This series supersedes a prior RFC [1]. This series refactors the arm64 signal handling implementation in order to make it easier to add support for new CPU architecture extensions, which may require new records in the signal frame and may require the signal frame to grow beyond its current size. This series provides infrastructure that will be needed in order to support the Scalable Vector Extension (SVE) [2] (patches to be posted separately). LTP testing is ongoing -- I will ping this series when I have results. Changes since RFC v2: * The frame link record has been moved back to the end of the full signal frame, instead of fixing its position at the end of __reserved[], after Catalin raised concerns about the space otherwise wasted by putting the extended signal frame data after it. Previous versions of the series had this behaviour, but concerns were expressed by reviewers about the potential ABI impact. This change has been informally tested for ABI impacts using the libgcc unwinder (exercised by throwing exceptions from a signal handler with g++ -fnon-call-exceptions) and gdb backtracing, neither of which showed any problem. Review of the libgcc and libunwind implementations and discussion with gdb developers suggest that no breakage is expected as a result of the change. (See patch 1 of this posting.) * Patches 4 (Allocate extra sigcontext space as needed) and 5 (Parse extra_context during sigreturn) of RFC v2 have been merged, since they are not fully bisectable and make more sense as a single patch. * Patch 6 (Report signal frame size to userspace via auxv) of RFC v2 has been dropped from the series, because this creates ABI that we don't need yet, and can benefit from further discussion. This patch will become relevant when merging support for SVE. * struct extra_context is explicitly padded up to a multiple of 16 bytes. This fixes a bug whereby a non-extra_context-aware unwinder would skip the wrong amount of data after extra_context and miss the terminator record. There is a de facto assumption that unwinders do not need to round _aarch64_ctx.size up to a multiple of 16 when stepping through the records in the signal frame, though this isn't documented anywhere. Original blurb: An architecture advertises the maximum possible signal frame size via the MINSIGSTKSZ #define (mandated by POSIX). However, CPU architecture extensions may increase the amount of space required to store the interrupted context when a signal is delivered. Eventually the amount of space needed in the signal frame may exceed MINSIGSTKSZ -- whether and when this happens is largely a matter of luck, depending on the initial guess for MINSIGSTKSZ and the evolution of that particular CPU architecture. Unfortunately MINSIGSTKSZ cannot be changed without an ABI break, and POSIX provides no mechanism for migration. arm64 initially reserved 4KB of space in the signal frame for extensions, of which about 0.5KB is allocated to the FP/SIMD registers initially. Depending on the vector length supported by the hardware, SVE requires up to around 8KB of space to store the full SIMD register context, which is too large to fit in the existing frame. This series adds a mechanism for optionally enlarging the signal frame (patches 4-5) and reporting the actual maximum signal frame size to userspace (patch 6). Patches 1-3 do some refactoring to support this change by abstracting the way signal frame records are allocated onto the user stack. Full backwards compatibility is not possible -- there is no way to hide the fact that the signal frame has grown -- so it is expected that support for new architecture extensions that can cause the signal frame to grow will be opt-in for userspace, in addition to using the extension mechanism defined by this series. [1] [RFC PATCH v2 0/6] Signal frame expansion support http://lists.infradead.org/pipermail/linux-arm-kernel/2017-April/501163.html [2] ARM Scalable Vector Extension https://community.arm.com/groups/processors/blog/2016/08/22/technology-update-the-scalable-vector-extension-sve-for-the-armv8-a-architecture https://developer.arm.com/docs/ddi0584/latest/arm-architecture-reference-manual-supplement-the-scalable-vector-extension-sve-for-armv8-a Dave Martin (5): arm64: signal: split frame link record from sigcontext structure arm64: signal: Refactor sigcontext parsing in rt_sigreturn arm64: signal: factor frame layout and population into separate passes arm64: signal: factor out signal frame record allocation arm64: signal: Allow expansion of the signal frame arch/arm64/include/uapi/asm/sigcontext.h | 55 +++++ arch/arm64/kernel/signal.c | 408 ++++++++++++++++++++++++++++--- 2 files changed, 427 insertions(+), 36 deletions(-) -- 2.1.4
