lib/vdso: Force inlining of __cvdso_clock_getres_common()
With CONFIG_CC_OPTIMIZE_FOR_SIZE=y, gcc may decide not to inline
__cvdso_clock_getres_common(). This introduces unnecessary stack
maintenance in the vDSO fastpath.
Furthermore, with automatic stack variable initialization
(CONFIG_INIT_STACK_ALL_ZERO or CONFIG_INIT_STACK_ALL_PATTERN)
gcc can emit a call to memset() which is not valid in the vDSO.
Mark the __cvdso_clock_getres_common() as __always_inline to avoid
both issues.
Paradoxically the inlining even reduces the size of the code:
$ ./scripts/bloat-o-meter arch/powerpc/kernel/vdso/vgettimeofday-32.o.before arch/powerpc/kernel/vdso/vgettimeofday-32.o.after
add/remove: 0/1 grow/shrink: 1/1 up/down: 52/-148 (-96)
Function old new delta
__c_kernel_clock_getres_time64 92 144 +52
__c_kernel_clock_getres 136 132 -4
__cvdso_clock_getres_common 144 - -144
Total: Before=2788, After=2692, chg -3.44%
With CONFIG_CC_OPTIMIZE_FOR_PERFORMANCE=y the functions are always
inlined and therefore the behaviour stays the same.
See also the equivalent change for clock_gettime() in commit b91c8c42ffdd
("lib/vdso: Force inlining of __cvdso_clock_gettime_common()").
Reported-by: Sverdlin, Alexander <alexander.sverdlin@siemens.com>
Closes: https://lore.kernel.org/lkml/f45316f65a46da638b3c6aa69effd8980e6677b9.camel@siemens.com/
Suggested-by: Christophe Leroy (CS GROUP) <chleroy@kernel.org>
Link: https://lore.kernel.org/lkml/230c749f-ebd6-4829-93ee-601d88000a45@kernel.org/
Fixes: 21bbfd74044f ("x86/vdso: Provide clock_getres_time64() for x86-32")
Fixes: 1149dcdfc9ef ("ARM: VDSO: Provide clock_getres_time64()")
Fixes: f10c2e72b5de ("arm64: vdso32: Provide clock_getres_time64()")
Fixes: bec06cd6a140 ("MIPS: vdso: Provide getres_time64() for 32-bit ABIs")
Fixes: 759a1f97373f ("powerpc/vdso: Provide clock_getres_time64()")
Signed-off-by: Thomas Weißschuh <thomas.weissschuh@linutronix.de>
1 file changed
