Fix SIGILL on arm64 when HWCAP_CPUID is not set

Origin: upstream, https://github.com/xianyi/OpenBLAS/commit/6fe0f1fab9d6a7f46d71d37ebb210fbf56924fbc
Bug-Debian: https://bugs.debian.org/986996
Last-Update: 2021-04-18

This is a crashing bug (SIGILL) that also affects numpy on arm64. One
line of processors affected are NVIDIA Tegra (Jetson devices).

On ARM64, openblas uses feature registers to detect the detailed
processor arch. It queries HWCAP_CPUID to check if the feature registers
are used. But due to a missing volatile declaration, gcc seems to break
this guarding in optimization.
Last-Update: 2021-04-18
Gbp-Pq: Name fix-arm64-sigill.patch

diff --git a/driver/others/dynamic_arm64.c b/driver/others/dynamic_arm64.c
index 4f1b12f27a83a253afec182abf4cc289e86dcc63..37c0694b6f76f497f80744841a7c81cddee6740e 100644 (file)
--- a/driver/others/dynamic_arm64.c
+++ b/driver/others/dynamic_arm64.c
@@ -68,7 +68,7 @@ extern void openblas_warning(int verbose, const char * msg);
 #endif
 
 #define get_cpu_ftr(id, var) ({                                        \
-               __asm__("mrs %0, "#id : "=r" (var));            \
+               __asm__ __volatile__("mrs %0, "#id : "=r" (var));               \
        })
 
 static char *corename[] = {