Erhard reported boot fails on this AMD machine when using clang and bisected it
to a commit introducing a few static_call()s. Turns out that when using clang
with -Os it it very likely to generate conditional tail calls like:
0000000000000350 <amd_pmu_add_event>:
350: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1) 351: R_X86_64_NONE __fentry__-0x4
355: 48 83 bf 20 01 00 00 00 cmpq $0x0,0x120(%rdi)
35d: 0f 85 00 00 00 00 jne 363 <amd_pmu_add_event+0x13> 35f: R_X86_64_PLT32 __SCT__amd_pmu_branch_add-0x4
363: e9 00 00 00 00 jmp 368 <amd_pmu_add_event+0x18> 364: R_X86_64_PLT32 __x86_return_thunk-0x4
And our inline static_call() patching code can't deal with those and BUG
happens -- really early.
These patches borrow the kprobe Jcc emulation to implement text_poke_bp() Jcc
support, which is then used to teach inline static_call() about this form.
---
arch/x86/include/asm/text-patching.h | 31 ++++++++++++++++++
arch/x86/kernel/alternative.c | 62 +++++++++++++++++++++++++++---------
arch/x86/kernel/kprobes/core.c | 38 +++++-----------------
arch/x86/kernel/static_call.c | 50 +++++++++++++++++++++++++++--
4 files changed, 133 insertions(+), 48 deletions(-)
Hi Peter and Ingo,
On Mon, Jan 23, 2023 at 09:59:15PM +0100, Peter Zijlstra wrote:
> Erhard reported boot fails on this AMD machine when using clang and bisected it
> to a commit introducing a few static_call()s. Turns out that when using clang
> with -Os it it very likely to generate conditional tail calls like:
>
> 0000000000000350 <amd_pmu_add_event>:
> 350: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1) 351: R_X86_64_NONE __fentry__-0x4
> 355: 48 83 bf 20 01 00 00 00 cmpq $0x0,0x120(%rdi)
> 35d: 0f 85 00 00 00 00 jne 363 <amd_pmu_add_event+0x13> 35f: R_X86_64_PLT32 __SCT__amd_pmu_branch_add-0x4
> 363: e9 00 00 00 00 jmp 368 <amd_pmu_add_event+0x18> 364: R_X86_64_PLT32 __x86_return_thunk-0x4
>
> And our inline static_call() patching code can't deal with those and BUG
> happens -- really early.
>
> These patches borrow the kprobe Jcc emulation to implement text_poke_bp() Jcc
> support, which is then used to teach inline static_call() about this form.
>
> ---
> arch/x86/include/asm/text-patching.h | 31 ++++++++++++++++++
> arch/x86/kernel/alternative.c | 62 +++++++++++++++++++++++++++---------
> arch/x86/kernel/kprobes/core.c | 38 +++++-----------------
> arch/x86/kernel/static_call.c | 50 +++++++++++++++++++++++++++--
> 4 files changed, 133 insertions(+), 48 deletions(-)
I noticed this series was applied to x86/alternatives versus
x86/urgent, even though this appears to be a regression since 6.1, as
Erhard hit this issue in that tree.
Additionally, a new change in LLVM main [1] causes conditional tail
calls to be emitted even at -O2, so this breakage will become more
noticeable over time. Is it possible to expedite this to mainline so
that it can be backported to 6.1? If not, no worries, but I figured I
would ask :)
I have a backport of this series to 6.1 prepared already [2], where it
appears to work for me but I will get wider testing before sending it
after this is in Linus' tree (regardless of when that is). I figured it
would not hurt to have other eyes on it ahead of time though.
[1]: https://github.com/llvm/llvm-project/commit/ee5585ed09aff2e54cb540fad4c33f0c93626b1b
[2]: https://git.kernel.org/nathan/l/cbl-1800-1774-6.1
Cheers,
Nathan