[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]

Re: please enable XFS in anaconda



Eric Sandeen wrote:
> Eric Sandeen wrote:
> 
>> The remaining issue may be xfs+4kstacks+complicated/deep IO stacks on
>> x86.  Honestly I've never much liked 4kstacks... layered filesystems
>> coming down the pipe (ecryptfs, unionfs) may well have trouble too.
>> I'd prefer to see the stack size boot-time selectable, maybe - or
>> perhaps disallow xfs (or issue a stern warning) on x86 boxes?  (x86_64 &
>> ia64, with sane stack size, work fine in this regard).
> 
> I can confirm that xfs on 4KSTACKS over lvm is still a problem.
> 
> Basic functionality is fine, but I can't make it through all of xfsqa
> w/o a stack overflow.  Even just mounting with -o quota gets us pretty
> close to the edge:
> 
>  mount used greatest stack depth: 180 bytes left
> 
> I'll look at some of the bigger offenders if I have any spare time in
> the evenings.  It's the callchains, too though, not just the individual
> functions that come into play.

Ok, some good news on this front.  With a handful of changes, I got
these stack reductions:

xfs_alloc_file_space [xfs]: 	184 176 :  -8
xfs_bmap_adjacent [xfs]: 	120 136 : +16
xfs_bmapi [xfs]: 		276 268 :  -8
xfs_bunmapi [xfs]: 		200 172 : -28
xfs_create [xfs]: 		116 100 : -16
xfs_free_file_space [xfs]: 	200 132 : -68
xfs_getbmap [xfs]: 		200 204 :  +4
xfs_iomap_write_direct [xfs]: 	180 168 : -12
xfs_link [xfs]: 		 96  88 :  -8
xfs_mkdir [xfs]: 		116  96 : -20
xfs_mountfs [xfs]: 		200 152 : -48
xfs_qm_dqusage_adjust [xfs]: 	 52  48 :  -4
xfs_qm_vop_chown_reserve [xfs]:  48  44 :  -4
xfs_read [xfs]: 		 72  48 : -24
xfs_remove [xfs]: 		 96  84 : -12
xfs_rename [xfs]: 		144 136 :  -8
xfs_rmdir [xfs]: 		 92  76 : -16
xfs_setattr [xfs]: 		132 116 : -16
xfs_splice_read [xfs]: 		 36  16 : -20
xfs_splice_write [xfs]: 	 56  36 : -20
xfs_unmount [xfs]: 		 52   4 : -48
xfs_write [xfs]: 		144 124 : -20
----------------------
Functions increased: 2
Functions decreased: 20
Net change: -388

and I can now survive xfsqa over lvm on 4KSTACKS.  I still get within
spitting distance of the end of the stack though.  :)

I've put the stack-reducing changes & a few bugfixes into
kernel-2.6.23-0.133.rc3.git6.fc8, building now.

-Eric


[Date Prev][Date Next]   [Thread Prev][Thread Next]   [Thread Index] [Date Index] [Author Index]