Build breakage due to incomplete addition of "gmb()" barrier |
||||
Issue descriptionError log: kernel/bpf/core.c: In function ‘__bpf_prog_run’: kernel/bpf/core.c:432:3: error: implicit declaration of function ‘gmb’ Affects test builds for ia64, metag, mips, powerpc, s390, sparc64, um, the virtio tool, and possibly others. Brought in as part of the fixes for CVE-2017-5753, but the patch series itself is not public nor, as it looks like, published (though integrated into Suse, Ubuntu, and other distribution kernels).
,
Feb 9 2018
,
Feb 9 2018
Interesting, I didn't realize we built our kernel for architectures we don't ship on. Want me to revert the CLs or just add the do {} while (0) place holder definition for those architectures?
,
Feb 9 2018
If I have a basket full of fruits, but only sell the apples, I tend to check all other fruit for bad ones anyway. Yes, I understand that there are people who argue that we should only check for bad apples, but I strongly believe that this would result in all fruit being bad at some point. Of course, not testing unrelated builds would prevent us from doing stable tree merges, since those would be impossible to test. It would also render 0day useless (which tests _all_ architectures), thus preventing us from being embarrassed by all the build failures visible to the public. Would this be good ? Those who are still opposed to stable tree merges might agree. As for what you want to do, your call. Again, I was not involved in the decision process, and no one explained to me why we decided to deviate from upstream, so I not in the position to make an informed decision either way.
,
Feb 14 2018
The following revision refers to this bug: https://chromium.googlesource.com/chromiumos/third_party/kernel/+/fadbcf53ec10906cc53076ce773ea1a926c7d785 commit fadbcf53ec10906cc53076ce773ea1a926c7d785 Author: Greg Kerr <kerrnel@chromium.org> Date: Wed Feb 14 05:17:11 2018 Revert "CHROMIUM: locking/barriers: introduce new memory barrier gmb()" This reverts the following commits: bae7a53cef9844107ca5f725947e5d9fc32d27dc ba8e989ba823ddf371a4123597cfc4e62442a71d 217b5f5e8605f0d50d28a0420f0603289f521c28 510fd35e03bcbb70921ca878e21170dac913fa07 Reason for revert: These patches break the test kernel builds on various architectures ( crbug.com/810789 ). The patches titles are: "FROMGIT: userns: add barrier to prevent bounds check bypass" "FROMGIT: x86, bpf, jit: add barrier to prevent bounds check bypass when JIT is enabled" "FROMGIT: bpf: add barrier to prevent bounds check bypass in eBPF interpreter" "CHROMIUM: locking/barriers: introduce new memory barrier gmb()" Bug: chromium:809607 , chromium:810789 Change-Id: I29511ea856cb2da12aab4fa985a9dff4d2c96e70 Reviewed-on: https://chromium-review.googlesource.com/912154 Commit-Ready: Greg Kerr <kerrnel@chromium.org> Tested-by: Greg Kerr <kerrnel@chromium.org> Reviewed-by: Guenter Roeck <groeck@chromium.org> [modify] https://crrev.com/fadbcf53ec10906cc53076ce773ea1a926c7d785/kernel/bpf/core.c [modify] https://crrev.com/fadbcf53ec10906cc53076ce773ea1a926c7d785/kernel/user_namespace.c [modify] https://crrev.com/fadbcf53ec10906cc53076ce773ea1a926c7d785/arch/arm64/include/asm/barrier.h [modify] https://crrev.com/fadbcf53ec10906cc53076ce773ea1a926c7d785/arch/x86/net/bpf_jit_comp.c [modify] https://crrev.com/fadbcf53ec10906cc53076ce773ea1a926c7d785/arch/arm/include/asm/barrier.h [modify] https://crrev.com/fadbcf53ec10906cc53076ce773ea1a926c7d785/include/asm-generic/barrier.h [modify] https://crrev.com/fadbcf53ec10906cc53076ce773ea1a926c7d785/arch/x86/include/asm/barrier.h
,
Feb 14 2018
|
||||
►
Sign in to add a comment |
||||
Comment 1 by groeck@chromium.org
, Feb 9 2018