linux-kernel - Re: [PATCH] f2fs: fix the discard thread sleep timeout under high utilization

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <0c7220d7-416e-32b7-96cb-effd3f84d6e2@huawei.com>
Date:   Mon, 15 Mar 2021 16:10:22 +0800
From:   Chao Yu <yuchao0@...wei.com>
To:     Sahitya Tummala <stummala@...eaurora.org>
CC:     Jaegeuk Kim <jaegeuk@...nel.org>,
        <linux-f2fs-devel@...ts.sourceforge.net>,
        <linux-kernel@...r.kernel.org>
Subject: Re: [PATCH] f2fs: fix the discard thread sleep timeout under high
 utilization

Hi Sahitya,

On 2021/3/15 15:46, Sahitya Tummala wrote:
> Hi Chao,
> 
> On Mon, Mar 15, 2021 at 02:12:44PM +0800, Chao Yu wrote:
>> Sahitya,
>>
>> On 2021/3/15 12:56, Sahitya Tummala wrote:
>>> When f2fs is heavily utilized over 80%, the current discard policy
>>> sets the max sleep timeout of discard thread as 50ms
>>> (DEF_MIN_DISCARD_ISSUE_TIME). But this is set even when there are
>>> no pending discard commands to be issued. This results into
>>> unnecessary frequent and periodic wake ups of the discard thread.
>>> This patch adds check for pending  discard commands in addition
>>> to heavy utilization condition to prevent those wake ups.
>>
>> Could this commit fix your issue?
>>
>> https://git.kernel.org/pub/scm/linux/kernel/git/jaegeuk/f2fs.git/commit/?h=dev&id=43f8c47ea7d59c7b2270835f1d7c4618a1ea27b6
>>
> I don't think it will help because we are changing the max timeout of the
> dpolicy itself in __init_discard_policy() when util > 80% as below -
> 
> dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
> 
> And issue_discard_thread() uses this value as wait_ms, when there
> are no more pending discard commands to be issued.
> <snip>
>                  } else {
>                          wait_ms = dpolicy.max_interval;
>                  }
> <snip>
> 
> The new patch posted above is not changing anything related to the  max_interval.
> Hence, I think it won't help the uncessary wakeup problem I am trying to solve
> for this condition - util > 80% and when there are no pending discards.
> 
> Please let me know if i am missing something.

Copied, thanks for the explanation.

But there is another case which can cause this issue in the case of
disk util < 80%.

wait_ms = DEF_MIN_DISCARD_ISSUE_TIME;

do {
	wait_event_interruptible_timeout(, wait_ms);

	...

	if (!atomic_read(&dcc->discard_cmd_cnt))
[1] new statement
		continue;

} while();

Then the loop will wakeup whenever 50ms timeout.

So, to avoid this case, shouldn't we reset wait_ms to dpolicy.max_interval
at [1]?

Meanwhile, how about relocating discard_cmd_cnt check after
__init_discard_policy(DPOLICY_FORCE)? and olny set .max_interval to
DEF_MAX_DISCARD_ISSUE_TIME if there is no discard command, and keep
.granularity to 1?

Thanks,

> 
> Thanks,
> Sahitya.
> 
>> Thanks,
>>
>>>
>>> Signed-off-by: Sahitya Tummala <stummala@...eaurora.org>
>>> ---
>>>   fs/f2fs/segment.c | 5 ++++-
>>>   1 file changed, 4 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/fs/f2fs/segment.c b/fs/f2fs/segment.c
>>> index dced46c..df30220 100644
>>> --- a/fs/f2fs/segment.c
>>> +++ b/fs/f2fs/segment.c
>>> @@ -1112,6 +1112,8 @@ static void __init_discard_policy(struct f2fs_sb_info *sbi,
>>>   				struct discard_policy *dpolicy,
>>>   				int discard_type, unsigned int granularity)
>>>   {
>>> +	struct discard_cmd_control *dcc = SM_I(sbi)->dcc_info;
>>> +
>>>   	/* common policy */
>>>   	dpolicy->type = discard_type;
>>>   	dpolicy->sync = true;
>>> @@ -1129,7 +1131,8 @@ static void __init_discard_policy(struct f2fs_sb_info *sbi,
>>>   		dpolicy->io_aware = true;
>>>   		dpolicy->sync = false;
>>>   		dpolicy->ordered = true;
>>> -		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL) {
>>> +		if (utilization(sbi) > DEF_DISCARD_URGENT_UTIL &&
>>> +				atomic_read(&dcc->discard_cmd_cnt)) {
>>>   			dpolicy->granularity = 1;
>>>   			dpolicy->max_interval = DEF_MIN_DISCARD_ISSUE_TIME;
>>>   		}
>>>
>