2021-09-13 10:42:12

by Eugene Syromiatnikov

[permalink] [raw]
Subject: [PATCH v2] io-wq: expose IO_WQ_ACCT_* enumeration items to UAPI

These are used to index elements in the argument
of IORING_REGISTER_IOWQ_MAX_WORKERS io_uring_register command,
so they are to be exposed in UAPI.

Complements: 2e480058ddc21ec5 ("io-wq: provide a way to limit max number of workers")
Signed-off-by: Eugene Syromiatnikov <[email protected]>
---
v2:
- IO_WQ_ACCT_NR is no longer exposed directly in UAPI, per Jens Axboe's
suggestion.

v1: https://lore.kernel.org/lkml/[email protected]/
---
fs/io-wq.c | 5 ++---
include/uapi/linux/io_uring.h | 8 ++++++++
2 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/fs/io-wq.c b/fs/io-wq.c
index 6c55362..eb5162d 100644
--- a/fs/io-wq.c
+++ b/fs/io-wq.c
@@ -14,6 +14,7 @@
#include <linux/rculist_nulls.h>
#include <linux/cpu.h>
#include <linux/tracehook.h>
+#include <uapi/linux/io_uring.h>

#include "io-wq.h"

@@ -78,9 +79,7 @@ struct io_wqe_acct {
};

enum {
- IO_WQ_ACCT_BOUND,
- IO_WQ_ACCT_UNBOUND,
- IO_WQ_ACCT_NR,
+ IO_WQ_ACCT_NR = __IO_WQ_ACCT_MAX
};

/*
diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 59ef351..dae1841 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -324,6 +324,14 @@ enum {
IORING_REGISTER_LAST
};

+/* io-wq worker limit categories */
+enum {
+ IO_WQ_ACCT_BOUND,
+ IO_WQ_ACCT_UNBOUND,
+
+ __IO_WQ_ACCT_MAX /* Non-UAPI */
+};
+
/* deprecated, see struct io_uring_rsrc_update */
struct io_uring_files_update {
__u32 offset;
--
2.1.4


2021-09-13 15:54:28

by Eugene Syromiatnikov

[permalink] [raw]
Subject: Re: [PATCH v2] io-wq: expose IO_WQ_ACCT_* enumeration items to UAPI

On Mon, Sep 13, 2021 at 07:28:11AM -0600, Jens Axboe wrote:
> This is really the same thing as before, just the names have changed.
> What I suggested was keeping the enum in io_uring, then just adding
>
> enum {
> IO_WQ_BOUND,
> IO_WQ_UNBOUND,
> };
>
> to uapi header. The ACCT stuff is io-wq specific too, that kind of naming
> shouldn't be propagated to userspace.

My apologies, I've overlooked the fact that the proposed names
are different. Updated and resent[1].

[1] https://lore.kernel.org/lkml/[email protected]/

> A BUILD_BUG_ON() could be added for them being different, but honestly
> I don't think that's worth it.

2021-09-13 22:59:21

by Jens Axboe

[permalink] [raw]
Subject: Re: [PATCH v2] io-wq: expose IO_WQ_ACCT_* enumeration items to UAPI

On 9/13/21 4:41 AM, Eugene Syromiatnikov wrote:
> These are used to index elements in the argument
> of IORING_REGISTER_IOWQ_MAX_WORKERS io_uring_register command,
> so they are to be exposed in UAPI.
>
> Complements: 2e480058ddc21ec5 ("io-wq: provide a way to limit max number of workers")
> Signed-off-by: Eugene Syromiatnikov <[email protected]>
> ---
> v2:
> - IO_WQ_ACCT_NR is no longer exposed directly in UAPI, per Jens Axboe's
> suggestion.
>
> v1: https://lore.kernel.org/lkml/[email protected]/
> ---
> fs/io-wq.c | 5 ++---
> include/uapi/linux/io_uring.h | 8 ++++++++
> 2 files changed, 10 insertions(+), 3 deletions(-)
>
> diff --git a/fs/io-wq.c b/fs/io-wq.c
> index 6c55362..eb5162d 100644
> --- a/fs/io-wq.c
> +++ b/fs/io-wq.c
> @@ -14,6 +14,7 @@
> #include <linux/rculist_nulls.h>
> #include <linux/cpu.h>
> #include <linux/tracehook.h>
> +#include <uapi/linux/io_uring.h>
>
> #include "io-wq.h"
>
> @@ -78,9 +79,7 @@ struct io_wqe_acct {
> };
>
> enum {
> - IO_WQ_ACCT_BOUND,
> - IO_WQ_ACCT_UNBOUND,
> - IO_WQ_ACCT_NR,
> + IO_WQ_ACCT_NR = __IO_WQ_ACCT_MAX
> };
>
> /*
> diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
> index 59ef351..dae1841 100644
> --- a/include/uapi/linux/io_uring.h
> +++ b/include/uapi/linux/io_uring.h
> @@ -324,6 +324,14 @@ enum {
> IORING_REGISTER_LAST
> };
>
> +/* io-wq worker limit categories */
> +enum {
> + IO_WQ_ACCT_BOUND,
> + IO_WQ_ACCT_UNBOUND,
> +
> + __IO_WQ_ACCT_MAX /* Non-UAPI */
> +};

This is really the same thing as before, just the names have changed.
What I suggested was keeping the enum in io_uring, then just adding

enum {
IO_WQ_BOUND,
IO_WQ_UNBOUND,
};

to uapi header. The ACCT stuff is io-wq specific too, that kind of naming
shouldn't be propagated to userspace.

A BUILD_BUG_ON() could be added for them being different, but honestly
I don't think that's worth it.

--
Jens Axboe