From: Jens Axboe <[email protected]>
To: Bijan Mottahedeh <[email protected]>
Cc: [email protected]
Subject: Re: [RFC 2/2] io_uring: report pinned memory usage
Date: Sun, 14 Jun 2020 09:57:40 -0600 [thread overview]
Message-ID: <[email protected]> (raw)
In-Reply-To: <[email protected]>
On 6/12/20 10:43 PM, Bijan Mottahedeh wrote:
> On 6/12/2020 8:19 AM, Jens Axboe wrote:
>> On 6/12/20 9:16 AM, Jens Axboe wrote:
>>> On 6/11/20 8:23 PM, Bijan Mottahedeh wrote:
>>>> Long term, it makes sense to separate reporting and enforcing of pinned
>>>> memory usage.
>>>>
>>>> Signed-off-by: Bijan Mottahedeh <[email protected]>
>>>>
>>>> It is useful to view
>>>> ---
>>>> fs/io_uring.c | 4 ++++
>>>> 1 file changed, 4 insertions(+)
>>>>
>>>> diff --git a/fs/io_uring.c b/fs/io_uring.c
>>>> index 4248726..cf3acaa 100644
>>>> --- a/fs/io_uring.c
>>>> +++ b/fs/io_uring.c
>>>> @@ -7080,6 +7080,8 @@ static int io_sq_offload_start(struct io_ring_ctx *ctx,
>>>> static void io_unaccount_mem(struct user_struct *user, unsigned long nr_pages)
>>>> {
>>>> atomic_long_sub(nr_pages, &user->locked_vm);
>>>> + if (current->mm)
>>>> + atomic_long_sub(nr_pages, ¤t->mm->pinned_vm);
>>>> }
>>>>
>>>> static int io_account_mem(struct user_struct *user, unsigned long nr_pages)
>>>> @@ -7096,6 +7098,8 @@ static int io_account_mem(struct user_struct *user, unsigned long nr_pages)
>>>> return -ENOMEM;
>>>> } while (atomic_long_cmpxchg(&user->locked_vm, cur_pages,
>>>> new_pages) != cur_pages);
>>>> + if (current->mm)
>>>> + atomic_long_add(nr_pages, ¤t->mm->pinned_vm);
>>>>
>>>> return 0;
>>>> }
>>> current->mm should always be valid for these, so I think you can skip the
>>> checking of that and just make it unconditional.
>> Two other issues with this:
>>
>> - It's an atomic64, so seems more appropriate to use the atomic64 helpers
>> for this one.
>> - The unaccount could potentially be a different mm, if the ring is shared
>> and one task sets it up while another tears it down. So we'd need something
>> to ensure consistency here.
>>
> Are you referring to a case where one process creates a ring and sends
> the ring fd to another process?
Or a simpler case, where someone has submissions and completions running
on separate threads, and it just so happens that the completion side is
the one to exit the ring.
--
Jens Axboe
prev parent reply other threads:[~2020-06-14 15:57 UTC|newest]
Thread overview: 11+ messages / expand[flat|nested] mbox.gz Atom feed top
2020-06-12 2:23 [RFC 0/2] io_uring: disallow overlapping ranges for buffer registration Bijan Mottahedeh
2020-06-12 2:23 ` [RFC 1/2] " Bijan Mottahedeh
2020-06-12 15:16 ` Jens Axboe
2020-06-12 18:22 ` Bijan Mottahedeh
2020-06-14 15:56 ` Jens Axboe
2020-06-12 2:23 ` [RFC 2/2] io_uring: report pinned memory usage Bijan Mottahedeh
2020-06-12 15:16 ` Jens Axboe
2020-06-12 15:19 ` Jens Axboe
2020-06-12 15:50 ` Jens Axboe
2020-06-13 4:43 ` Bijan Mottahedeh
2020-06-14 15:57 ` Jens Axboe [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
[email protected] \
[email protected] \
[email protected] \
[email protected] \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox