linux-kernel - Re: [PATCH v8 6/7] rust: Add read_poll

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: <20250123.162520.1427672620079645345.fujita.tomonori@gmail.com>
Date: Thu, 23 Jan 2025 16:25:20 +0900 (JST)
From: FUJITA Tomonori <fujita.tomonori@...il.com>
To: gary@...yguo.net
Cc: fujita.tomonori@...il.com, linux-kernel@...r.kernel.org,
 boqun.feng@...il.com, rust-for-linux@...r.kernel.org,
 netdev@...r.kernel.org, andrew@...n.ch, hkallweit1@...il.com,
 tmgross@...ch.edu, ojeda@...nel.org, alex.gaynor@...il.com,
 bjorn3_gh@...tonmail.com, benno.lossin@...ton.me, a.hindborg@...sung.com,
 aliceryhl@...gle.com, anna-maria@...utronix.de, frederic@...nel.org,
 tglx@...utronix.de, arnd@...db.de, jstultz@...gle.com, sboyd@...nel.org,
 mingo@...hat.com, peterz@...radead.org, juri.lelli@...hat.com,
 vincent.guittot@...aro.org, dietmar.eggemann@....com, rostedt@...dmis.org,
 bsegall@...gle.com, mgorman@...e.de, vschneid@...hat.com
Subject: Re: [PATCH v8 6/7] rust: Add read_poll_timeout functions

On Wed, 22 Jan 2025 18:36:12 +0000
Gary Guo <gary@...yguo.net> wrote:

>> +#[track_caller]
>> +pub fn read_poll_timeout<Op, Cond, T: Copy>(
> 
> I wonder if we can lift the `T: Copy` restriction and have `Cond` take
> `&T` instead. I can see this being useful in many cases.
> 
> I know that quite often `T` is just an integer so you'd want to pass by
> value, but I think almost always `Cond` is a very simple closure so
> inlining would take place and they won't make a performance difference.

Yeah, we can. More handy for the users of this function. I'll do.

>> +    mut op: Op,
>> +    cond: Cond,
>> +    sleep_delta: Delta,
>> +    timeout_delta: Delta,
>> +) -> Result<T>
>> +where
>> +    Op: FnMut() -> Result<T>,
>> +    Cond: Fn(T) -> bool,
>> +{
>> +    let start = Instant::now();
>> +    let sleep = !sleep_delta.is_zero();
>> +    let timeout = !timeout_delta.is_zero();
>> +
>> +    might_sleep(Location::caller());
> 
> This should only be called if `timeout` is true?

Oops, I messed up this in v6 somehow. I'll fix.

>> +    let val = loop {
>> +        let val = op()?;
>> +        if cond(val) {
>> +            // Unlike the C version, we immediately return.
>> +            // We know a condition is met so we don't need to check again.
>> +            return Ok(val);
>> +        }
>> +        if timeout && start.elapsed() > timeout_delta {
>> +            // Should we return Err(ETIMEDOUT) here instead of call op() again
>> +            // without a sleep between? But we follow the C version. op() could
>> +            // take some time so might be worth checking again.
>> +            break op()?;
> 
> Maybe the reason is `ktime_get` can take some time (due to its use of
> seqlock and thus may require retrying?) Although this logic breaks down
> when `read_poll_timeout_atomic` also has this extra `op(args)` despite
> the condition being trivial.

ktime_get() might do retrying (read_seqcount) but compared to the op
function, I think that ktime_get() is fast (usually an op function
waits for hardware).

> So I really can't convince myself that this additional `op()` call is
> needed. I can't think of any case where this behaviour would be
> depended on by a driver, so I'd be tempted just to return ETIMEOUT
> straight.

As I commented in the code, I just mimic the logic of the C version,
which has been used for a long time. But as you said, looks like we
can return Err(ETIMEOUT) immediately here. I'll do that in the next
version.