diff options
| author | cinap_lenrek <cinap_lenrek@felloff.net> | 2020-10-18 02:51:32 +0200 |
|---|---|---|
| committer | cinap_lenrek <cinap_lenrek@felloff.net> | 2020-10-18 02:51:32 +0200 |
| commit | cf8ff0e71369afdf600e36924c2d046ecf783cbf (patch) | |
| tree | 248cf8379c175c580a9552104808bf12a25ef081 | |
| parent | 5fd2e746e1a2a00d2acbc0c3b1924b1ea22101e8 (diff) | |
| download | plan9front-cf8ff0e71369afdf600e36924c2d046ecf783cbf.tar.xz | |
sdnvme: handle machines with more cpu's than submit queues (thanks mischief)
We used to assume a 1:1 pairing of processors to submit queues.
With recent machines, we now got more cpu cores than what some
nvme drives support so we need to distribute the queues across
these cpu's which requires locking on command submission.
There is a feature get/set command to probe the number of submit
and completion queues, but we decided to just handling
submission queue create command error gracefully as it is simpler
and has less chance of regression with existing setups.
Thanks to mischief for investigating and writing the code.
| -rw-r--r-- | sys/src/9/pc/sdnvme.c | 30 |
1 files changed, 25 insertions, 5 deletions
diff --git a/sys/src/9/pc/sdnvme.c b/sys/src/9/pc/sdnvme.c index 152dba187..1909eabdb 100644 --- a/sys/src/9/pc/sdnvme.c +++ b/sys/src/9/pc/sdnvme.c @@ -41,6 +41,7 @@ struct SQ u32int *base; WS **wait; Ctlr *ctlr; + Lock; }; struct Ctlr @@ -63,6 +64,8 @@ struct Ctlr u32int mpsshift; u32int dstrd; + u32int nsq; + CQ cq[1+1]; SQ sq[1+MAXMACH]; @@ -99,7 +102,9 @@ qcmd(WS *ws, Ctlr *ctlr, int adm, u32int opc, u32int nsid, void *mptr, void *dat if(!adm){ Retry: splhi(); - sq = &ctlr->sq[1+m->machno]; + sq = &ctlr->sq[1+(m->machno % ctlr->nsq)]; + if(conf.nmach > ctlr->nsq) + lock(sq); } else { qlock(ctlr); sq = &ctlr->sq[0]; @@ -207,7 +212,9 @@ wcmd(WS *ws) coherence(); ctlr->reg[DBell + ((sq-ctlr->sq)*2+0 << ctlr->dstrd)] = sq->tail & sq->mask; if(sq > ctlr->sq) { - assert(sq == &ctlr->sq[1+m->machno]); + assert(sq == &ctlr->sq[1+(m->machno % ctlr->nsq)]); + if(conf.nmach > ctlr->nsq) + unlock(sq); spllo(); } else qunlock(sq->ctlr); @@ -381,7 +388,7 @@ sqalloc(Ctlr *ctlr, SQ *sq, u32int lgsize) static void setupqueues(Ctlr *ctlr) { - u32int lgsize, *e; + u32int lgsize, st, *e; CQ *cq; SQ *sq; WS ws; @@ -400,6 +407,8 @@ setupqueues(Ctlr *ctlr) e[11] = 3; /* IEN | PC */ checkstatus(wcmd(&ws), "create completion queue"); + st = 0; + /* SQID[1..nmach]: submission queue per cpu */ for(i=1; i<=conf.nmach; i++){ sq = &ctlr->sq[i]; @@ -407,8 +416,19 @@ setupqueues(Ctlr *ctlr) e = qcmd(&ws, ctlr, 1, 0x01, 0, nil, sq->base, 0x1000); e[10] = i | sq->mask<<16; e[11] = (cq - ctlr->cq)<<16 | 1; /* CQID<<16 | PC */ - checkstatus(wcmd(&ws), "create submission queue"); + + st = wcmd(&ws); + if(st != 0){ + free(sq->base); + free(sq->wait); + memset(sq, 0, sizeof(*sq)); + break; + } } + + ctlr->nsq = i - 1; + if(ctlr->nsq < 1) + checkstatus(st, "create submission queues"); ilock(&ctlr->intr); ctlr->ints |= 1<<(cq - ctlr->cq); @@ -544,7 +564,7 @@ nvmeenable(SDev *sd) Ready: identify(ctlr); setupqueues(ctlr); - + print("%s: using %d submit queues\n", name, ctlr->nsq); poperror(); return 1; |
