#
✕ nick:arnd
#asahi · 2021-03-26 19:24
arndI can also give some Reviewed-by replies for the patches I have looked at in detail, though they are not that meaningful if I'm also the one to merge the patches
#asahi · 2021-03-26 19:21
arndOnce you have those, I can pull it into the soc tree for linux-next, and whatever comes up later can be add-on patches
#asahi · 2021-03-26 19:20
arndI'd hope to get a Reviewed-by from maz, plus an Acked-by from tglx for merging it through the soc tree instead of the tip/irqchip tree
#asahi · 2021-03-26 19:19
arnd- finally, I see the irqchip driver is still missing an Ack from the irqchip maintainers. I don't know if you already point this out in the cover letter, otherwise please add it
#asahi · 2021-03-26 19:14
arndactually all of the commits
#asahi · 2021-03-26 19:13
arnde.g. on https://github.com/AsahiLinux/linux/commit/07f6504a82ffe78aba8fd78618444a1b8895f3ee
#asahi · 2021-03-26 19:13
arnd- When you add Reviewed-by or similar tags, please keep your own Signed-off-by as the last tag in the list
#asahi · 2021-03-26 19:12
arnd- for the merge commits, please add a little more detail on why it gets merged in the changelog. I would also add a Signed-off-by for the merge, though that is not strictly required
#asahi · 2021-03-26 19:12
arndmarcan: a few formal details:
#asahi · 2021-03-26 15:35
arndeta: rerere only remembers the changes you did yourself, it doesn't get put in the history
#asahi · 2021-03-26 09:36
arndwhat I'd suggest is to merge 71b25f4df984 into your branch and resolve the conflict the same way that gregkh did in his tree. Then try to merge tty-next on top of that. If that is a clean merge, it should be fine, otherwise I can see if I have a better idea
#asahi · 2021-03-26 09:35
arndit might still work
#asahi · 2021-03-26 09:35
arndit doesn't consider them identical if they have different commit IDs and there was a merge conflict in the old version
#asahi · 2021-03-26 09:34
arndno, cherry-picking makes it worse because then torvalds has to resolve that merge again
#asahi · 2021-03-26 09:29
arndI'd prefer not to have the same backmerge of in my tree (it messes up the history), so maybe either merge in 71b25f4df984 (the last commit for the samsung serial driver) or 3d82be8be6fd (the last commit before the backmerge) instead of the entire tty-next
#asahi · 2021-03-26 09:28
arndmarcan: tty-next seems to have a backmerge of -rc3 into that, with the previous patches based on -rc2
#asahi · 2021-03-26 09:25
arndLet me check out tty-next to see if there is a sensible way of having a subset of that
#asahi · 2021-03-26 09:24
arndmarcan: I would start out with -rc3, and then "git pull --no-ff" the other branches into that one at a time, with a merge changeset describing why you need then.
#asahi · 2021-03-26 09:14
arndmarcan: if nothing of the additional feedback is required for correct operation of the driver, I'd suggest you keep those fixups separate from the rest of the series and just send them to gregkh as cleanups
#asahi · 2021-03-26 09:08
arndI hope there are not too many additional branches you depend on
#asahi · 2021-03-26 09:08
arndmarcan: cmarinas mentioned that he also has picked up maz' patches for GENERIC_IRQ_MULTI_HANDLER into the arm64 tree (I don't know what patches those are, but I assume you do), so you'll want to merge that branch and the serial driver on top of -rc3, and then rebased the rest on top of that
#asahi · 2021-03-26 09:06
arndthe usual procedure for the serial drivers is that gregkh would wait for an Acked-by or Reviewed-by tag from krzk and then apply the patches directly. There is rarely the need for gregkh to require a pull request from a driver maintainer
#asahi · 2021-03-26 09:04
arndwhat we could do in theory is to have your branch based on the serial git, or a merge of that into another branch. I'd probably reserve a top-level branch in the soc tree for m1, so that this time we have "arm/dt", "arm/defconfig", "arm/drivers", "arm/soc" and "arm/m1" branches, and after 5.13 you split the patches into dt, defconfig, drivers, etc as everyone else
#asahi · 2021-03-26 09:01
arndnote that gregkh doesn't remove stuff from his branch, so it's now part of the git history, though he could revert the patches if necessary
#asahi · 2021-03-26 08:59
arndI think gregkh already merged the serial driver patches into his tree, not sure if that is a problem for having everything in one branch
#asahi · 2021-03-26 08:59
arndok, good
#asahi · 2021-03-26 08:58
arndmarcan: do you think there are any show-stoppers left? Final consensus on the "non-posted" property is the only thing I still see as necessary, not sure where we actually are on that, or if there are any other issues. Once there are only minor style issues left, I would like to just merge a branch in to my for-next tree and let you follow up with patches on top of that
#asahi · 2021-03-13 20:57
arndStefano Stabellini has done some interesting work on trimming down Xen to the point where it mainly does static partitioning and is overall very small, see https://xenproject.org/2019/12/16/true-static-partitioning-with-xen-dom0-less/ for an older talk from him
#asahi · 2021-03-13 20:54
arndjeffmiw: right, rustvmm is a set of library that can do different parts of the whole solution, but I suppose the part you need most is probably one that is not there
#asahi · 2021-03-13 20:43
arndjeffmiw: all the cool kids these days build on top of rustvmm
#asahi · 2021-03-10 19:27
arndhttps://www.blopeur.com/2020/04/08/Intel-x86-patent-never-ending.html links to an sse3 patent expiring in 2026, sse2 would clearly be earlier than that, but may not be expired yet. Which of these are actually emulated in rosetta2?
#asahi · 2021-03-10 19:13
arndmaybe it's not really a coincidence that microsoft and apple both start emulating 64-bit x86 around the same time, after years of only 32-bit emulation in windows on arm
#asahi · 2021-03-08 13:13
arndok, so it's not just emulated in qemu then
#asahi · 2021-03-08 12:43
arnddoes it need a similar hack?
#asahi · 2021-03-08 12:43
arndmaz: how does qemu/hvf do it?
#asahi · 2021-03-05 15:21
arndthe 30% improvement came from shrinking the size of the average preprocessed file by 80%, which means that clang spends way too much time parsing and analyzing code that it clearly never uses
#asahi · 2021-03-05 15:20
arndbut fixing the indirect header inclusions would also help a ton with marcan's rebuild since that avoids rebuilding a lot of files that don't care about the changed headers
#asahi · 2021-03-05 15:19
arnd30% was for clang, the difference on gcc was much less because gcc does dead code elimination much earlier during a compile job
#asahi · 2021-03-05 15:17
arndI had a series that improved build times by about 30%, by shrinking recursive header inclusions in kernel headers, but I never got that to the point of submitting upstream
#asahi · 2021-03-05 15:16
arndmy randoconfig builds have recently gotten a little slower, need to investigate. I'm down to around 20 kernels per hour using clang-13 on a 16-core tr1950x, it used to be over 30 kernels per hour
#asahi · 2021-03-05 15:14
arndmarcan: for this kind of work, I find ccache helps the most, as you'll constantly be rebuilding the same files you have built before. the few times I tried distcc, I didn't have much success, as it tended to be faster to rely on everything being in RAM on a reasonably fast box
#asahi · 2021-03-04 20:20
arndmarcan: yes, I was going to suggest that as well, he usually doesn't care about the bindings
#asahi · 2021-03-04 20:18
arndah right
#asahi · 2021-03-04 20:15
arndwhy is davem on Cc?
#asahi · 2021-03-04 20:15
arndmarcan: maybe find one more to drop to allow for replies to replies
#asahi · 2021-03-04 20:11
arnddrivers/video/fbdev is basically orphaned, the gpu maintainers apply the patches if nobody else will, but I assume they are happy not to have to do it
#asahi · 2021-03-04 20:10
arndI'm happy to take that one in with an Ack for the binding change from robher
#asahi · 2021-03-04 20:09
arndforget about the last suggesting, moving names into 'to' means your email goes through, but any 'reply-to-all' won't
#asahi · 2021-03-04 20:08
arndyou can try dropping the gpu maintainers (they won't care), or dropping the cleartext names, or moving some more names names into To:
#asahi · 2021-03-04 20:07
arndmarcan: you should stay below 1024 characters for the Cc: line, I count 1137 characters, which means some mailing lists with discard the email

plus ›