40 x16: ffffffc00012d0c0
[44042.623027] x15: 0000007f8dde04d8 x14: 676f6e6179632f6c
[44042.623037] x13: 656e72656b2f2e2e x12: 2f2e2e2f2e2e2f2e
[44042.623046] x11: 2e2f2e2e2f2e2e20 x10: 7461206572756c69
[44042.623055] x9 : 6166203a4755425d x8 : 00000000001f0cc5
[44042.623064] x7 : ffffffc000f3d5a0 x6 : 000000000000008b
[44042.623073] x5 : 0000000000000000 x4 : 0000000000000004
[44042.623082] x3 : 0000000000000002 x2 : 0000000000000001
[44042.623091] x1 : 0000000000000aee x0 : 000000000000dead
This patch makes the lookup resilient to cache inconsistencies by doing
the expensive recomputation in cases where the bug would otherwise be
triggered.
Ticket: PORRDIGE-527
Change-Id: I5dd701960057983a42d3d3bd57521e8d17c03d7f
Diffstat (limited to 'block')
| -rw-r--r-- | block/bfq-sched.c | 3 |
1 files changed, 3 insertions, 0 deletions
diff --git a/block/bfq-sched.c b/block/bfq-sched.c index d0890c6d4..c3828a45b 100644 --- a/block/bfq-sched.c +++ b/block/bfq-sched.c @@ -1046,6 +1046,9 @@ static struct bfq_entity *bfq_lookup_next_entity(struct bfq_sched_data *sd, entity = __bfq_lookup_next_entity(st + i, false); if (entity != NULL) { if (extract) { + if (sd->next_in_service != entity) { + entity = __bfq_lookup_next_entity(st + i, true); + } bfq_check_next_in_service(sd, entity); bfq_active_extract(st + i, entity); sd->in_service_entity = entity; |