xfstests, bad generic tests 009 and 308

[Yann Dupont - Veille Techno]

Le 22/09/2015 00:52, Dave Chinner a écrit :
> As it is, I highly recommend that you try a current 4.3 kernel, as 
> there are several code fixes in the XFS kernel code that work around 
> compiler issues we know about. AFAIA, the do_div() asm bug that trips 
> recent gcc optimisations isn't in the upstream kernel yet, but that 
> can be worked around by setting CONFIG_CC_OPTIMIZE_FOR_SIZE=y in your 
> build. 

Hi dave,

I can confirm that CONFIG_CC_OPTIMIZE_FOR_SIZE=y is (was ?) the only way 
for me to have reliable XFS kernel code on different arm platforms 
(Marvell kirkwood, Allwinner A20, Amlogic S805), no matter what recent 
gcc version I've been using.

I must admit I was cross-compiling from X86-64 too, but I think (not 
sure) that it was also the case with native gcc.

I must also admit that I didn't tried since some months, because 
CONFIG_CC_OPTIMIZE_FOR_SIZE=y was the silver bullet for arm xfs kernel 
crashes. This crash was difficult to understand because it occurs quite 
randomly (I.e it can take several hours to trigger)

If there's a patch floating around for gcc (or kernel), I'm interested 
to test.

Cheers,