Debian Bug report logs - #976185
i386 baseline violation in QCMS code

Reply or subscribe to this bug.

Display info messages

Message #5 received at [email protected] (full text, mbox, reply):

From: Fanael Linithien <[email protected]>

To: [email protected]

Subject: i386 baseline violation in QCMS code

Date: Tue, 1 Dec 2020 05:48:08 +0100

Package: firefox
Version: 83.0-1

Firefox crashes with SIGILL on CPUs that don't support SSE2, even
though Debian's baseline doesn't require it.

The crash manifests itself in qcms_transform_data_bgra_out_lut_sse2,
which uses SSE2 intrinsics, and so should never be reached on CPUs
that don't support SSE2, but is, at least on my Pentium III.

I'm 95% sure that the origin of the problem is sse_version_available,
in gfx/qcms/src/transform.rs: it's incorrectly defined to always
return 2, which is correct on AMD64, but not on IA-32:

    /* we know at build time that 64-bit CPUs always have SSE2
     * this tells the compiler that non-SSE2 branches will never be
     * taken (i.e. OK to optimze away the SSE1 and non-SIMD code */
    return 2;

The upstream commit that introduced the regression appears to be
25a50264ae3204f2df9add308fc712df45947637, which also incorrectly
dropped support for SSE 1, which is IIRC still supported by upstream,
it's just not tier 1 anymore and upstream *binaries* don't support it.

Message #10 received at [email protected] (full text, mbox, reply):

From: Fanael Linithien <[email protected]>

To: [email protected]

Subject: i386 baseline violation in QCMS code

Date: Mon, 29 Mar 2021 12:23:07 +0200

control: fixed -1 86.0.1-1

This bug has been fixed upstream, likely by
https://phabricator.services.mozilla.com/D100293, and the Firefox
version currently in sid works fine on that hardware. As I don't know
how to close bug reports, or if it's even possible at all, I'm leaving
that to you.

Send a report that this bug log contains spam.