The following changes since commit 41508de67c06661ff1d473d108a8a01912ade114:
fio/server: fix confusing sk_out check (2023-07-03 09:16:45 -0600)
are available in the Git repository at:
git://git.kernel.dk/fio.git master
for you to fetch changes up to 8e2b81b854286f32eae7951a434dddebd968f9d5:
zbd: Support finishing zones on Android (2023-07-05 15:48:11 -0600)
----------------------------------------------------------------
Bart Van Assche (1):
zbd: Support finishing zones on Android
Jens Axboe (1):
Merge branch 'makefile-hardening-cpp-flags' of https://github.com/proact-de/fio
Martin Steigerwald (1):
Keep C pre processor hardening build flags.
Vincent Fu (4):
engines/io_uring_cmd: make trims async
engines/io_uring: remove dead code related to trim
t/nvmept: add check for iodepth
t/nvmept: add trim test with ioengine options enabled
Makefile | 2 +-
engines/io_uring.c | 49 ++++++++++----------------
engines/nvme.c | 96 ++++++++++++++++++++++++--------------------------
engines/nvme.h | 5 +--
oslib/linux-blkzoned.c | 24 ++++++-------
t/nvmept.py | 21 +++++++++++
6 files changed, 100 insertions(+), 97 deletions(-)
---
Diff of recent changes:
diff --git a/Makefile b/Makefile
index 6d7fd4e2..cc8164b2 100644
--- a/Makefile
+++ b/Makefile
@@ -20,7 +20,7 @@ include config-host.mak
endif
DEBUGFLAGS = -DFIO_INC_DEBUG
-CPPFLAGS= -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -DFIO_INTERNAL $(DEBUGFLAGS)
+CPPFLAGS+= -D_LARGEFILE_SOURCE -D_FILE_OFFSET_BITS=64 -DFIO_INTERNAL $(DEBUGFLAGS)
OPTFLAGS= -g -ffast-math
FIO_CFLAGS= -std=gnu99 -Wwrite-strings -Wall -Wdeclaration-after-statement $(OPTFLAGS) $(EXTFLAGS) $(BUILD_CFLAGS) -I. -I$(SRCDIR)
LIBS += -lm $(EXTLIBS)
diff --git a/engines/io_uring.c b/engines/io_uring.c
index 73e4a27a..5021239e 100644
--- a/engines/io_uring.c
+++ b/engines/io_uring.c
@@ -78,6 +78,8 @@ struct ioring_data {
struct ioring_mmap mmap[3];
struct cmdprio cmdprio;
+
+ struct nvme_dsm_range *dsm;
};
struct ioring_options {
@@ -410,7 +412,7 @@ static int fio_ioring_cmd_prep(struct thread_data *td, struct io_u *io_u)
if (o->cmd_type != FIO_URING_CMD_NVME)
return -EINVAL;
- if (io_u->ddir == DDIR_TRIM)
+ if (io_u->ddir == DDIR_TRIM && td->io_ops->flags & FIO_ASYNCIO_SYNC_TRIM)
return 0;
sqe = &ld->sqes[(io_u->index) << 1];
@@ -444,7 +446,8 @@ static int fio_ioring_cmd_prep(struct thread_data *td, struct io_u *io_u)
cmd = (struct nvme_uring_cmd *)sqe->cmd;
return fio_nvme_uring_cmd_prep(cmd, io_u,
- o->nonvectored ? NULL : &ld->iovecs[io_u->index]);
+ o->nonvectored ? NULL : &ld->iovecs[io_u->index],
+ &ld->dsm[io_u->index]);
}
static struct io_u *fio_ioring_event(struct thread_data *td, int event)
@@ -561,27 +564,6 @@ static inline void fio_ioring_cmdprio_prep(struct thread_data *td,
ld->sqes[io_u->index].ioprio = io_u->ioprio;
}
-static int fio_ioring_cmd_io_u_trim(struct thread_data *td,
- struct io_u *io_u)
-{
- struct fio_file *f = io_u->file;
- int ret;
-
- if (td->o.zone_mode == ZONE_MODE_ZBD) {
- ret = zbd_do_io_u_trim(td, io_u);
- if (ret == io_u_completed)
- return io_u->xfer_buflen;
- if (ret)
- goto err;
- }
-
- return fio_nvme_trim(td, f, io_u->offset, io_u->xfer_buflen);
-
-err:
- io_u->error = ret;
- return 0;
-}
-
static enum fio_q_status fio_ioring_queue(struct thread_data *td,
struct io_u *io_u)
{
@@ -594,14 +576,11 @@ static enum fio_q_status fio_ioring_queue(struct thread_data *td,
if (ld->queued == ld->iodepth)
return FIO_Q_BUSY;
- if (io_u->ddir == DDIR_TRIM) {
+ if (io_u->ddir == DDIR_TRIM && td->io_ops->flags & FIO_ASYNCIO_SYNC_TRIM) {
if (ld->queued)
return FIO_Q_BUSY;
- if (!strcmp(td->io_ops->name, "io_uring_cmd"))
- fio_ioring_cmd_io_u_trim(td, io_u);
- else
- do_io_u_trim(td, io_u);
+ do_io_u_trim(td, io_u);
io_u_mark_submit(td, 1);
io_u_mark_complete(td, 1);
@@ -734,6 +713,7 @@ static void fio_ioring_cleanup(struct thread_data *td)
free(ld->io_u_index);
free(ld->iovecs);
free(ld->fds);
+ free(ld->dsm);
free(ld);
}
}
@@ -1146,6 +1126,16 @@ static int fio_ioring_init(struct thread_data *td)
return 1;
}
+ /*
+ * For io_uring_cmd, trims are async operations unless we are operating
+ * in zbd mode where trim means zone reset.
+ */
+ if (!strcmp(td->io_ops->name, "io_uring_cmd") && td_trim(td) &&
+ td->o.zone_mode == ZONE_MODE_ZBD)
+ td->io_ops->flags |= FIO_ASYNCIO_SYNC_TRIM;
+ else
+ ld->dsm = calloc(ld->iodepth, sizeof(*ld->dsm));
+
return 0;
}
@@ -1361,8 +1351,7 @@ static struct ioengine_ops ioengine_uring = {
static struct ioengine_ops ioengine_uring_cmd = {
.name = "io_uring_cmd",
.version = FIO_IOOPS_VERSION,
- .flags = FIO_ASYNCIO_SYNC_TRIM | FIO_NO_OFFLOAD |
- FIO_MEMALIGN | FIO_RAWIO |
+ .flags = FIO_NO_OFFLOAD | FIO_MEMALIGN | FIO_RAWIO |
FIO_ASYNCIO_SETS_ISSUE_TIME,
.init = fio_ioring_init,
.post_init = fio_ioring_cmd_post_init,
diff --git a/engines/nvme.c b/engines/nvme.c
index 1047ade2..b18ad4c2 100644
--- a/engines/nvme.c
+++ b/engines/nvme.c
@@ -5,8 +5,41 @@
#include "nvme.h"
+static inline __u64 get_slba(struct nvme_data *data, struct io_u *io_u)
+{
+ if (data->lba_ext)
+ return io_u->offset / data->lba_ext;
+ else
+ return io_u->offset >> data->lba_shift;
+}
+
+static inline __u32 get_nlb(struct nvme_data *data, struct io_u *io_u)
+{
+ if (data->lba_ext)
+ return io_u->xfer_buflen / data->lba_ext - 1;
+ else
+ return (io_u->xfer_buflen >> data->lba_shift) - 1;
+}
+
+void fio_nvme_uring_cmd_trim_prep(struct nvme_uring_cmd *cmd, struct io_u *io_u,
+ struct nvme_dsm_range *dsm)
+{
+ struct nvme_data *data = FILE_ENG_DATA(io_u->file);
+
+ cmd->opcode = nvme_cmd_dsm;
+ cmd->nsid = data->nsid;
+ cmd->cdw10 = 0;
+ cmd->cdw11 = NVME_ATTRIBUTE_DEALLOCATE;
+ cmd->addr = (__u64) (uintptr_t) dsm;
+ cmd->data_len = sizeof(*dsm);
+
+ dsm->slba = get_slba(data, io_u);
+ /* nlb is a 1-based value for deallocate */
+ dsm->nlb = get_nlb(data, io_u) + 1;
+}
+
int fio_nvme_uring_cmd_prep(struct nvme_uring_cmd *cmd, struct io_u *io_u,
- struct iovec *iov)
+ struct iovec *iov, struct nvme_dsm_range *dsm)
{
struct nvme_data *data = FILE_ENG_DATA(io_u->file);
__u64 slba;
@@ -14,21 +47,23 @@ int fio_nvme_uring_cmd_prep(struct nvme_uring_cmd *cmd, struct io_u *io_u,
memset(cmd, 0, sizeof(struct nvme_uring_cmd));
- if (io_u->ddir == DDIR_READ)
+ switch (io_u->ddir) {
+ case DDIR_READ:
cmd->opcode = nvme_cmd_read;
- else if (io_u->ddir == DDIR_WRITE)
+ break;
+ case DDIR_WRITE:
cmd->opcode = nvme_cmd_write;
- else
+ break;
+ case DDIR_TRIM:
+ fio_nvme_uring_cmd_trim_prep(cmd, io_u, dsm);
+ return 0;
+ default:
return -ENOTSUP;
-
- if (data->lba_ext) {
- slba = io_u->offset / data->lba_ext;
- nlb = (io_u->xfer_buflen / data->lba_ext) - 1;
- } else {
- slba = io_u->offset >> data->lba_shift;
- nlb = (io_u->xfer_buflen >> data->lba_shift) - 1;
}
+ slba = get_slba(data, io_u);
+ nlb = get_nlb(data, io_u);
+
/* cdw10 and cdw11 represent starting lba */
cmd->cdw10 = slba & 0xffffffff;
cmd->cdw11 = slba >> 32;
@@ -48,45 +83,6 @@ int fio_nvme_uring_cmd_prep(struct nvme_uring_cmd *cmd, struct io_u *io_u,
return 0;
}
-static int nvme_trim(int fd, __u32 nsid, __u32 nr_range, __u32 data_len,
- void *data)
-{
- struct nvme_passthru_cmd cmd = {
- .opcode = nvme_cmd_dsm,
- .nsid = nsid,
- .addr = (__u64)(uintptr_t)data,
- .data_len = data_len,
- .cdw10 = nr_range - 1,
- .cdw11 = NVME_ATTRIBUTE_DEALLOCATE,
- };
-
- return ioctl(fd, NVME_IOCTL_IO_CMD, &cmd);
-}
-
-int fio_nvme_trim(const struct thread_data *td, struct fio_file *f,
- unsigned long long offset, unsigned long long len)
-{
- struct nvme_data *data = FILE_ENG_DATA(f);
- struct nvme_dsm_range dsm;
- int ret;
-
- if (data->lba_ext) {
- dsm.nlb = len / data->lba_ext;
- dsm.slba = offset / data->lba_ext;
- } else {
- dsm.nlb = len >> data->lba_shift;
- dsm.slba = offset >> data->lba_shift;
- }
-
- ret = nvme_trim(f->fd, data->nsid, 1, sizeof(struct nvme_dsm_range),
- &dsm);
- if (ret)
- log_err("%s: nvme_trim failed for offset %llu and len %llu, err=%d\n",
- f->file_name, offset, len, ret);
-
- return ret;
-}
-
static int nvme_identify(int fd, __u32 nsid, enum nvme_identify_cns cns,
enum nvme_csi csi, void *data)
{
diff --git a/engines/nvme.h b/engines/nvme.h
index f7cb820d..238471dd 100644
--- a/engines/nvme.h
+++ b/engines/nvme.h
@@ -216,9 +216,6 @@ struct nvme_dsm_range {
__le64 slba;
};
-int fio_nvme_trim(const struct thread_data *td, struct fio_file *f,
- unsigned long long offset, unsigned long long len);
-
int fio_nvme_iomgmt_ruhs(struct thread_data *td, struct fio_file *f,
struct nvme_fdp_ruh_status *ruhs, __u32 bytes);
@@ -226,7 +223,7 @@ int fio_nvme_get_info(struct fio_file *f, __u32 *nsid, __u32 *lba_sz,
__u32 *ms, __u64 *nlba);
int fio_nvme_uring_cmd_prep(struct nvme_uring_cmd *cmd, struct io_u *io_u,
- struct iovec *iov);
+ struct iovec *iov, struct nvme_dsm_range *dsm);
int fio_nvme_get_zoned_model(struct thread_data *td, struct fio_file *f,
enum zbd_zoned_model *model);
diff --git a/oslib/linux-blkzoned.c b/oslib/linux-blkzoned.c
index c3130d0e..722e0992 100644
--- a/oslib/linux-blkzoned.c
+++ b/oslib/linux-blkzoned.c
@@ -22,6 +22,9 @@
#include "zbd_types.h"
#include <linux/blkzoned.h>
+#ifndef BLKFINISHZONE
+#define BLKFINISHZONE _IOW(0x12, 136, struct blk_zone_range)
+#endif
/*
* If the uapi headers installed on the system lacks zone capacity support,
@@ -312,7 +315,6 @@ int blkzoned_reset_wp(struct thread_data *td, struct fio_file *f,
int blkzoned_finish_zone(struct thread_data *td, struct fio_file *f,
uint64_t offset, uint64_t length)
{
-#ifdef BLKFINISHZONE
struct blk_zone_range zr = {
.sector = offset >> 9,
.nr_sectors = length >> 9,
@@ -327,21 +329,19 @@ int blkzoned_finish_zone(struct thread_data *td, struct fio_file *f,
return -errno;
}
- if (ioctl(fd, BLKFINISHZONE, &zr) < 0)
+ if (ioctl(fd, BLKFINISHZONE, &zr) < 0) {
ret = -errno;
+ /*
+ * Kernel versions older than 5.5 do not support BLKFINISHZONE
+ * and return the ENOTTY error code. These old kernels only
+ * support block devices that close zones automatically.
+ */
+ if (ret == ENOTTY)
+ ret = 0;
+ }
if (f->fd < 0)
close(fd);
return ret;
-#else
- /*
- * Kernel versions older than 5.5 does not support BLKFINISHZONE. These
- * old kernels assumed zones are closed automatically at max_open_zones
- * limit. Also they did not support max_active_zones limit. Then there
- * was no need to finish zones to avoid errors caused by max_open_zones
- * or max_active_zones. For those old versions, just do nothing.
- */
- return 0;
-#endif
}
diff --git a/t/nvmept.py b/t/nvmept.py
index e235d160..cc26d152 100755
--- a/t/nvmept.py
+++ b/t/nvmept.py
@@ -80,6 +80,10 @@ class PassThruTest(FioJobCmdTest):
print(f"Unhandled rw value {self.fio_opts['rw']}")
self.passed = False
+ if job['iodepth_level']['8'] < 95:
+ print("Did not achieve requested iodepth")
+ self.passed = False
+
TEST_LIST = [
{
@@ -232,6 +236,23 @@ TEST_LIST = [
},
"test_class": PassThruTest,
},
+ {
+ # We can't enable fixedbufs because for trim-only
+ # workloads fio actually does not allocate any buffers
+ "test_id": 15,
+ "fio_opts": {
+ "rw": 'randtrim',
+ "timebased": 1,
+ "runtime": 3,
+ "fixedbufs": 0,
+ "nonvectored": 1,
+ "force_async": 1,
+ "registerfiles": 1,
+ "sqthread_poll": 1,
+ "output-format": "json",
+ },
+ "test_class": PassThruTest,
+ },
]
def parse_args():
