[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <5b37e8b1da534ebb8abc9b7b0f7022d2@AcuMS.aculab.com>
Date: Thu, 9 May 2024 21:06:54 +0000
From: David Laight <David.Laight@...LAB.COM>
To: 'Linus Torvalds' <torvalds@...ux-foundation.org>, Theodore Ts'o
<tytso@....edu>
CC: Kees Cook <keescook@...omium.org>, Justin Stitt <justinstitt@...gle.com>,
Peter Zijlstra <peterz@...radead.org>, Mark Rutland <mark.rutland@....com>,
"linux-hardening@...r.kernel.org" <linux-hardening@...r.kernel.org>,
"linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
"llvm@...ts.linux.dev" <llvm@...ts.linux.dev>
Subject: RE: [RFC] Mitigating unexpected arithmetic overflow
...
> Going the other way is similar:
>
> all_bits = low_bits + ((u64) high_bits << 16) << 16);
>
> and again, the compiler will recognize this idiom and do the right
> thing (and if 'all_bits' is only 32-bit, the compiler will optimize
> the high bit noise away).
On a 32bit system the compiler might not do the expected thing.
I had terrible trouble with the 32bit div_u64_u32() code I was
playing with getting the compiler to do 'something sensible' for
that.
I just couldn't get it to stop generating two 64bit values
(in two register pairs) and then oring them together.
I didn't try using a union - that might work.
On x64 the asm "A" (edx:eax) and "a" and "d" constraints will DTRT
but force the values into edx:eax which is ok if you are doing divides.
David
-
Registered Address Lakeside, Bramley Road, Mount Farm, Milton Keynes, MK1 1PT, UK
Registration No: 1397386 (Wales)
Powered by blists - more mailing lists