* [PATCH v2 1/5] os-linux: Cleanup IO priority class and value macros
2023-07-20 1:31 [PATCH v2 0/5] Add support for I/O priority hints Damien Le Moal
@ 2023-07-20 1:31 ` Damien Le Moal
2023-07-20 9:18 ` Niklas Cassel
2023-07-20 1:31 ` [PATCH v2 2/5] os-linux: add initial support for IO priority hints Damien Le Moal
` (3 subsequent siblings)
4 siblings, 1 reply; 10+ messages in thread
From: Damien Le Moal @ 2023-07-20 1:31 UTC (permalink / raw)
To: fio, Vincent Fu, Jens Axboe; +Cc: Niklas Cassel
In os/os-linux.h, define the ioprio() macro using the already defined
IOPRIO_MAX_PRIO macro instead of hard coding the maximum priority value
again. Also move the definitions of the ioprio_class() and ioprio()
macros before the ioprio_value() function and use ioprio_class() inside
ioprio_value_is_class_rt() instead of re-coding the iopriority class
extraction again in that function.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
---
os/os-linux.h | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/os/os-linux.h b/os/os-linux.h
index 2f9f7e79..72727ac3 100644
--- a/os/os-linux.h
+++ b/os/os-linux.h
@@ -131,6 +131,9 @@ enum {
#define IOPRIO_MIN_PRIO_CLASS 0
#define IOPRIO_MAX_PRIO_CLASS 3
+#define ioprio_class(ioprio) ((ioprio) >> IOPRIO_CLASS_SHIFT)
+#define ioprio(ioprio) ((ioprio) & IOPRIO_MAX_PRIO)
+
static inline int ioprio_value(int ioprio_class, int ioprio)
{
/*
@@ -144,7 +147,7 @@ static inline int ioprio_value(int ioprio_class, int ioprio)
static inline bool ioprio_value_is_class_rt(unsigned int priority)
{
- return (priority >> IOPRIO_CLASS_SHIFT) == IOPRIO_CLASS_RT;
+ return ioprio_class(priority) == IOPRIO_CLASS_RT;
}
static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
@@ -153,9 +156,6 @@ static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
ioprio_value(ioprio_class, ioprio));
}
-#define ioprio_class(ioprio) ((ioprio) >> IOPRIO_CLASS_SHIFT)
-#define ioprio(ioprio) ((ioprio) & 7)
-
#ifndef CONFIG_HAVE_GETTID
static inline int gettid(void)
{
--
2.41.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* Re: [PATCH v2 1/5] os-linux: Cleanup IO priority class and value macros
2023-07-20 1:31 ` [PATCH v2 1/5] os-linux: Cleanup IO priority class and value macros Damien Le Moal
@ 2023-07-20 9:18 ` Niklas Cassel
0 siblings, 0 replies; 10+ messages in thread
From: Niklas Cassel @ 2023-07-20 9:18 UTC (permalink / raw)
To: Damien Le Moal; +Cc: fio@vger.kernel.org, Vincent Fu, Jens Axboe
On Thu, Jul 20, 2023 at 10:31:10AM +0900, Damien Le Moal wrote:
> In os/os-linux.h, define the ioprio() macro using the already defined
> IOPRIO_MAX_PRIO macro instead of hard coding the maximum priority value
> again. Also move the definitions of the ioprio_class() and ioprio()
> macros before the ioprio_value() function and use ioprio_class() inside
> ioprio_value_is_class_rt() instead of re-coding the iopriority class
> extraction again in that function.
>
> Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
> ---
> os/os-linux.h | 8 ++++----
> 1 file changed, 4 insertions(+), 4 deletions(-)
>
> diff --git a/os/os-linux.h b/os/os-linux.h
> index 2f9f7e79..72727ac3 100644
> --- a/os/os-linux.h
> +++ b/os/os-linux.h
> @@ -131,6 +131,9 @@ enum {
> #define IOPRIO_MIN_PRIO_CLASS 0
> #define IOPRIO_MAX_PRIO_CLASS 3
>
> +#define ioprio_class(ioprio) ((ioprio) >> IOPRIO_CLASS_SHIFT)
> +#define ioprio(ioprio) ((ioprio) & IOPRIO_MAX_PRIO)
> +
> static inline int ioprio_value(int ioprio_class, int ioprio)
> {
> /*
> @@ -144,7 +147,7 @@ static inline int ioprio_value(int ioprio_class, int ioprio)
>
> static inline bool ioprio_value_is_class_rt(unsigned int priority)
> {
> - return (priority >> IOPRIO_CLASS_SHIFT) == IOPRIO_CLASS_RT;
> + return ioprio_class(priority) == IOPRIO_CLASS_RT;
> }
>
> static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
> @@ -153,9 +156,6 @@ static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
> ioprio_value(ioprio_class, ioprio));
> }
>
> -#define ioprio_class(ioprio) ((ioprio) >> IOPRIO_CLASS_SHIFT)
> -#define ioprio(ioprio) ((ioprio) & 7)
> -
> #ifndef CONFIG_HAVE_GETTID
> static inline int gettid(void)
> {
> --
> 2.41.0
>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH v2 2/5] os-linux: add initial support for IO priority hints
2023-07-20 1:31 [PATCH v2 0/5] Add support for I/O priority hints Damien Le Moal
2023-07-20 1:31 ` [PATCH v2 1/5] os-linux: Cleanup IO priority class and value macros Damien Le Moal
@ 2023-07-20 1:31 ` Damien Le Moal
2023-07-20 1:31 ` [PATCH v2 3/5] options: add priohint option Damien Le Moal
` (2 subsequent siblings)
4 siblings, 0 replies; 10+ messages in thread
From: Damien Le Moal @ 2023-07-20 1:31 UTC (permalink / raw)
To: fio, Vincent Fu, Jens Axboe; +Cc: Niklas Cassel
Add initial support for Linux to allow specifying a hint for any
priority value. With this change, a priority value becomes the
combination of a priority class, a priority level and a hint.
The generic os.h ioprio manipulation macros, as well as the
os-dragonfly.h ioprio manipulation macros are modified to ignore this
hint.
For all other OSes that do not support priority classes, priotity hints
are ignored and always equal to 0.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
---
backend.c | 5 +++--
engines/cmdprio.c | 4 ++--
options.c | 2 +-
os/os-dragonfly.h | 4 ++--
os/os-linux.h | 19 +++++++++++++++----
os/os.h | 7 +++++--
6 files changed, 28 insertions(+), 13 deletions(-)
diff --git a/backend.c b/backend.c
index b06a11a5..268b7825 100644
--- a/backend.c
+++ b/backend.c
@@ -1800,12 +1800,13 @@ static void *thread_main(void *data)
/* ioprio_set() has to be done before td_io_init() */
if (fio_option_is_set(o, ioprio) ||
fio_option_is_set(o, ioprio_class)) {
- ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class, o->ioprio);
+ ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class,
+ o->ioprio, 0);
if (ret == -1) {
td_verror(td, errno, "ioprio_set");
goto err;
}
- td->ioprio = ioprio_value(o->ioprio_class, o->ioprio);
+ td->ioprio = ioprio_value(o->ioprio_class, o->ioprio, 0);
td->ts.ioprio = td->ioprio;
}
diff --git a/engines/cmdprio.c b/engines/cmdprio.c
index 979a81b6..e6ff1fc2 100644
--- a/engines/cmdprio.c
+++ b/engines/cmdprio.c
@@ -342,7 +342,7 @@ static int fio_cmdprio_gen_perc(struct thread_data *td, struct cmdprio *cmdprio)
prio = &cmdprio->perc_entry[ddir];
prio->perc = options->percentage[ddir];
prio->prio = ioprio_value(options->class[ddir],
- options->level[ddir]);
+ options->level[ddir], 0);
assign_clat_prio_index(prio, &values[ddir]);
ret = init_ts_clat_prio(ts, ddir, &values[ddir]);
@@ -400,7 +400,7 @@ static int fio_cmdprio_parse_and_gen_bssplit(struct thread_data *td,
goto err;
implicit_cmdprio = ioprio_value(options->class[ddir],
- options->level[ddir]);
+ options->level[ddir], 0);
ret = fio_cmdprio_generate_bsprio_desc(&cmdprio->bsprio_desc[ddir],
&parse_res[ddir],
diff --git a/options.c b/options.c
index 0f739317..143d3583 100644
--- a/options.c
+++ b/options.c
@@ -344,7 +344,7 @@ static int parse_cmdprio_bssplit_entry(struct thread_options *o,
case 4: /* bs/perc/class/level case */
class = min(class, (unsigned int) IOPRIO_MAX_PRIO_CLASS);
level = min(level, (unsigned int) IOPRIO_MAX_PRIO);
- entry->prio = ioprio_value(class, level);
+ entry->prio = ioprio_value(class, level, 0);
break;
default:
log_err("fio: invalid cmdprio_bssplit format\n");
diff --git a/os/os-dragonfly.h b/os/os-dragonfly.h
index bde39101..4ce72539 100644
--- a/os/os-dragonfly.h
+++ b/os/os-dragonfly.h
@@ -171,8 +171,8 @@ static inline int fio_getaffinity(int pid, os_cpu_mask_t *mask)
* ioprio_set() with 4 arguments, so define fio's ioprio_set() as a macro.
* Note that there is no idea of class within ioprio_set(2) unlike Linux.
*/
-#define ioprio_value(ioprio_class, ioprio) (ioprio)
-#define ioprio_set(which, who, ioprio_class, ioprio) \
+#define ioprio_value(ioprio_class, ioprio, ioprio_hint) (ioprio)
+#define ioprio_set(which, who, ioprio_class, ioprio, ioprio_hint) \
ioprio_set(which, who, ioprio)
#define ioprio(ioprio) (ioprio)
diff --git a/os/os-linux.h b/os/os-linux.h
index 72727ac3..c5cd6515 100644
--- a/os/os-linux.h
+++ b/os/os-linux.h
@@ -125,16 +125,24 @@ enum {
#define IOPRIO_BITS 16
#define IOPRIO_CLASS_SHIFT 13
+#define IOPRIO_HINT_BITS 10
+#define IOPRIO_HINT_SHIFT 3
+
#define IOPRIO_MIN_PRIO 0 /* highest priority */
#define IOPRIO_MAX_PRIO 7 /* lowest priority */
#define IOPRIO_MIN_PRIO_CLASS 0
#define IOPRIO_MAX_PRIO_CLASS 3
+#define IOPRIO_MIN_PRIO_HINT 0
+#define IOPRIO_MAX_PRIO_HINT ((1 << IOPRIO_HINT_BITS) - 1)
+
#define ioprio_class(ioprio) ((ioprio) >> IOPRIO_CLASS_SHIFT)
#define ioprio(ioprio) ((ioprio) & IOPRIO_MAX_PRIO)
+#define ioprio_hint(ioprio) \
+ (((ioprio) >> IOPRIO_HINT_SHIFT) & IOPRIO_MAX_PRIO_HINT)
-static inline int ioprio_value(int ioprio_class, int ioprio)
+static inline int ioprio_value(int ioprio_class, int ioprio, int ioprio_hint)
{
/*
* If no class is set, assume BE
@@ -142,7 +150,9 @@ static inline int ioprio_value(int ioprio_class, int ioprio)
if (!ioprio_class)
ioprio_class = IOPRIO_CLASS_BE;
- return (ioprio_class << IOPRIO_CLASS_SHIFT) | ioprio;
+ return (ioprio_class << IOPRIO_CLASS_SHIFT) |
+ (ioprio_hint << IOPRIO_HINT_SHIFT) |
+ ioprio;
}
static inline bool ioprio_value_is_class_rt(unsigned int priority)
@@ -150,10 +160,11 @@ static inline bool ioprio_value_is_class_rt(unsigned int priority)
return ioprio_class(priority) == IOPRIO_CLASS_RT;
}
-static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio)
+static inline int ioprio_set(int which, int who, int ioprio_class, int ioprio,
+ int ioprio_hint)
{
return syscall(__NR_ioprio_set, which, who,
- ioprio_value(ioprio_class, ioprio));
+ ioprio_value(ioprio_class, ioprio, ioprio_hint));
}
#ifndef CONFIG_HAVE_GETTID
diff --git a/os/os.h b/os/os.h
index 036fc233..0f182324 100644
--- a/os/os.h
+++ b/os/os.h
@@ -120,11 +120,14 @@ extern int fio_cpus_split(os_cpu_mask_t *mask, unsigned int cpu);
#define ioprio_value_is_class_rt(prio) (false)
#define IOPRIO_MIN_PRIO_CLASS 0
#define IOPRIO_MAX_PRIO_CLASS 0
+#define ioprio_hint(prio) 0
+#define IOPRIO_MIN_PRIO_HINT 0
+#define IOPRIO_MAX_PRIO_HINT 0
#endif
#ifndef FIO_HAVE_IOPRIO
-#define ioprio_value(prioclass, prio) (0)
+#define ioprio_value(prioclass, prio, priohint) (0)
#define ioprio(ioprio) 0
-#define ioprio_set(which, who, prioclass, prio) (0)
+#define ioprio_set(which, who, prioclass, prio, priohint) (0)
#define IOPRIO_MIN_PRIO 0
#define IOPRIO_MAX_PRIO 0
#endif
--
2.41.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* [PATCH v2 3/5] options: add priohint option
2023-07-20 1:31 [PATCH v2 0/5] Add support for I/O priority hints Damien Le Moal
2023-07-20 1:31 ` [PATCH v2 1/5] os-linux: Cleanup IO priority class and value macros Damien Le Moal
2023-07-20 1:31 ` [PATCH v2 2/5] os-linux: add initial support for IO priority hints Damien Le Moal
@ 2023-07-20 1:31 ` Damien Le Moal
2023-07-20 19:33 ` Jens Axboe
2023-07-20 1:31 ` [PATCH v2 4/5] cmdprio: Add support for per I/O priority hint Damien Le Moal
2023-07-20 1:31 ` [PATCH v2 5/5] stats: Add hint information to per priority level stats Damien Le Moal
4 siblings, 1 reply; 10+ messages in thread
From: Damien Le Moal @ 2023-07-20 1:31 UTC (permalink / raw)
To: fio, Vincent Fu, Jens Axboe; +Cc: Niklas Cassel
Introduce the new option priohint to allow a user to specify an I/O
priority hint applying to all IOs issued by a job. This increases fio
server version (FIO_SERVER_VER) to 101.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
---
HOWTO.rst | 6 ++++++
backend.c | 8 +++++---
cconv.c | 2 ++
fio.1 | 5 +++++
options.c | 18 ++++++++++++++++++
server.h | 2 +-
thread_options.h | 3 ++-
7 files changed, 39 insertions(+), 5 deletions(-)
diff --git a/HOWTO.rst b/HOWTO.rst
index 7ae8ea7b..9b9472a8 100644
--- a/HOWTO.rst
+++ b/HOWTO.rst
@@ -3424,6 +3424,12 @@ Threads, processes and job synchronization
priority setting, see I/O engine specific :option:`cmdprio_percentage`
and :option:`cmdprio_class` options.
+.. option:: priohint=int
+
+ Set the I/O priority hint. This has effects only for devices with
+ features controlled through priority hints (e.g. block devices
+ supporting command duration limits).
+
.. option:: cpus_allowed=str
Controls the same options as :option:`cpumask`, but accepts a textual
diff --git a/backend.c b/backend.c
index 268b7825..624d7c6f 100644
--- a/backend.c
+++ b/backend.c
@@ -1799,14 +1799,16 @@ static void *thread_main(void *data)
/* ioprio_set() has to be done before td_io_init() */
if (fio_option_is_set(o, ioprio) ||
- fio_option_is_set(o, ioprio_class)) {
+ fio_option_is_set(o, ioprio_class) ||
+ fio_option_is_set(o, ioprio_hint)) {
ret = ioprio_set(IOPRIO_WHO_PROCESS, 0, o->ioprio_class,
- o->ioprio, 0);
+ o->ioprio, o->ioprio_hint);
if (ret == -1) {
td_verror(td, errno, "ioprio_set");
goto err;
}
- td->ioprio = ioprio_value(o->ioprio_class, o->ioprio, 0);
+ td->ioprio = ioprio_value(o->ioprio_class, o->ioprio,
+ o->ioprio_hint);
td->ts.ioprio = td->ioprio;
}
diff --git a/cconv.c b/cconv.c
index 1bfa770f..ce6acbe6 100644
--- a/cconv.c
+++ b/cconv.c
@@ -281,6 +281,7 @@ int convert_thread_options_to_cpu(struct thread_options *o,
o->nice = le32_to_cpu(top->nice);
o->ioprio = le32_to_cpu(top->ioprio);
o->ioprio_class = le32_to_cpu(top->ioprio_class);
+ o->ioprio_hint = le32_to_cpu(top->ioprio_hint);
o->file_service_type = le32_to_cpu(top->file_service_type);
o->group_reporting = le32_to_cpu(top->group_reporting);
o->stats = le32_to_cpu(top->stats);
@@ -496,6 +497,7 @@ void convert_thread_options_to_net(struct thread_options_pack *top,
top->nice = cpu_to_le32(o->nice);
top->ioprio = cpu_to_le32(o->ioprio);
top->ioprio_class = cpu_to_le32(o->ioprio_class);
+ top->ioprio_hint = cpu_to_le32(o->ioprio_hint);
top->file_service_type = cpu_to_le32(o->file_service_type);
top->group_reporting = cpu_to_le32(o->group_reporting);
top->stats = cpu_to_le32(o->stats);
diff --git a/fio.1 b/fio.1
index da875276..0f7ebc7f 100644
--- a/fio.1
+++ b/fio.1
@@ -3132,6 +3132,11 @@ Set the I/O priority class. See man \fBionice\fR\|(1). For per-command
priority setting, see the I/O engine specific `cmdprio_percentage` and
`cmdprio_class` options.
.TP
+.BI priohint \fR=\fPint
+Set the I/O priority hint. This has effects only for devices with
+features controlled through priority hints (e.g. block devices supporting
+command duration limits).
+.TP
.BI cpus_allowed \fR=\fPstr
Controls the same options as \fBcpumask\fR, but accepts a textual
specification of the permitted CPUs instead and CPUs are indexed from 0. So
diff --git a/options.c b/options.c
index 143d3583..56672960 100644
--- a/options.c
+++ b/options.c
@@ -3806,6 +3806,18 @@ struct fio_option fio_options[FIO_MAX_OPTS] = {
.category = FIO_OPT_C_GENERAL,
.group = FIO_OPT_G_CRED,
},
+ {
+ .name = "priohint",
+ .lname = "I/O nice priority hint",
+ .type = FIO_OPT_INT,
+ .off1 = offsetof(struct thread_options, ioprio_hint),
+ .help = "Set job IO priority hint",
+ .minval = IOPRIO_MIN_PRIO_HINT,
+ .maxval = IOPRIO_MAX_PRIO_HINT,
+ .interval = 1,
+ .category = FIO_OPT_C_GENERAL,
+ .group = FIO_OPT_G_CRED,
+ },
#else
{
.name = "prioclass",
@@ -3813,6 +3825,12 @@ struct fio_option fio_options[FIO_MAX_OPTS] = {
.type = FIO_OPT_UNSUPPORTED,
.help = "Your platform does not support IO priority classes",
},
+ {
+ .name = "priohint",
+ .lname = "I/O nice priority hint",
+ .type = FIO_OPT_UNSUPPORTED,
+ .help = "Your platform does not support IO priority hints",
+ },
#endif
{
.name = "thinktime",
diff --git a/server.h b/server.h
index 601d3340..ad706118 100644
--- a/server.h
+++ b/server.h
@@ -51,7 +51,7 @@ struct fio_net_cmd_reply {
};
enum {
- FIO_SERVER_VER = 100,
+ FIO_SERVER_VER = 101,
FIO_SERVER_MAX_FRAGMENT_PDU = 1024,
FIO_SERVER_MAX_CMD_MB = 2048,
diff --git a/thread_options.h b/thread_options.h
index 1715b36c..38a9993d 100644
--- a/thread_options.h
+++ b/thread_options.h
@@ -248,6 +248,7 @@ struct thread_options {
unsigned int nice;
unsigned int ioprio;
unsigned int ioprio_class;
+ unsigned int ioprio_hint;
unsigned int file_service_type;
unsigned int group_reporting;
unsigned int stats;
@@ -568,6 +569,7 @@ struct thread_options_pack {
uint32_t nice;
uint32_t ioprio;
uint32_t ioprio_class;
+ uint32_t ioprio_hint;
uint32_t file_service_type;
uint32_t group_reporting;
uint32_t stats;
@@ -601,7 +603,6 @@ struct thread_options_pack {
uint32_t lat_percentiles;
uint32_t slat_percentiles;
uint32_t percentile_precision;
- uint32_t pad5;
fio_fp64_t percentile_list[FIO_IO_U_LIST_MAX_LEN];
uint8_t read_iolog_file[FIO_TOP_STR_MAX];
--
2.41.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* Re: [PATCH v2 3/5] options: add priohint option
2023-07-20 1:31 ` [PATCH v2 3/5] options: add priohint option Damien Le Moal
@ 2023-07-20 19:33 ` Jens Axboe
0 siblings, 0 replies; 10+ messages in thread
From: Jens Axboe @ 2023-07-20 19:33 UTC (permalink / raw)
To: Damien Le Moal, fio, Vincent Fu; +Cc: Niklas Cassel
On 7/19/23 7:31?PM, Damien Le Moal wrote:
> diff --git a/HOWTO.rst b/HOWTO.rst
> index 7ae8ea7b..9b9472a8 100644
> --- a/HOWTO.rst
> +++ b/HOWTO.rst
> @@ -3424,6 +3424,12 @@ Threads, processes and job synchronization
> priority setting, see I/O engine specific :option:`cmdprio_percentage`
> and :option:`cmdprio_class` options.
>
> +.. option:: priohint=int
> +
> + Set the I/O priority hint. This has effects only for devices with
> + features controlled through priority hints (e.g. block devices
> + supporting command duration limits).
> +
I would rephrase that as:
Set the I/O priority hint. This is only applicable to devices with
features controlled through priority hints (e.g. block devices
supporting command duration limits, or CDL). CDL is a way to XXX
and finish the XXX, and apply to the man page as well.
Assume people don't know what CDL is, because that will generally be
true!
--
Jens Axboe
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH v2 4/5] cmdprio: Add support for per I/O priority hint
2023-07-20 1:31 [PATCH v2 0/5] Add support for I/O priority hints Damien Le Moal
` (2 preceding siblings ...)
2023-07-20 1:31 ` [PATCH v2 3/5] options: add priohint option Damien Le Moal
@ 2023-07-20 1:31 ` Damien Le Moal
2023-07-20 19:36 ` Jens Axboe
2023-07-20 1:31 ` [PATCH v2 5/5] stats: Add hint information to per priority level stats Damien Le Moal
4 siblings, 1 reply; 10+ messages in thread
From: Damien Le Moal @ 2023-07-20 1:31 UTC (permalink / raw)
To: fio, Vincent Fu, Jens Axboe; +Cc: Niklas Cassel
Introduce the new option cmdprio_hint to allow specifying I/O priority
hints per IO with the iouring and libaio IO engines. A third acceptable
format for the cmdprio_bssplit option is also introduced to allow
specifying an IO hint in addition to a priority class and level.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
---
HOWTO.rst | 28 +++++++++++++++++++++++-----
engines/cmdprio.c | 9 ++++++---
engines/cmdprio.h | 1 +
engines/io_uring.c | 25 +++++++++++++++++++++++--
engines/libaio.c | 21 +++++++++++++++++++++
fio.1 | 27 +++++++++++++++++++++++----
options.c | 13 ++++++++++---
7 files changed, 107 insertions(+), 17 deletions(-)
diff --git a/HOWTO.rst b/HOWTO.rst
index 9b9472a8..a3b5bf2e 100644
--- a/HOWTO.rst
+++ b/HOWTO.rst
@@ -2275,6 +2275,16 @@ with the caveat that when used on the command line, they must come after the
reads and writes. See :manpage:`ionice(1)`. See also the
:option:`prioclass` option.
+.. option:: cmdprio_hint=int[,int] : [io_uring] [libaio]
+
+ Set the I/O priority hint to use for I/Os that must be issued with
+ a priority when :option:`cmdprio_percentage` or
+ :option:`cmdprio_bssplit` is set. If not specified when
+ :option:`cmdprio_percentage` or :option:`cmdprio_bssplit` is set,
+ this defaults to 0 (no hint). A single value applies to reads and
+ writes. Comma-separated values may be specified for reads and writes.
+ See also the :option:`priohint` option.
+
.. option:: cmdprio=int[,int] : [io_uring] [libaio]
Set the I/O priority value to use for I/Os that must be issued with
@@ -2301,9 +2311,9 @@ with the caveat that when used on the command line, they must come after the
cmdprio_bssplit=blocksize/percentage:blocksize/percentage
- In this case, each entry will use the priority class and priority
- level defined by the options :option:`cmdprio_class` and
- :option:`cmdprio` respectively.
+ In this case, each entry will use the priority class, priority hint
+ and priority level defined by the options :option:`cmdprio_class`,
+ :option:`cmdprio` and :option:`cmdprio_hint` respectively.
The second accepted format for this option is:
@@ -2314,7 +2324,14 @@ with the caveat that when used on the command line, they must come after the
accepted format does not restrict all entries to have the same priority
class and priority level.
- For both formats, only the read and write data directions are supported,
+ The third accepted format for this option is:
+
+ cmdprio_bssplit=blocksize/percentage/class/level/hint:...
+
+ This is an extension of the second accepted format that allows to also
+ specify a priority hint.
+
+ For all formats, only the read and write data directions are supported,
values for trim IOs are ignored. This option is mutually exclusive with
the :option:`cmdprio_percentage` option.
@@ -3428,7 +3445,8 @@ Threads, processes and job synchronization
Set the I/O priority hint. This has effects only for devices with
features controlled through priority hints (e.g. block devices
- supporting command duration limits).
+ supporting command duration limits). For per-I/O priority hint
+ setting, see the I/O engine specific :option:`cmdprio_hint` option.
.. option:: cpus_allowed=str
diff --git a/engines/cmdprio.c b/engines/cmdprio.c
index e6ff1fc2..153e3691 100644
--- a/engines/cmdprio.c
+++ b/engines/cmdprio.c
@@ -267,7 +267,8 @@ static int fio_cmdprio_percentage(struct cmdprio *cmdprio, struct io_u *io_u,
* to be set. If the random percentage value is within the user specified
* percentage of I/Os that should use a cmdprio priority value (rather than
* the default priority), then this function updates the io_u with an ioprio
- * value as defined by the cmdprio/cmdprio_class or cmdprio_bssplit options.
+ * value as defined by the cmdprio/cmdprio_hint/cmdprio_class or
+ * cmdprio_bssplit options.
*
* Return true if the io_u ioprio was changed and false otherwise.
*/
@@ -342,7 +343,8 @@ static int fio_cmdprio_gen_perc(struct thread_data *td, struct cmdprio *cmdprio)
prio = &cmdprio->perc_entry[ddir];
prio->perc = options->percentage[ddir];
prio->prio = ioprio_value(options->class[ddir],
- options->level[ddir], 0);
+ options->level[ddir],
+ options->hint[ddir]);
assign_clat_prio_index(prio, &values[ddir]);
ret = init_ts_clat_prio(ts, ddir, &values[ddir]);
@@ -400,7 +402,8 @@ static int fio_cmdprio_parse_and_gen_bssplit(struct thread_data *td,
goto err;
implicit_cmdprio = ioprio_value(options->class[ddir],
- options->level[ddir], 0);
+ options->level[ddir],
+ options->hint[ddir]);
ret = fio_cmdprio_generate_bsprio_desc(&cmdprio->bsprio_desc[ddir],
&parse_res[ddir],
diff --git a/engines/cmdprio.h b/engines/cmdprio.h
index 755da8d0..9b351828 100644
--- a/engines/cmdprio.h
+++ b/engines/cmdprio.h
@@ -39,6 +39,7 @@ struct cmdprio_options {
unsigned int percentage[CMDPRIO_RWDIR_CNT];
unsigned int class[CMDPRIO_RWDIR_CNT];
unsigned int level[CMDPRIO_RWDIR_CNT];
+ unsigned int hint[CMDPRIO_RWDIR_CNT];
char *bssplit_str;
};
diff --git a/engines/io_uring.c b/engines/io_uring.c
index f30a3c00..d55af6bc 100644
--- a/engines/io_uring.c
+++ b/engines/io_uring.c
@@ -157,6 +157,21 @@ static struct fio_option options[] = {
.category = FIO_OPT_C_ENGINE,
.group = FIO_OPT_G_IOURING,
},
+ {
+ .name = "cmdprio_hint",
+ .lname = "Asynchronous I/O priority hint",
+ .type = FIO_OPT_INT,
+ .off1 = offsetof(struct ioring_options,
+ cmdprio_options.hint[DDIR_READ]),
+ .off2 = offsetof(struct ioring_options,
+ cmdprio_options.hint[DDIR_WRITE]),
+ .help = "Set asynchronous IO priority hint",
+ .minval = IOPRIO_MIN_PRIO_HINT,
+ .maxval = IOPRIO_MAX_PRIO_HINT,
+ .interval = 1,
+ .category = FIO_OPT_C_ENGINE,
+ .group = FIO_OPT_G_IOURING,
+ },
{
.name = "cmdprio",
.lname = "Asynchronous I/O priority level",
@@ -195,6 +210,12 @@ static struct fio_option options[] = {
.type = FIO_OPT_UNSUPPORTED,
.help = "Your platform does not support I/O priority classes",
},
+ {
+ .name = "cmdprio_hint",
+ .lname = "Asynchronous I/O priority hint",
+ .type = FIO_OPT_UNSUPPORTED,
+ .help = "Your platform does not support I/O priority classes",
+ },
{
.name = "cmdprio",
.lname = "Asynchronous I/O priority level",
@@ -365,8 +386,8 @@ static int fio_ioring_prep(struct thread_data *td, struct io_u *io_u)
/*
* Since io_uring can have a submission context (sqthread_poll)
* that is different from the process context, we cannot rely on
- * the IO priority set by ioprio_set() (option prio/prioclass)
- * to be inherited.
+ * the IO priority set by ioprio_set() (options prio, prioclass,
+ * and priohint) to be inherited.
* td->ioprio will have the value of the "default prio", so set
* this unconditionally. This value might get overridden by
* fio_ioring_cmdprio_prep() if the option cmdprio_percentage or
diff --git a/engines/libaio.c b/engines/libaio.c
index 6a0745aa..c52d05fe 100644
--- a/engines/libaio.c
+++ b/engines/libaio.c
@@ -102,6 +102,21 @@ static struct fio_option options[] = {
.category = FIO_OPT_C_ENGINE,
.group = FIO_OPT_G_LIBAIO,
},
+ {
+ .name = "cmdprio_hint",
+ .lname = "Asynchronous I/O priority hint",
+ .type = FIO_OPT_INT,
+ .off1 = offsetof(struct libaio_options,
+ cmdprio_options.hint[DDIR_READ]),
+ .off2 = offsetof(struct libaio_options,
+ cmdprio_options.hint[DDIR_WRITE]),
+ .help = "Set asynchronous IO priority hint",
+ .minval = IOPRIO_MIN_PRIO_HINT,
+ .maxval = IOPRIO_MAX_PRIO_HINT,
+ .interval = 1,
+ .category = FIO_OPT_C_ENGINE,
+ .group = FIO_OPT_G_LIBAIO,
+ },
{
.name = "cmdprio",
.lname = "Asynchronous I/O priority level",
@@ -140,6 +155,12 @@ static struct fio_option options[] = {
.type = FIO_OPT_UNSUPPORTED,
.help = "Your platform does not support I/O priority classes",
},
+ {
+ .name = "cmdprio_hint",
+ .lname = "Asynchronous I/O priority hint",
+ .type = FIO_OPT_UNSUPPORTED,
+ .help = "Your platform does not support I/O priority classes",
+ },
{
.name = "cmdprio",
.lname = "Asynchronous I/O priority level",
diff --git a/fio.1 b/fio.1
index 0f7ebc7f..c53cb27c 100644
--- a/fio.1
+++ b/fio.1
@@ -2072,6 +2072,14 @@ is set, this defaults to the highest priority class. A single value applies
to reads and writes. Comma-separated values may be specified for reads and
writes. See man \fBionice\fR\|(1). See also the \fBprioclass\fR option.
.TP
+.BI (io_uring,libaio)cmdprio_hint \fR=\fPint[,int]
+Set the I/O priority hint to use for I/Os that must be issued with a
+priority when \fBcmdprio_percentage\fR or \fBcmdprio_bssplit\fR is set.
+If not specified when \fBcmdprio_percentage\fR or \fBcmdprio_bssplit\fR
+is set, this defaults to 0 (no hint). A single value applies to reads and
+writes. Comma-separated values may be specified for reads and writes.
+See also the \fBpriohint\fR option.
+.TP
.BI (io_uring,libaio)cmdprio \fR=\fPint[,int]
Set the I/O priority value to use for I/Os that must be issued with a
priority when \fBcmdprio_percentage\fR or \fBcmdprio_bssplit\fR is set.
@@ -2097,8 +2105,9 @@ The first accepted format for this option is the same as the format of the
cmdprio_bssplit=blocksize/percentage:blocksize/percentage
.RE
.P
-In this case, each entry will use the priority class and priority level defined
-by the options \fBcmdprio_class\fR and \fBcmdprio\fR respectively.
+In this case, each entry will use the priority class, priority hint and
+priority level defined by the options \fBcmdprio_class\fR, \fBcmdprio\fR
+and \fBcmdprio_hint\fR respectively.
.P
The second accepted format for this option is:
.RS
@@ -2111,7 +2120,16 @@ entry. In comparison with the first accepted format, the second accepted format
does not restrict all entries to have the same priority class and priority
level.
.P
-For both formats, only the read and write data directions are supported, values
+The third accepted format for this option is:
+.RS
+.P
+cmdprio_bssplit=blocksize/percentage/class/level/hint:...
+.RE
+.P
+This is an extension of the second accepted format that allows to also
+specify a priority hint.
+.P
+For all formats, only the read and write data directions are supported, values
for trim IOs are ignored. This option is mutually exclusive with the
\fBcmdprio_percentage\fR option.
.RE
@@ -3135,7 +3153,8 @@ priority setting, see the I/O engine specific `cmdprio_percentage` and
.BI priohint \fR=\fPint
Set the I/O priority hint. This has effects only for devices with
features controlled through priority hints (e.g. block devices supporting
-command duration limits).
+command duration limits). For per-I/O priority hint setting, see the I/O
+engine specific \fBcmdprio_hint\fB option.
.TP
.BI cpus_allowed \fR=\fPstr
Controls the same options as \fBcpumask\fR, but accepts a textual
diff --git a/options.c b/options.c
index 56672960..48aa0d7b 100644
--- a/options.c
+++ b/options.c
@@ -313,15 +313,17 @@ static int parse_cmdprio_bssplit_entry(struct thread_options *o,
int matches = 0;
char *bs_str = NULL;
long long bs_val;
- unsigned int perc = 0, class, level;
+ unsigned int perc = 0, class, level, hint;
/*
* valid entry formats:
* bs/ - %s/ - set perc to 0, prio to -1.
* bs/perc - %s/%u - set prio to -1.
* bs/perc/class/level - %s/%u/%u/%u
+ * bs/perc/class/level/hint - %s/%u/%u/%u/%u
*/
- matches = sscanf(str, "%m[^/]/%u/%u/%u", &bs_str, &perc, &class, &level);
+ matches = sscanf(str, "%m[^/]/%u/%u/%u/%u",
+ &bs_str, &perc, &class, &level, &hint);
if (matches < 1) {
log_err("fio: invalid cmdprio_bssplit format\n");
return 1;
@@ -342,9 +344,14 @@ static int parse_cmdprio_bssplit_entry(struct thread_options *o,
case 2: /* bs/perc case */
break;
case 4: /* bs/perc/class/level case */
+ case 5: /* bs/perc/class/level/hint case */
class = min(class, (unsigned int) IOPRIO_MAX_PRIO_CLASS);
level = min(level, (unsigned int) IOPRIO_MAX_PRIO);
- entry->prio = ioprio_value(class, level, 0);
+ if (matches == 5)
+ hint = min(hint, (unsigned int) IOPRIO_MAX_PRIO_HINT);
+ else
+ hint = 0;
+ entry->prio = ioprio_value(class, level, hint);
break;
default:
log_err("fio: invalid cmdprio_bssplit format\n");
--
2.41.0
^ permalink raw reply related [flat|nested] 10+ messages in thread* Re: [PATCH v2 4/5] cmdprio: Add support for per I/O priority hint
2023-07-20 1:31 ` [PATCH v2 4/5] cmdprio: Add support for per I/O priority hint Damien Le Moal
@ 2023-07-20 19:36 ` Jens Axboe
2023-07-21 1:07 ` Damien Le Moal
0 siblings, 1 reply; 10+ messages in thread
From: Jens Axboe @ 2023-07-20 19:36 UTC (permalink / raw)
To: Damien Le Moal, fio, Vincent Fu; +Cc: Niklas Cassel
On 7/19/23 7:31?PM, Damien Le Moal wrote:
> diff --git a/engines/io_uring.c b/engines/io_uring.c
> index f30a3c00..d55af6bc 100644
> --- a/engines/io_uring.c
> +++ b/engines/io_uring.c
> @@ -157,6 +157,21 @@ static struct fio_option options[] = {
> .category = FIO_OPT_C_ENGINE,
> .group = FIO_OPT_G_IOURING,
> },
> + {
> + .name = "cmdprio_hint",
> + .lname = "Asynchronous I/O priority hint",
> + .type = FIO_OPT_INT,
> + .off1 = offsetof(struct ioring_options,
> + cmdprio_options.hint[DDIR_READ]),
> + .off2 = offsetof(struct ioring_options,
> + cmdprio_options.hint[DDIR_WRITE]),
> + .help = "Set asynchronous IO priority hint",
> + .minval = IOPRIO_MIN_PRIO_HINT,
> + .maxval = IOPRIO_MAX_PRIO_HINT,
> + .interval = 1,
> + .category = FIO_OPT_C_ENGINE,
> + .group = FIO_OPT_G_IOURING,
> + },
> {
> .name = "cmdprio",
> .lname = "Asynchronous I/O priority level",
> @@ -195,6 +210,12 @@ static struct fio_option options[] = {
> .type = FIO_OPT_UNSUPPORTED,
> .help = "Your platform does not support I/O priority classes",
> },
> + {
> + .name = "cmdprio_hint",
> + .lname = "Asynchronous I/O priority hint",
> + .type = FIO_OPT_UNSUPPORTED,
> + .help = "Your platform does not support I/O priority classes",
> + },
> {
> .name = "cmdprio",
> .lname = "Asynchronous I/O priority level",
Since neither this nor libaio actually do anything with these values,
why isn't this just a generic option? You could flag the two engines
where it actually works with an engine flag, and check if it's set and
we don't have that flag and error out in the generic code.
--
Jens Axboe
^ permalink raw reply [flat|nested] 10+ messages in thread* Re: [PATCH v2 4/5] cmdprio: Add support for per I/O priority hint
2023-07-20 19:36 ` Jens Axboe
@ 2023-07-21 1:07 ` Damien Le Moal
0 siblings, 0 replies; 10+ messages in thread
From: Damien Le Moal @ 2023-07-21 1:07 UTC (permalink / raw)
To: Jens Axboe, fio, Vincent Fu; +Cc: Niklas Cassel
On 7/21/23 04:36, Jens Axboe wrote:
> On 7/19/23 7:31?PM, Damien Le Moal wrote:
>> diff --git a/engines/io_uring.c b/engines/io_uring.c
>> index f30a3c00..d55af6bc 100644
>> --- a/engines/io_uring.c
>> +++ b/engines/io_uring.c
>> @@ -157,6 +157,21 @@ static struct fio_option options[] = {
>> .category = FIO_OPT_C_ENGINE,
>> .group = FIO_OPT_G_IOURING,
>> },
>> + {
>> + .name = "cmdprio_hint",
>> + .lname = "Asynchronous I/O priority hint",
>> + .type = FIO_OPT_INT,
>> + .off1 = offsetof(struct ioring_options,
>> + cmdprio_options.hint[DDIR_READ]),
>> + .off2 = offsetof(struct ioring_options,
>> + cmdprio_options.hint[DDIR_WRITE]),
>> + .help = "Set asynchronous IO priority hint",
>> + .minval = IOPRIO_MIN_PRIO_HINT,
>> + .maxval = IOPRIO_MAX_PRIO_HINT,
>> + .interval = 1,
>> + .category = FIO_OPT_C_ENGINE,
>> + .group = FIO_OPT_G_IOURING,
>> + },
>> {
>> .name = "cmdprio",
>> .lname = "Asynchronous I/O priority level",
>> @@ -195,6 +210,12 @@ static struct fio_option options[] = {
>> .type = FIO_OPT_UNSUPPORTED,
>> .help = "Your platform does not support I/O priority classes",
>> },
>> + {
>> + .name = "cmdprio_hint",
>> + .lname = "Asynchronous I/O priority hint",
>> + .type = FIO_OPT_UNSUPPORTED,
>> + .help = "Your platform does not support I/O priority classes",
>> + },
>> {
>> .name = "cmdprio",
>> .lname = "Asynchronous I/O priority level",
>
> Since neither this nor libaio actually do anything with these values,
> why isn't this just a generic option? You could flag the two engines
> where it actually works with an engine flag, and check if it's set and
> we don't have that flag and error out in the generic code.
Hmmm. I would like to keep the async IO prio option definitions and storage
together with the cmdprio_options struct, which is common to both libaio and
iouring engines. If we move the hint option definition to be a generic one, we
cannot use that struct and would need a job field. I can make that work, but
that makes the code a little messier by spreading out the controls for async IO
prio.
I agree that we can drop the error output though as indeed the hint value is
unused on systems without hint support. But warning the user about the lack of
support is nice I think.
Checking the code again though, one thing I noticed is that iouring and libaio
cmdprio_xxx options definition is identical, modulo the group. So all these
definitions could be moved to engines/cmdprio.h as macros and IO engines
supporting these options can use these to add the options. The other option is
to move the definition of all these options to be generic ones and use the
engine flag trick you mention to indicate support.
--
Damien Le Moal
Western Digital Research
^ permalink raw reply [flat|nested] 10+ messages in thread
* [PATCH v2 5/5] stats: Add hint information to per priority level stats
2023-07-20 1:31 [PATCH v2 0/5] Add support for I/O priority hints Damien Le Moal
` (3 preceding siblings ...)
2023-07-20 1:31 ` [PATCH v2 4/5] cmdprio: Add support for per I/O priority hint Damien Le Moal
@ 2023-07-20 1:31 ` Damien Le Moal
4 siblings, 0 replies; 10+ messages in thread
From: Damien Le Moal @ 2023-07-20 1:31 UTC (permalink / raw)
To: fio, Vincent Fu, Jens Axboe; +Cc: Niklas Cassel
Modify the json and standard per-priority output stats to display the
hint value together with the priority class and level.
Signed-off-by: Damien Le Moal <dlemoal@kernel.org>
Reviewed-by: Niklas Cassel <niklas.cassel@wdc.com>
---
stat.c | 10 +++++++---
1 file changed, 7 insertions(+), 3 deletions(-)
diff --git a/stat.c b/stat.c
index 7fad73d1..7b791628 100644
--- a/stat.c
+++ b/stat.c
@@ -597,10 +597,11 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
continue;
snprintf(buf, sizeof(buf),
- "%s prio %u/%u",
+ "%s prio %u/%u/%u",
clat_type,
ioprio_class(ts->clat_prio[ddir][i].ioprio),
- ioprio(ts->clat_prio[ddir][i].ioprio));
+ ioprio(ts->clat_prio[ddir][i].ioprio),
+ ioprio_hint(ts->clat_prio[ddir][i].ioprio));
display_lat(buf, min, max, mean, dev, out);
}
}
@@ -640,10 +641,11 @@ static void show_ddir_status(struct group_run_stats *rs, struct thread_stat *ts,
continue;
snprintf(prio_name, sizeof(prio_name),
- "%s prio %u/%u (%.2f%% of IOs)",
+ "%s prio %u/%u/%u (%.2f%% of IOs)",
clat_type,
ioprio_class(ts->clat_prio[ddir][i].ioprio),
ioprio(ts->clat_prio[ddir][i].ioprio),
+ ioprio_hint(ts->clat_prio[ddir][i].ioprio),
100. * (double) prio_samples / (double) samples);
show_clat_percentiles(ts->clat_prio[ddir][i].io_u_plat,
prio_samples, ts->percentile_list,
@@ -1533,6 +1535,8 @@ static void add_ddir_status_json(struct thread_stat *ts,
ioprio_class(ts->clat_prio[ddir][i].ioprio));
json_object_add_value_int(obj, "prio",
ioprio(ts->clat_prio[ddir][i].ioprio));
+ json_object_add_value_int(obj, "priohint",
+ ioprio_hint(ts->clat_prio[ddir][i].ioprio));
tmp_object = add_ddir_lat_json(ts,
ts->clat_percentiles | ts->lat_percentiles,
--
2.41.0
^ permalink raw reply related [flat|nested] 10+ messages in thread