1522203 - Clang warns about unsequenced code in Pentium Pro workaround

Status: RESOLVED FIXED

(NSS :: Libraries, defect, P1)

Description

[(Away)]

https://searchfox.org/mozilla-central/rev/39265dd58298c40bed029617ad408bf806cce761/security/nss/lib/pk11wrap/pk11slot.c#2099-2102,2128-2137

#if defined(XP_WIN32)
/* This is works around some horrible cache/page thrashing problems
** on Win32. Without this, this loop can take up to 6 seconds at
** 100% CPU on a Pentium-Pro 200. The thing this changes is to
** increase the size of the stack frame and modify it.
** Moving the loop code itself seems to have no effect.
** Dunno why this combination makes a difference, but it does.
*/
waste[j & 0xf] = j++;
#endif

I only noticed it because I happened to see this warning scroll past during a build:

17:12:55 INFO - z:/build/build/src/security/nss/lib/pk11wrap/pk11slot.c(2136,27): warning: unsequenced modification and access to 'j' [-Wunsequenced]
17:12:55 INFO - waste[j & 0xf] = j++;

Do we still need this workaround?

Comment 1

[J.C. Jones [:jcj] (he/they)]

That's amazing.

I'll patch this out in NSS 3.43.

Comment 2

[(Away)]

(In reply to J.C. Jones [:jcj] (he/him) from comment #1)

I'll patch this out in NSS 3.43.

Is this still the plan? :-)

Comment 3

[J.C. Jones [:jcj] (he/they)]

Attachment: Bug 1522203 - Remove Pentium Pro workaround for PK11_GetAllTokens r?kjacobs

The comment indicated the wasted effort was to work around a cache issue on the
Pentium Pro. I think it has served its purpose.

Comment 4

[J.C. Jones [:jcj] (he/they)]

https://hg.mozilla.org/projects/nss/rev/27d9fb4ac69b7541ea2051e1cda85799e9f099ce