On Sat, Oct 22, 2005 at 04:56:10PM -0400, Dave Jones wrote: > On Sat, Oct 22, 2005 at 07:25:57AM -0400, Steve Snyder wrote: > > I'm having a hard time understanding the thinking of the FC developers > > regarding the kernel "Optimize for size" option. > > > > Earlier 2.6.x kernels had this disabled, then they enabled it for a while, now > > it's disabled in current FC4 kernels. Meanwhile, the current RHEL4 kernel > > does enable the "Optimize for size" option, albeit with earlier compiler > > (GCC v3.4.4) and kernel (2.6.9. + patches) version. > > > > What's going on here? Does the value of this option change with differing > > kernel versions? How does one determine empirically whether this option > > should be enabled or not? > > As a rule, it's meant to be enabled. The gcc in rawhide hit a compiler > bug, so whilst that got worked out, it got disabled for a while. > Whilst it was disabled however, the rebase to newer upstream happened, > and the disable crept into FC4 too. > > Thanks for spotting it, I'll flip it back on in the next FC4 errata. > It's probably time to try it again in rawhide now too. And time to turn it back off again. Turns out that its quite broken on x86-64 right now in rawhide. Tracking down what exactly is getting miscompiled is going to be tricky, as the symptoms are random crashes. It's going to require compiling some parts of the kernel with -Os and the rest with -O2 and testing until we find the area to narrow down into, which is tedious work (to put it lightly). For now, I'm just disabling it again, as we've bigger fish to fry in rawhide right now. I'm also somewhat hesitant about enabling it in FC4 again given the recent rawhide results, so it's likely that I'm going to go back on what I said last week, and leave it off for the next errata there too. Dave
