Hi,
As I promised on the "AUTOSEL process" centi-thread
(https://lore.kernel.org/stable/[email protected]/T/#u),
I've developed some new scripts that can be found at
https://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/stable-tools.git :
- `find-orig-patch`: Finds the original patch email from a git commit.
It first checks for a matching "Message-Id:" or "Link:" from the git commit.
If that fails, it falls back to a search of https://lore.kernel.org by
commit title and uses some heuristics to try to find the right patch email.
- `find-orig-series`: Like find-orig-patch but outputs the full series.
- `find-missing-prereqs`: Finds commits that were backported without previous
patches in their original series also being backported. It accepts a range of
git commits. I also added an option to filter the results by AUTOSEL only.
For more information, see the README at
https://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/stable-tools.git/tree/README.md
Note: since it wasn't clear where to put these or how to integrate them into
anything else, for now this is a completely standalone project. Perhaps the
find-orig-patch functionality would be a nice feature for b4, but the more
stable-kernel-maintenance-specific logic should be in a separate project. BTW,
I used the same language and license as b4: Python and GPLv2+.
I wrote some regression tests that show that find-missing-prereqs is able to
detect the missing patches for a couple examples of missed backports that broke
users, including the recent blk-cgroup one
(https://lore.kernel.org/linux-block/CAOCAAm4reGhz400DSVrh0BetYD3Ljr2CZen7_3D4gXYYdB4SKQ@mail.gmail.com/T/#u).
For another example, at the end of this email I've also pasted the output of
'find-missing-prereqs v6.1.24..v6.1.26', which covers the last couple weeks of
6.1. I don't immediately see anything super interesting in there, and it picked
up a few very long patchsets which is a bit annoying (maybe very long patchsets
generally aren't interesting and should be skipped?). But it's just an example.
Something that could be built on top of this is a script that applies the
patches from the stable-queue repository for each stable kernel version, and
generates a report about each one. BTW, I can work on these scripts more, but
what I can't commit to doing is manually sending out reports every week... I
hope that this can be automated and/or adopted by the stable maintainers.
Here's the output of 'find-missing-prereqs v6.1.24..v6.1.26':
The following commit(s):
[PATCH 24/33] commit 779fd2a575cc ("drm/amd/display: Pass the right info to drm_dp_remove_payload")
... are backported without earlier commit(s) in series:
[PATCH 1/33] commit de534c1cb031 ("drm/amd/display: Add height granularity limitation for dsc slice height calculation")
[PATCH 2/33] commit aee0c07a74d3 ("drm/amd/display: Unify DC logging for BW Alloc")
[PATCH 3/33] commit 67d198da2fd4 ("drm/amd/display: When blanking during init loop to find OPP index")
[PATCH 4/33] commit c93aa7f33e94 ("drm/amd/display: 3.2.225")
[PATCH 5/33] commit 0db13eae41fc ("drm/amd/display: Add minimum Z8 residency debug option")
[PATCH 6/33] commit 0215ce9057ed ("drm/amd/display: Update minimum stutter residency for DCN314 Z8")
[PATCH 7/33] commit c0a561d96a28 ("drm/amd/display: Drop CONFIG_DRM_AMD_DC_HDR")
[PATCH 8/33] commit 11efe095dfe0 ("drm/amd/display: Fix no-DCN build")
[PATCH 9/33] commit ab487ea8910d ("drm/amd/display: fix typo in dc_dsc_config_options structure")
[PATCH 10/33] commit 1e88eb1b2c25 ("drm/amd/display: Drop CONFIG_DRM_AMD_DC_HDCP")
[PATCH 11/33] commit efa4c4df864e ("drm/amd/display: call remove_stream_from_ctx from res_pool funcs")
[PATCH 12/33] commit 84c03df58d8b ("drm/amd/display: Build DSC without DCN config")
[PATCH 13/33] commit 36516001a7c9 ("drm/amd/display: move dc_link functions in accessories folder to dc_link_exports")
[PATCH 14/33] commit 76f5dc40ebb1 ("drm/amd/display: move dc_link functions in link root folder to dc_link_exports")
[PATCH 15/33] commit 6455cb522191 ("drm/amd/display: link link_dp_dpia_bw.o in makefile")
[PATCH 16/33] commit 202a3816f37e ("drm/amd/display: move dc_link functions in protocols folder to dc_link_exports")
[PATCH 17/33] commit 788c6e2ce5c7 ("drm/amd/display: replace all dc_link function call in link with link functions")
[PATCH 18/33] commit 34fd6df78869 ("drm/amd/display: Simplify register offsets")
[PATCH 19/33] commit 2b02d746c181 ("drm/amd/display: Keep PHY active for dp config")
[PATCH 21/33] commit bf77fda02411 ("drm/amd/display: Drop unnecessary DCN guards")
[PATCH 22/33] commit 4652ae7a51b7 ("drm/amd/display: Rename DCN config to FP")
[PATCH 23/33] commit de930140bb57 ("drm/amd/display: Update to correct min FCLK when construction BB")
Original patch series is "[PATCH 00/33] DC Patches Mar 6th, 2023"
(https://lore.kernel.org/r/[email protected])
The following commit(s):
[PATCH 2/4] commit 3570f3cc4aab ("RDMA/erdma: Update default EQ depth to 4096 and max_send_wr to 8192")
[PATCH 3/4] commit d682c9bc41fa ("RDMA/erdma: Inline mtt entries into WQE if supported")
[PATCH 4/4] commit 132918e08e86 ("RDMA/erdma: Defer probing if netdevice can not be found")
... are backported without earlier commit(s) in series:
[PATCH 1/4] commit 3fe26c0493e4 ("RDMA/erdma: Fix some typos")
Original patch series is "[PATCH for-rc 0/4] RDMA/erdma: erdma fixes 3-20-2023"
(https://lore.kernel.org/r/[email protected])
The following commit(s):
[PATCH 15/26] commit 361b02e68181 ("KVM: arm64: Initialise hypervisor copies of host symbols unconditionally")
... are backported without earlier commit(s) in series:
[PATCH 1/26] commit 0f4f7ae10ee4 ("KVM: arm64: Move hyp refcount manipulation helpers to common header file")
[PATCH 2/26] commit 72a5bc0f153c ("KVM: arm64: Allow attaching of non-coalescable pages to a hyp pool")
[PATCH 3/26] commit 8e6bcc3a4502 ("KVM: arm64: Back the hypervisor 'struct hyp_page' array for all memory")
[PATCH 4/26] commit 0d16d12eb26e ("KVM: arm64: Fix-up hyp stage-1 refcounts for all pages mapped at EL2")
[PATCH 5/26] commit 33bc332d4061 ("KVM: arm64: Unify identifiers used to distinguish host and hypervisor")
[PATCH 6/26] commit 1ed5c24c26f4 ("KVM: arm64: Implement do_donate() helper for donating memory")
[PATCH 7/26] commit 43c1ff8b7501 ("KVM: arm64: Prevent the donation of no-map pages")
[PATCH 8/26] commit 9926cfce8dcb ("KVM: arm64: Add helpers to pin memory shared with the hypervisor at EL2")
[PATCH 9/26] commit 4d968b12e6bb ("KVM: arm64: Include asm/kvm_mmu.h in nvhe/mem_protect.h")
[PATCH 10/26] commit 1c80002e3264 ("KVM: arm64: Add hyp_spinlock_t static initializer")
[PATCH 11/26] commit 5304002dc375 ("KVM: arm64: Rename 'host_kvm' to 'host_mmu'")
[PATCH 12/26] commit a1ec5c70d3f6 ("KVM: arm64: Add infrastructure to create and track pKVM instances at EL2")
[PATCH 13/26] commit 9d0c063a4d1d ("KVM: arm64: Instantiate pKVM hypervisor VM and vCPU structures from EL1")
[PATCH 14/26] commit aa6948f82f0b ("KVM: arm64: Add per-cpu fixmap infrastructure at EL2")
Original patch series is "[PATCH v6 00/26] KVM: arm64: Introduce pKVM hyp VM and vCPU state at EL2"
(https://lore.kernel.org/r/[email protected])
The following commit(s):
[PATCH 2/2] commit 2fcfd51add22 ("Bluetooth: SCO: Fix possible circular locking dependency sco_sock_getsockopt")
... are backported without earlier commit(s) in series:
[PATCH 1/2] commit 9a8ec9e8ebb5 ("Bluetooth: SCO: Fix possible circular locking dependency on sco_connect_cfm")
The following commit(s):
[PATCH 3/4] commit 5620eeb379d1 ("tracing: Add trace_array_puts() to write into instance")
... are backported without earlier commit(s) in series:
[PATCH 1/4] commit cb1f98c5e574 ("tracing: Add creation of instances at boot command line")
[PATCH 2/4] commit c4846480831e ("tracing: Add enabling of events to boot instances")
Original patch series is "[PATCH v2 0/4] tracing: Addition of tracing instances via kernel command line"
(https://lore.kernel.org/r/[email protected])
The following commit(s):
[PATCH 36/66] commit 4ac57c3fe2c0 ("drm/amd/display: set dcn315 lb bpp to 48")
... are backported without earlier commit(s) in series:
[PATCH 31/66] commit 0b5dfe12755f ("drm/amd/display: fix a divided-by-zero error")
[PATCH 35/66] commit 1e994cc0956b ("drm/amd/display: limit timing for single dimm memory")
Original patch series is "[PATCH 00/66] DC Patches Apr 17th, 2023"
(https://lore.kernel.org/r/[email protected])
On Sat, Apr 29, 2023 at 11:49:48PM -0700, Eric Biggers wrote:
>Hi,
>
>As I promised on the "AUTOSEL process" centi-thread
>(https://lore.kernel.org/stable/[email protected]/T/#u),
>I've developed some new scripts that can be found at
>https://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/stable-tools.git :
>
>- `find-orig-patch`: Finds the original patch email from a git commit.
> It first checks for a matching "Message-Id:" or "Link:" from the git commit.
> If that fails, it falls back to a search of https://lore.kernel.org by
> commit title and uses some heuristics to try to find the right patch email.
>
>- `find-orig-series`: Like find-orig-patch but outputs the full series.
>
>- `find-missing-prereqs`: Finds commits that were backported without previous
> patches in their original series also being backported. It accepts a range of
> git commits. I also added an option to filter the results by AUTOSEL only.
>
>For more information, see the README at
>https://git.kernel.org/pub/scm/linux/kernel/git/ebiggers/stable-tools.git/tree/README.md
Thanks! This is really great!
>Note: since it wasn't clear where to put these or how to integrate them into
>anything else, for now this is a completely standalone project. Perhaps the
>find-orig-patch functionality would be a nice feature for b4, but the more
>stable-kernel-maintenance-specific logic should be in a separate project. BTW,
>I used the same language and license as b4: Python and GPLv2+.
>
>I wrote some regression tests that show that find-missing-prereqs is able to
>detect the missing patches for a couple examples of missed backports that broke
>users, including the recent blk-cgroup one
>(https://lore.kernel.org/linux-block/CAOCAAm4reGhz400DSVrh0BetYD3Ljr2CZen7_3D4gXYYdB4SKQ@mail.gmail.com/T/#u).
>
>For another example, at the end of this email I've also pasted the output of
>'find-missing-prereqs v6.1.24..v6.1.26', which covers the last couple weeks of
>6.1. I don't immediately see anything super interesting in there, and it picked
>up a few very long patchsets which is a bit annoying (maybe very long patchsets
>generally aren't interesting and should be skipped?). But it's just an example.
The results here are interesting: the script points out a lot of
patchsets that are just sets of unrelated patches mashed together, I'll
comment more below on each of the findings.
I'm not sure the right thing to do would be to ignore long patchsets in
general, but rather learn to ignore sets that are declared to be
unrelated.
>Something that could be built on top of this is a script that applies the
>patches from the stable-queue repository for each stable kernel version, and
>generates a report about each one. BTW, I can work on these scripts more, but
>what I can't commit to doing is manually sending out reports every week... I
>hope that this can be automated and/or adopted by the stable maintainers.
Right, I'm going to incorporate it into my scripts - I don't think that
this is something a human should be running going forward, we have quite
a few bots :)
>Here's the output of 'find-missing-prereqs v6.1.24..v6.1.26':
>
>The following commit(s):
> [PATCH 24/33] commit 779fd2a575cc ("drm/amd/display: Pass the right info to drm_dp_remove_payload")
>... are backported without earlier commit(s) in series:
> [PATCH 1/33] commit de534c1cb031 ("drm/amd/display: Add height granularity limitation for dsc slice height calculation")
> [PATCH 2/33] commit aee0c07a74d3 ("drm/amd/display: Unify DC logging for BW Alloc")
> [PATCH 3/33] commit 67d198da2fd4 ("drm/amd/display: When blanking during init loop to find OPP index")
> [PATCH 4/33] commit c93aa7f33e94 ("drm/amd/display: 3.2.225")
> [PATCH 5/33] commit 0db13eae41fc ("drm/amd/display: Add minimum Z8 residency debug option")
> [PATCH 6/33] commit 0215ce9057ed ("drm/amd/display: Update minimum stutter residency for DCN314 Z8")
> [PATCH 7/33] commit c0a561d96a28 ("drm/amd/display: Drop CONFIG_DRM_AMD_DC_HDR")
> [PATCH 8/33] commit 11efe095dfe0 ("drm/amd/display: Fix no-DCN build")
> [PATCH 9/33] commit ab487ea8910d ("drm/amd/display: fix typo in dc_dsc_config_options structure")
> [PATCH 10/33] commit 1e88eb1b2c25 ("drm/amd/display: Drop CONFIG_DRM_AMD_DC_HDCP")
> [PATCH 11/33] commit efa4c4df864e ("drm/amd/display: call remove_stream_from_ctx from res_pool funcs")
> [PATCH 12/33] commit 84c03df58d8b ("drm/amd/display: Build DSC without DCN config")
> [PATCH 13/33] commit 36516001a7c9 ("drm/amd/display: move dc_link functions in accessories folder to dc_link_exports")
> [PATCH 14/33] commit 76f5dc40ebb1 ("drm/amd/display: move dc_link functions in link root folder to dc_link_exports")
> [PATCH 15/33] commit 6455cb522191 ("drm/amd/display: link link_dp_dpia_bw.o in makefile")
> [PATCH 16/33] commit 202a3816f37e ("drm/amd/display: move dc_link functions in protocols folder to dc_link_exports")
> [PATCH 17/33] commit 788c6e2ce5c7 ("drm/amd/display: replace all dc_link function call in link with link functions")
> [PATCH 18/33] commit 34fd6df78869 ("drm/amd/display: Simplify register offsets")
> [PATCH 19/33] commit 2b02d746c181 ("drm/amd/display: Keep PHY active for dp config")
> [PATCH 21/33] commit bf77fda02411 ("drm/amd/display: Drop unnecessary DCN guards")
> [PATCH 22/33] commit 4652ae7a51b7 ("drm/amd/display: Rename DCN config to FP")
> [PATCH 23/33] commit de930140bb57 ("drm/amd/display: Update to correct min FCLK when construction BB")
>Original patch series is "[PATCH 00/33] DC Patches Mar 6th, 2023"
>(https://lore.kernel.org/r/[email protected])
There seem to be weekly "DC Patches $DATE" patchsets composed of
unrelated patches, we could probably ignore those.
>The following commit(s):
> [PATCH 2/4] commit 3570f3cc4aab ("RDMA/erdma: Update default EQ depth to 4096 and max_send_wr to 8192")
> [PATCH 3/4] commit d682c9bc41fa ("RDMA/erdma: Inline mtt entries into WQE if supported")
> [PATCH 4/4] commit 132918e08e86 ("RDMA/erdma: Defer probing if netdevice can not be found")
>... are backported without earlier commit(s) in series:
> [PATCH 1/4] commit 3fe26c0493e4 ("RDMA/erdma: Fix some typos")
>Original patch series is "[PATCH for-rc 0/4] RDMA/erdma: erdma fixes 3-20-2023"
>(https://lore.kernel.org/r/[email protected])
Similar to above, unrelated patches and the one we skipped is just a
typo fix.
>The following commit(s):
> [PATCH 15/26] commit 361b02e68181 ("KVM: arm64: Initialise hypervisor copies of host symbols unconditionally")
>... are backported without earlier commit(s) in series:
> [PATCH 1/26] commit 0f4f7ae10ee4 ("KVM: arm64: Move hyp refcount manipulation helpers to common header file")
> [PATCH 2/26] commit 72a5bc0f153c ("KVM: arm64: Allow attaching of non-coalescable pages to a hyp pool")
> [PATCH 3/26] commit 8e6bcc3a4502 ("KVM: arm64: Back the hypervisor 'struct hyp_page' array for all memory")
> [PATCH 4/26] commit 0d16d12eb26e ("KVM: arm64: Fix-up hyp stage-1 refcounts for all pages mapped at EL2")
> [PATCH 5/26] commit 33bc332d4061 ("KVM: arm64: Unify identifiers used to distinguish host and hypervisor")
> [PATCH 6/26] commit 1ed5c24c26f4 ("KVM: arm64: Implement do_donate() helper for donating memory")
> [PATCH 7/26] commit 43c1ff8b7501 ("KVM: arm64: Prevent the donation of no-map pages")
> [PATCH 8/26] commit 9926cfce8dcb ("KVM: arm64: Add helpers to pin memory shared with the hypervisor at EL2")
> [PATCH 9/26] commit 4d968b12e6bb ("KVM: arm64: Include asm/kvm_mmu.h in nvhe/mem_protect.h")
> [PATCH 10/26] commit 1c80002e3264 ("KVM: arm64: Add hyp_spinlock_t static initializer")
> [PATCH 11/26] commit 5304002dc375 ("KVM: arm64: Rename 'host_kvm' to 'host_mmu'")
> [PATCH 12/26] commit a1ec5c70d3f6 ("KVM: arm64: Add infrastructure to create and track pKVM instances at EL2")
> [PATCH 13/26] commit 9d0c063a4d1d ("KVM: arm64: Instantiate pKVM hypervisor VM and vCPU structures from EL1")
> [PATCH 14/26] commit aa6948f82f0b ("KVM: arm64: Add per-cpu fixmap infrastructure at EL2")
>Original patch series is "[PATCH v6 00/26] KVM: arm64: Introduce pKVM hyp VM and vCPU state at EL2"
>(https://lore.kernel.org/r/[email protected])
This one is the most interesting of the bunch IMO: it's actually a long
patchset that enables new functionality, and we've picked a "random"
patch from the middle of it.
In this case this seems to be safe.
>The following commit(s):
> [PATCH 2/2] commit 2fcfd51add22 ("Bluetooth: SCO: Fix possible circular locking dependency sco_sock_getsockopt")
>... are backported without earlier commit(s) in series:
> [PATCH 1/2] commit 9a8ec9e8ebb5 ("Bluetooth: SCO: Fix possible circular locking dependency on sco_connect_cfm")
This script is a great way to identify missing fixes too :)
>The following commit(s):
> [PATCH 3/4] commit 5620eeb379d1 ("tracing: Add trace_array_puts() to write into instance")
>... are backported without earlier commit(s) in series:
> [PATCH 1/4] commit cb1f98c5e574 ("tracing: Add creation of instances at boot command line")
> [PATCH 2/4] commit c4846480831e ("tracing: Add enabling of events to boot instances")
>Original patch series is "[PATCH v2 0/4] tracing: Addition of tracing instances via kernel command line"
>(https://lore.kernel.org/r/[email protected])
Similar to the KVM one, but here we picked a patch that adds a new
function we used in the dependant commit. Looks okay in this case.
>The following commit(s):
> [PATCH 36/66] commit 4ac57c3fe2c0 ("drm/amd/display: set dcn315 lb bpp to 48")
>... are backported without earlier commit(s) in series:
> [PATCH 31/66] commit 0b5dfe12755f ("drm/amd/display: fix a divided-by-zero error")
> [PATCH 35/66] commit 1e994cc0956b ("drm/amd/display: limit timing for single dimm memory")
>Original patch series is "[PATCH 00/66] DC Patches Apr 17th, 2023"
>(https://lore.kernel.org/r/[email protected])
Same as the first finding: patchset of unrelated patches.
--
Thanks,
Sasha
On Sat, Apr 29, 2023 at 11:49:48PM -0700, Eric Biggers wrote:
> - `find-orig-patch`: Finds the original patch email from a git commit.
> It first checks for a matching "Message-Id:" or "Link:" from the git commit.
> If that fails, it falls back to a search of https://lore.kernel.org by
> commit title and uses some heuristics to try to find the right patch email.
You should also try matching by patch-id. E.g., taking the commit from the
example:
$ git show de534c1cb031 | git patch-id --stable
7d23fb78b4bb892155827a612631168cb28584cc de534c1cb0313a070f45938a53b53927cd34e5b9
Then we can search by patchid:7d23fb78b4bb892155827a612631168cb28584cc:
https://lore.kernel.org/all/?q=patchid%3A7d23fb78b4bb892155827a612631168cb28584cc
-K