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

bdev: retry IOs on ENOMEM from pull/append_copy

Browse Source 
The IOs will now be retried after ENOMEM is received when doing memory
domain pull or appending an accel copy.  The retries are performed using
the mechanism that's already in place for IOs completed with
SPDK_BDEV_IO_STATUS_NOMEM.

Signed-off-by: Konrad Sztyber <konrad.sztyber@intel.com>
Change-Id: I284643bf9971338094e14617974f7511f745f24e
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/17761
Community-CI: Mellanox Build Bot
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Ben Walker <benjamin.walker@intel.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This commit is contained in:
Konrad Sztyber 2023-04-19 15:35:46 +02:00 committed by Jim Harris
parent 7952ef88e0
commit 28bcf6a760
2 changed files with 86 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

62
lib/bdev/bdev.c
View File

@ -368,6 +368,8 @@ struct spdk_bdev_io_error_stat {
enum bdev_io_retry_state {
BDEV_IO_RETRY_STATE_INVALID,
BDEV_IO_RETRY_STATE_PULL,
BDEV_IO_RETRY_STATE_PULL_MD,
BDEV_IO_RETRY_STATE_SUBMIT,
};
@ -414,6 +416,8 @@ static bool claim_type_is_v2(enum spdk_bdev_claim_type type);
static void bdev_desc_release_claims(struct spdk_bdev_desc *desc);
static void claim_reset(struct spdk_bdev *bdev);
static void bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch);
#define bdev_get_ext_io_opt(opts, field, defval) \
(((opts) != NULL && offsetof(struct spdk_bdev_ext_io_opts, field) + \
sizeof((opts)->field) <= sizeof(*(opts))) ? (opts)->field : (defval))
@ -1061,6 +1065,11 @@ bdev_io_exec_sequence_cb(void *ctx, int status)
TAILQ_REMOVE(&bdev_io->internal.ch->io_accel_exec, bdev_io, internal.link);
bdev_io_decrement_outstanding(ch, ch->shared_resource);
if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
bdev_ch_retry_io(ch);
}
bdev_io->internal.data_transfer_cpl(bdev_io, status);
}
@ -1126,6 +1135,11 @@ bdev_io_pull_md_buf_done(void *ctx, int status)
TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
bdev_io_decrement_outstanding(ch, ch->shared_resource);
if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
bdev_ch_retry_io(ch);
}
assert(bdev_io->internal.data_transfer_cpl);
bdev_io->internal.data_transfer_cpl(bdev_io, status);
}
@ -1151,8 +1165,11 @@ bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
}
bdev_io_decrement_outstanding(ch, ch->shared_resource);
TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
spdk_memory_domain_get_dma_device_id(bdev_io->internal.memory_domain), rc);
if (rc != -ENOMEM) {
SPDK_ERRLOG("Failed to pull data from memory domain %s, rc %d\n",
spdk_memory_domain_get_dma_device_id(
bdev_io->internal.memory_domain), rc);
}
} else {
memcpy(bdev_io->internal.bounce_md_iov.iov_base,
bdev_io->internal.orig_md_iov.iov_base,
@ -1160,8 +1177,12 @@ bdev_io_pull_md_buf(struct spdk_bdev_io *bdev_io)
}
}
assert(bdev_io->internal.data_transfer_cpl);
bdev_io->internal.data_transfer_cpl(bdev_io, rc);
if (spdk_unlikely(rc == -ENOMEM)) {
bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL_MD);
} else {
assert(bdev_io->internal.data_transfer_cpl);
bdev_io->internal.data_transfer_cpl(bdev_io, rc);
}
}
static void
@ -1228,6 +1249,10 @@ _bdev_io_pull_bounce_data_buf_done(void *ctx, int status)
TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
bdev_io_decrement_outstanding(ch, ch->shared_resource);
if (spdk_unlikely(!TAILQ_EMPTY(&ch->shared_resource->nomem_io))) {
bdev_ch_retry_io(ch);
}
bdev_io_pull_bounce_data_buf_done(ctx, status);
}
@ -1261,7 +1286,7 @@ bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
NULL, NULL, 0, NULL, NULL);
}
if (spdk_unlikely(rc != 0)) {
if (spdk_unlikely(rc != 0 && rc != -ENOMEM)) {
SPDK_ERRLOG("Failed to append copy to accel sequence: %p\n",
bdev_io->internal.accel_sequence);
}
@ -1283,8 +1308,11 @@ bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
}
TAILQ_REMOVE(&ch->io_memory_domain, bdev_io, internal.link);
bdev_io_decrement_outstanding(ch, ch->shared_resource);
SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
spdk_memory_domain_get_dma_device_id(bdev_io->internal.memory_domain));
if (rc != -ENOMEM) {
SPDK_ERRLOG("Failed to pull data from memory domain %s\n",
spdk_memory_domain_get_dma_device_id(
bdev_io->internal.memory_domain));
}
} else {
assert(bdev_io->u.bdev.iovcnt == 1);
spdk_copy_iovs_to_buf(bdev_io->u.bdev.iovs[0].iov_base,
@ -1294,13 +1322,19 @@ bdev_io_pull_data(struct spdk_bdev_io *bdev_io)
}
}
bdev_io_pull_bounce_data_buf_done(bdev_io, rc);
if (spdk_unlikely(rc == -ENOMEM)) {
bdev_queue_nomem_io_head(ch->shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
} else {
bdev_io_pull_bounce_data_buf_done(bdev_io, rc);
}
}
static void
_bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t len,
bdev_copy_bounce_buffer_cpl cpl_cb)
{
struct spdk_bdev_shared_resource *shared_resource = bdev_io->internal.ch->shared_resource;
bdev_io->internal.data_transfer_cpl = cpl_cb;
/* save original iovec */
bdev_io->internal.orig_iovs = bdev_io->u.bdev.iovs;
@ -1312,7 +1346,11 @@ _bdev_io_pull_bounce_data_buf(struct spdk_bdev_io *bdev_io, void *buf, size_t le
bdev_io->u.bdev.iovs[0].iov_base = buf;
bdev_io->u.bdev.iovs[0].iov_len = len;
bdev_io_pull_data(bdev_io);
if (spdk_unlikely(!TAILQ_EMPTY(&shared_resource->nomem_io))) {
bdev_queue_nomem_io_tail(shared_resource, bdev_io, BDEV_IO_RETRY_STATE_PULL);
} else {
bdev_io_pull_data(bdev_io);
}
}
static void
@ -1438,6 +1476,12 @@ bdev_ch_retry_io(struct spdk_bdev_channel *bdev_ch)
case BDEV_IO_RETRY_STATE_SUBMIT:
bdev_ch_resubmit_io(bdev_ch, bdev_io);
break;
case BDEV_IO_RETRY_STATE_PULL:
bdev_io_pull_data(bdev_io);
break;
case BDEV_IO_RETRY_STATE_PULL_MD:
bdev_io_pull_md_buf(bdev_io);
break;
default:
assert(0 && "invalid retry state");
break;

34
test/unit/lib/bdev/bdev.c/bdev_ut.c
View File

@ -5795,7 +5795,7 @@ bdev_io_ext_bounce_buffer(void)
struct spdk_io_channel *io_ch;
char io_buf[512];
struct iovec iov = { .iov_base = io_buf, .iov_len = 512 };
struct ut_expected_io *expected_io;
struct ut_expected_io *expected_io, *aux_io;
struct spdk_bdev_ext_io_opts ext_io_opts = {
.metadata = (void *)0xFF000000,
.size = sizeof(ext_io_opts)
@ -5849,6 +5849,38 @@ bdev_io_ext_bounce_buffer(void)
stub_complete_io(1);
CU_ASSERT(g_io_done == true);
/* Verify the request is queued after receiving ENOMEM from pull */
g_io_done = false;
aux_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 1);
ut_expected_io_set_iov(aux_io, 0, iov.iov_base, iov.iov_len);
TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, aux_io, link);
rc = spdk_bdev_writev_blocks(desc, io_ch, &iov, 1, 32, 14, io_done, NULL);
CU_ASSERT(rc == 0);
CU_ASSERT(g_io_done == false);
CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
expected_io = ut_alloc_expected_io(SPDK_BDEV_IO_TYPE_WRITE, 32, 14, 1);
ut_expected_io_set_iov(expected_io, 0, iov.iov_base, iov.iov_len);
TAILQ_INSERT_TAIL(&g_bdev_ut_channel->expected_io, expected_io, link);
MOCK_SET(spdk_memory_domain_pull_data, -ENOMEM);
rc = spdk_bdev_writev_blocks_ext(desc, io_ch, &iov, 1, 32, 14, io_done, NULL, &ext_io_opts);
CU_ASSERT(rc == 0);
CU_ASSERT(g_io_done == false);
/* The second IO has been queued */
CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
MOCK_CLEAR(spdk_memory_domain_pull_data);
g_memory_domain_pull_data_called = false;
stub_complete_io(1);
CU_ASSERT(g_io_done == true);
CU_ASSERT(g_memory_domain_pull_data_called == true);
/* The second IO should be submitted now */
CU_ASSERT(g_bdev_ut_channel->outstanding_io_count == 1);
g_io_done = false;
stub_complete_io(1);
CU_ASSERT(g_io_done == true);
spdk_put_io_channel(io_ch);
spdk_bdev_close(desc);
free_bdev(bdev);