[PATCH] btrfs: Disable BTRFS on platforms having 256K pages
David Sterba
dsterba at suse.cz
Fri Jun 11 02:20:46 AEST 2021
On Thu, Jun 10, 2021 at 04:50:09PM +0200, Christophe Leroy wrote:
>
>
> Le 10/06/2021 à 15:54, Chris Mason a écrit :
> >
> >> On Jun 10, 2021, at 1:23 AM, Christophe Leroy <christophe.leroy at csgroup.eu> wrote:
> >>
> >> With a config having PAGE_SIZE set to 256K, BTRFS build fails
> >> with the following message
> >>
> >> include/linux/compiler_types.h:326:38: error: call to '__compiletime_assert_791' declared with attribute error: BUILD_BUG_ON failed: (BTRFS_MAX_COMPRESSED % PAGE_SIZE) != 0
> >>
> >> BTRFS_MAX_COMPRESSED being 128K, BTRFS cannot support platforms with
> >> 256K pages at the time being.
> >>
> >> There are two platforms that can select 256K pages:
> >> - hexagon
> >> - powerpc
> >>
> >> Disable BTRFS when 256K page size is selected.
> >>
> >
> > We’ll have other subpage blocksize concerns with 256K pages, but this BTRFS_MAX_COMPRESSED #define is arbitrary. It’s just trying to have an upper bound on the amount of memory we’ll need to uncompress a single page’s worth of random reads.
> >
> > We could change it to max(PAGE_SIZE, 128K) or just bump to 256K.
> >
>
> But if 256K is problematic in other ways, is it worth bumping BTRFS_MAX_COMPRESSED to 256K ?
>
> David, in below mail, said that 256K support would require deaper changes. So disabling BTRFS
> support seems the easiest solution for the time being, at least for Stable (I forgot the Fixes: tag
> and the CC: to stable).
>
> On powerpc, 256k pages is a corner case, it requires customised binutils, so I don't think disabling
> BTRFS is a issue there. For hexagon I don't know.
That it blew up due to the max compressed size is a coincidence. We
could have explicit BUILD_BUG_ONs for page size or other constraints
derived from the page size like INLINE_EXTENT_BUFFER_PAGES.
And there's no such thing like "just bump BTRFS_MAX_COMPRESSED to 256K".
The constant is part of on-disk format for lzo and otherwise changing it
would impact performance so this would need proper evaluation.
More information about the Linuxppc-dev
mailing list