Message ID | 87zglefhxd.fsf@mpe.ellerman.id.au (mailing list archive) |
---|---|
State | Not Applicable |
Headers | show |
Series | [GIT,PULL] Please pull powerpc/linux.git powerpc-5.18-1 tag | expand |
On Fri, Mar 25, 2022 at 3:25 AM Michael Ellerman <mpe@ellerman.id.au> wrote: > > Livepatch support for 32-bit is probably the standout new feature, otherwise mostly just > lots of bits and pieces all over the board. Heh. I would have expected 32-bit ppc to be entirely legacy by now, so it's a bit surprising to see that being a standout feature. That said: > There's a series of commits cleaning up function descriptor handling, For some reason I also thought that powerpc had actually moved away from function descriptors, so I'm clearly not keeping up with the times. Anyway, unification in that odd area is all good. Pulled, Linus
The pull request you sent on Fri, 25 Mar 2022 21:25:02 +1100:
> https://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux.git tags/powerpc-5.18-1
has been merged into torvalds/linux.git:
https://git.kernel.org/torvalds/c/1f1c153e406a7375ae0fc3d6000b64e7ba27cf8a
Thank you!
Linus Torvalds <torvalds@linux-foundation.org> writes: > On Fri, Mar 25, 2022 at 3:25 AM Michael Ellerman <mpe@ellerman.id.au> wrote: >> >> Livepatch support for 32-bit is probably the standout new feature, otherwise mostly just >> lots of bits and pieces all over the board. > > Heh. I would have expected 32-bit ppc to be entirely legacy by now, so > it's a bit surprising to see that being a standout feature. We still get the odd bug report from people running mainline, or recent stable kernels, on 32-bit. And obviously Christophe has been doing lots of work on the Linux side, so in that sense 32-bit is alive and well. Having said that I don't think we'll see any new 32-bit CPU designs, so the clock is slowly ticking. > That said: > >> There's a series of commits cleaning up function descriptor handling, > > For some reason I also thought that powerpc had actually moved away > from function descriptors, so I'm clearly not keeping up with the > times. No you're right, we have moved away from them, but not entirely. Functions descriptors are still used for 64-bit big endian, but they're not used for 64-bit little endian, or 32-bit. cheers
On Mon, Mar 28, 2022 at 08:07:13PM +1100, Michael Ellerman wrote: > Linus Torvalds <torvalds@linux-foundation.org> writes: > > On Fri, Mar 25, 2022 at 3:25 AM Michael Ellerman <mpe@ellerman.id.au> wrote: > > > That said: > > > >> There's a series of commits cleaning up function descriptor handling, > > > > For some reason I also thought that powerpc had actually moved away > > from function descriptors, so I'm clearly not keeping up with the > > times. > > No you're right, we have moved away from them, but not entirely. > > Functions descriptors are still used for 64-bit big endian, but they're > not used for 64-bit little endian, or 32-bit. There was a patch to use ABIv2 for ppc64 big endian. I suppose that would rid usof the gunction descriptors for good. Somehow the discussion of that change tralied off without any results. Maybe it's worth resurrecting? Thanks Michal
Michal Suchánek <msuchanek@suse.de> writes: > On Mon, Mar 28, 2022 at 08:07:13PM +1100, Michael Ellerman wrote: >> Linus Torvalds <torvalds@linux-foundation.org> writes: >> > On Fri, Mar 25, 2022 at 3:25 AM Michael Ellerman <mpe@ellerman.id.au> wrote: >> >> > That said: >> > >> >> There's a series of commits cleaning up function descriptor handling, >> > >> > For some reason I also thought that powerpc had actually moved away >> > from function descriptors, so I'm clearly not keeping up with the >> > times. >> >> No you're right, we have moved away from them, but not entirely. >> >> Functions descriptors are still used for 64-bit big endian, but they're >> not used for 64-bit little endian, or 32-bit. > > There was a patch to use ABIv2 for ppc64 big endian. I suppose that > would rid usof the gunction descriptors for good. It would be nice. The hesitation in the past was that the GNU toolchain developers don't officially support BE+ELFv2, though it is in use so it does work. > Maybe it's worth resurrecting? Yeah maybe we should think about it again. If it builds with clang/lld that would be a real plus. cheers
On Wed, Mar 30, 2022 at 3:21 PM Michael Ellerman <mpe@ellerman.id.au> wrote: > Michal Suchánek <msuchanek@suse.de> writes: > > On Mon, Mar 28, 2022 at 08:07:13PM +1100, Michael Ellerman wrote: > >> No you're right, we have moved away from them, but not entirely. > >> > >> Functions descriptors are still used for 64-bit big endian, but they're > >> not used for 64-bit little endian, or 32-bit. > > > > There was a patch to use ABIv2 for ppc64 big endian. I suppose that > > would rid usof the gunction descriptors for good. > > It would be nice. > > The hesitation in the past was that the GNU toolchain developers don't > officially support BE+ELFv2, though it is in use so it does work. It clearly made sense to wait while BE+ELFv1 was commonly used and well tested, but as that is getting less common each year, getting ELFv1 out of the picture would appear to make the setup less obscure, not more. Arnd
On Thu, Mar 31, 2022 at 12:21:03AM +1100, Michael Ellerman wrote: > Michal Suchánek <msuchanek@suse.de> writes: > > On Mon, Mar 28, 2022 at 08:07:13PM +1100, Michael Ellerman wrote: > >> Linus Torvalds <torvalds@linux-foundation.org> writes: > >> > On Fri, Mar 25, 2022 at 3:25 AM Michael Ellerman <mpe@ellerman.id.au> wrote: > >> > >> > That said: > >> > > >> >> There's a series of commits cleaning up function descriptor handling, > >> > > >> > For some reason I also thought that powerpc had actually moved away > >> > from function descriptors, so I'm clearly not keeping up with the > >> > times. > >> > >> No you're right, we have moved away from them, but not entirely. > >> > >> Functions descriptors are still used for 64-bit big endian, but they're > >> not used for 64-bit little endian, or 32-bit. > > > > There was a patch to use ABIv2 for ppc64 big endian. I suppose that > > would rid usof the gunction descriptors for good. > > It would be nice. > > The hesitation in the past was that the GNU toolchain developers don't > officially support BE+ELFv2, though it is in use so it does work. We do not officially support ELFv2 BE because there are no significant users, so we cannot have the same confidence it works correctly. It isn't tested often with GCC for example, mainly because it isn't convenient to do without pre-packaged user space for it (and on the other hand, there isn't much demand for it). > > Maybe it's worth resurrecting? > > Yeah maybe we should think about it again. If it builds with clang/lld > that would be a real plus. With GCC it should work fine still. But no doubt you will find some edge cases... which you won't find until you try :-) Segher