Received-SPF: pass (google.com: domain of linux-kernel+bounces-158713-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Feedback-ID: iad51458e:Fastmail
Date: Thu, 25 Apr 2024 07:28:45 -0700
From: Boqun Feng <boqun.feng@gmail.com>
To: Andreas Hindborg <nmi@metaspace.dk>
Cc: Alice Ryhl <aliceryhl@google.com>, Thomas Gleixner <tglx@linutronix.de>,
	Miguel Ojeda <ojeda@kernel.org>, John Stultz <jstultz@google.com>,
	Stephen Boyd <sboyd@kernel.org>,
	Alex Gaynor <alex.gaynor@gmail.com>,
	Wedson Almeida Filho <wedsonaf@gmail.com>,
	Gary Guo <gary@garyguo.net>, bjorn3_gh@protonmail.com,
	Benno Lossin <benno.lossin@proton.me>,
	Andreas Hindborg <a.hindborg@samsung.com>,
	rust-for-linux@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 2/2] rust: time: Use wrapping_sub() for Ktime::sub()
Message-ID: <ZiponXmzv1iIO3dJ@Boquns-Mac-mini.home>
References: <20240411230801.1504496-1-boqun.feng@gmail.com>
 <20240411230801.1504496-3-boqun.feng@gmail.com>
 <CAH5fLgg0CkZO9KF58boui6Y5ajpXHuGh0OxyNru2aEG5pNObwQ@mail.gmail.com>
 <Zhk0q-0eM4z5JHqn@Boquns-Mac-mini.home>
 <CAH5fLghL=G-ihevf1_D0aGffmJMtxtSpMDoTGtrmdiDfhwpKnw@mail.gmail.com>
 <877cgln7f4.fsf@metaspace.dk>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=utf-8
Content-Disposition: inline
Content-Transfer-Encoding: 8bit
In-Reply-To: <877cgln7f4.fsf@metaspace.dk>

On Thu, Apr 25, 2024 at 11:00:31AM +0200, Andreas Hindborg wrote:
> Alice Ryhl <aliceryhl@google.com> writes:
> 
> > On Fri, Apr 12, 2024 at 3:18 PM Boqun Feng <boqun.feng@gmail.com> wrote:
> >>
> >> On Fri, Apr 12, 2024 at 10:36:05AM +0200, Alice Ryhl wrote:
> >> > On Fri, Apr 12, 2024 at 1:08 AM Boqun Feng <boqun.feng@gmail.com> wrote:
> >> > >
> >> > > Currently since Rust code is compiled with "-Coverflow-checks=y", so a
> >> > > normal substraction may be compiled as an overflow checking and panic
> >> > > if overflow happens:
> >> > >
> >> > >         subq    %rsi, %rdi
> >> > >         jo      .LBB0_2
> >> > >         movq    %rdi, %rax
> >> > >         retq
> >> > > .LBB0_2:
> >> > >         pushq   %rax
> >> > >         leaq    str.0(%rip), %rdi
> >> > >         leaq    .L__unnamed_1(%rip), %rdx
> >> > >         movl    $33, %esi
> >> > >         callq   *core::panicking::panic::h59297120e85ea178@GOTPCREL(%rip)
> >> > >
> >> > > although overflow detection is nice to have, however this makes
> >> > > `Ktime::sub()` behave differently than `ktime_sub()`, moreover it's not
> >> > > clear that the overflow checking is helpful, since for example, the
> >> > > current binder usage[1] doesn't have the checking.
> >> >
> >> > I don't think this is a good idea at all. Any code that triggers an
> >> > overflow in Ktime::sub is wrong, and anyone who enables
> >> > CONFIG_RUST_OVERFLOW_CHECKS does so because they want such bugs to be
> >> > caught. You may have been able to find one example of a subtraction
> >> > that doesn't have a risk of overflow, but overflow bugs really do
> >>
> >> The point is you won't panic the kernel because of an overflow. I
> >> agree that overflow is something we want to catch, but currently
> >> ktime_t doesn't panic if overflow happens.
> >
> > What the CONFIG_RUST_OVERFLOW_CHECKS option does is enable panics on
> > overflow. So I don't understand how "it panics on overflow" is an
> > argument for removing the overflow check. That's what you asked for!
> > One could perhaps argue about whether CONFIG_RUST_OVERFLOW_CHECKS is a
> > good idea (I think it is), but that is orthogonal. When
> > CONFIG_RUST_OVERFLOW_CHECKS is enabled, you should respect the flag.
> 
> I would agree. If users do not want panics on overflow, they disable
> RUST_OVERFLOW_CHECKS. If the config is enabled, overflows in ktime sub
> should panic, even if it does not do so in equivalent C code.
> 

What's reason then? If we think a piece of API should have a different
semantics than its C counterpart, that suggests we also need to change
the C side for the same reason. Don't keep good stuffs only to ourselves
;-)

Plus, what if users don't want to panic on ktime API but still want
overflow checks for other parts? Using RUST_OVERFLOW_CHECKS to determine
whether ktime should perform overflow checkings seems a bad to me
regardless of what semantics we end up with. For reference, the Rust std
`Duration` performs always overflow checking:

	https://doc.rust-lang.org/src/std/time.rs.html#429-435

We shouldn't define that overflow checking of ktime follows the general
rule of overflow checking of any i64, instead, we should have a clear
rule for it.

Regards,
Boqun

> BR Andreas