Linus, I'm not sure what's going on here, this patch set seems to be in some sort of "ignore list" on Peter's side with no indication on its destiny. I'd really like for this change to go into the new release with the rest of uprobe improvements that happened this cycle, as they all nicely complement each other. This patch set has been done-done since Oct 24 when Suren sent the final version of mm-side changes ([0]), which I subsequently resent as part of this mm+uprobe patch set on Oct 27, after coordinating that this will go through uprobe subsystem with Andrew Morton ([1]). The uprobe part was effectively unchanged since this summer, when this speculative uprobe lookup logic was posted as part of an earlier RFC series ([2]). That's just to say that this was thoroughly reviewed, discussed, and stress-tested, meanwhile, and I see no reason to delay landing it for so long. I've even written a separate overview email with a summary of all the uprobe-related work and how it all fits together ([3]), realizing that there are a few seemingly independent email threads and patch sets, trying to engage involved maintainers. The outcome was: - two patch sets did land (uretprobe + SRCU and Jiri's uprobe session prerequisites) after a bunch of extra pings, but that's at least something; - Liao's siglock optimization ([4]) still hasn't landed with no explanation what's the delay; - this patch set is also stuck in limbo for weeks now; - there was little engagement on arm64 front for Liao's optimization of uprobes on STP instructions [5], which is perhaps a separate topic for another email, but just another instance of maintainers not engaging in timely fashion. In short, I hope to get your help with the next steps. What can I do to help land this patch set (and hopefully also others I mentioned above)? More broadly, what should be contributors' expectations on timeliness of maintainers' engagement? Maintainer record in MAINTAINERS can't be just a veto power, right? It is also a responsibility before others to move the kernel development along. I'd like to understand what you think is reasonable to expect here? Same question for patch handling (applying, reviewing, rejecting, etc.) latency. Thank you! [0] https://lore.kernel.org/linux-mm/20241024205231.1944747-1-surenb@xxxxxxxxxx/ [1] https://lore.kernel.org/linux-mm/20241028204822.6638f330fad809381eafb49c@xxxxxxxxxxxxxxxxxxxx/ [2] https://lore.kernel.org/linux-trace-kernel/20240813042917.506057-14-andrii@xxxxxxxxxx/ [3] https://lore.kernel.org/linux-trace-kernel/CAEf4BzY-0Eu27jyT_s2kRO1UuUPOkE9_SRrBOqu2gJfmxsv+3A@xxxxxxxxxxxxxx/ [4] https://lore.kernel.org/linux-trace-kernel/CAEf4BzarhiBHAQXECJzP5e-z0fbSaTpfQNPaSXwdgErz2f0vUA@xxxxxxxxxxxxxx/ [5] https://lore.kernel.org/linux-trace-kernel/CAEf4BzZ3trjMWjvWX4Zy1GzW5RN1ihXZSnLZax7V-mCzAUg2cg@xxxxxxxxxxxxxx/ [6] https://lore.kernel.org/all/172074397710.247544.17045299807723238107.stgit@devnote2/ On Mon, Nov 11, 2024 at 9:26 AM Andrii Nakryiko <andrii.nakryiko@xxxxxxxxx> wrote: > > On Tue, Nov 5, 2024 at 6:01 PM Andrii Nakryiko > <andrii.nakryiko@xxxxxxxxx> wrote: > > > > On Sun, Oct 27, 2024 at 6:09 PM Andrii Nakryiko <andrii@xxxxxxxxxx> wrote: > > > > > > Implement speculative (lockless) resolution of VMA to inode to uprobe, > > > bypassing the need to take mmap_lock for reads, if possible. First two patches > > > by Suren adds mm_struct helpers that help detect whether mm_struct was > > > changed, which is used by uprobe logic to validate that speculative results > > > can be trusted after all the lookup logic results in a valid uprobe instance. > > > > > > Patch #3 is a simplification to uprobe VMA flag checking, suggested by Oleg. > > > > > > And, finally, patch #4 is the speculative VMA-to-uprobe resolution logic > > > itself, and is the focal point of this patch set. It makes entry uprobes in > > > common case scale very well with number of CPUs, as we avoid any locking or > > > cache line bouncing between CPUs. See corresponding patch for details and > > > benchmarking results. > > > > > > Note, this patch set assumes that FMODE_BACKING files were switched to have > > > SLAB_TYPE_SAFE_BY_RCU semantics, which was recently done by Christian Brauner > > > in [0]. This change can be pulled into perf/core through stable > > > tags/vfs-6.13.for-bpf.file tag from [1]. > > > > > > [0] https://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git/commit/?h=vfs-6.13.for-bpf.file&id=8b1bc2590af61129b82a189e9dc7c2804c34400e > > > [1] git://git.kernel.org/pub/scm/linux/kernel/git/vfs/vfs.git > > > > > > v3->v4: > > > - rebased and dropped data_race(), given mm_struct uses real seqcount (Peter); > > > v2->v3: > > > - dropped kfree_rcu() patch (Christian); > > > - added data_race() annotations for fields of vma and vma->vm_file which could > > > be modified during speculative lookup (Oleg); > > > - fixed int->long problem in stubs for mmap_lock_speculation_{start,end}(), > > > caught by Kernel test robot; > > > v1->v2: > > > - adjusted vma_end_write_all() comment to point out it should never be called > > > manually now, but I wasn't sure how ACQUIRE/RELEASE comments should be > > > reworded (previously requested by Jann), so I'd appreciate some help there > > > (Jann); > > > - int -> long change for mm_lock_seq, as agreed at LPC2024 (Jann, Suren, Liam); > > > - kfree_rcu_mightsleep() for FMODE_BACKING (Suren, Christian); > > > - vm_flags simplification in find_active_uprobe_rcu() and > > > find_active_uprobe_speculative() (Oleg); > > > - guard(rcu)() simplified find_active_uprobe_speculative() implementation. > > > > > > Andrii Nakryiko (2): > > > uprobes: simplify find_active_uprobe_rcu() VMA checks > > > uprobes: add speculative lockless VMA-to-inode-to-uprobe resolution > > > > > > Suren Baghdasaryan (2): > > > mm: Convert mm_lock_seq to a proper seqcount > > > mm: Introduce mmap_lock_speculation_{begin|end} > > > > > > include/linux/mm.h | 12 ++--- > > > include/linux/mm_types.h | 7 ++- > > > include/linux/mmap_lock.h | 87 ++++++++++++++++++++++++-------- > > > kernel/events/uprobes.c | 47 ++++++++++++++++- > > > kernel/fork.c | 5 +- > > > mm/init-mm.c | 2 +- > > > tools/testing/vma/vma.c | 4 +- > > > tools/testing/vma/vma_internal.h | 4 +- > > > 8 files changed, 129 insertions(+), 39 deletions(-) > > > > > > -- > > > 2.43.5 > > > > > > > Hi! > > > > What's the status of this patch set? Are there any blockers for it to > > be applied to perf/core? MM folks are OK with landing the first two > > patches in perf/core, so hopefully we should be good to go? > > Another week, another ping. Peter, what can I do to make this land? MM > parts are clearly ok with Andrew Morton, uprobe-side logic didn't > change (modulo inconsequential data_race() back and forth) since at > least August, was approved by Oleg, and seems to be very stable in > testing. I think it's time to let me forget about this patch set and > make actual use of it in production, please.
