ivampiresp/Spdk
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nvme: add tracker prefetching in completion path

Browse Source 
At 10M IO/s, we see a lot of CPU cycles wasted getting
the next tracker into cache.  If we only get one
completion at a time, this is unavoidable, but when
there are multiple completions pending, we can prefetch
the second tracker while processing the completion for
the first.

Signed-off-by: Jim Harris <james.r.harris@intel.com>
Change-Id: I9de702bee3719e4494eec6f05b09be3672f1e0ac

Reviewed-on: https://review.gerrithub.io/c/spdk/spdk/+/456097
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
This commit is contained in:
Jim Harris 2019-05-28 08:52:36 -07:00
parent f2d46446ca
commit 6c820f84cb
1 changed files with 19 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

21
lib/nvme/nvme_pcie.c
View File

@ -2048,9 +2048,12 @@ nvme_pcie_qpair_process_completions(struct spdk_nvme_qpair *qpair, uint32_t max_
{
struct nvme_pcie_qpair *pqpair = nvme_pcie_qpair(qpair);
struct nvme_tracker *tr;
struct spdk_nvme_cpl *cpl;
struct spdk_nvme_cpl *cpl, *next_cpl;
uint32_t num_completions = 0;
struct spdk_nvme_ctrlr *ctrlr = qpair->ctrlr;
uint16_t next_cq_head;
uint8_t next_phase;
bool next_is_valid = false;
if (spdk_unlikely(nvme_qpair_is_admin_queue(qpair))) {
nvme_robust_mutex_lock(&ctrlr->ctrlr_lock);
@ -2068,9 +2071,23 @@ nvme_pcie_qpair_process_completions(struct spdk_nvme_qpair *qpair, uint32_t max_
while (1) {
cpl = &pqpair->cpl[pqpair->cq_head];
if (cpl->status.p != pqpair->flags.phase) {
if (!next_is_valid && cpl->status.p != pqpair->flags.phase) {
break;
}
if (spdk_likely(pqpair->cq_head + 1 != pqpair->num_entries)) {
next_cq_head = pqpair->cq_head + 1;
next_phase = pqpair->flags.phase;
} else {
next_cq_head = 0;
next_phase = !pqpair->flags.phase;
}
next_cpl = &pqpair->cpl[next_cq_head];
next_is_valid = (next_cpl->status.p == next_phase);
if (next_is_valid) {
__builtin_prefetch(&pqpair->tr[next_cpl->cid]);
}
#ifdef __PPC64__
/*
* This memory barrier prevents reordering of: