Messages in this thread Patch in this message | | | From | "H. Peter Anvin (Intel)" <> | Subject | [PATCH v3 2/2] x86/asm: avoid adding register pressure for the init case in static_cpu_has() | Date | Fri, 10 Sep 2021 12:59:10 -0700 |
| |
From: "H. Peter Anvin" <hpa@zytor.com>
gcc will sometimes manifest the address of boot_cpu_data in a register as part of constant propagation. When multiple static_cpu_has() are used this may foul the mainline code with a register load which will only be used on the fallback path, which is unused after initialization.
Explicitly force gcc to use immediate (rip-relative) addressing for the fallback path, thus removing any possible register use from static_cpu_has().
While making changes, modernize the code to use .pushsection...popsection instead of .section...previous.
Signed-off-by: H. Peter Anvin (Intel) <hpa@zytor.com> --- arch/x86/include/asm/cpufeature.h | 13 +++++++++---- 1 file changed, 9 insertions(+), 4 deletions(-)
diff --git a/arch/x86/include/asm/cpufeature.h b/arch/x86/include/asm/cpufeature.h index 16a51e7288d5..1261842d006c 100644 --- a/arch/x86/include/asm/cpufeature.h +++ b/arch/x86/include/asm/cpufeature.h @@ -173,20 +173,25 @@ extern void clear_cpu_cap(struct cpuinfo_x86 *c, unsigned int bit); * means that the boot_cpu_has() variant is already fast enough for the * majority of cases and you should stick to using it as it is generally * only two instructions: a RIP-relative MOV and a TEST. + * + * Do not use an "m" constraint for [cap_byte] here: gcc doesn't know + * that this is only used on a fallback path and will sometimes cause + * it to manifest the address of boot_cpu_data in a register, fouling + * the mainline (post-initialization) code. */ static __always_inline bool _static_cpu_has(u16 bit) { asm_volatile_goto( ALTERNATIVE_TERNARY("jmp 6f", %P[feature], "", "jmp %l[t_no]") - ".section .altinstr_aux,\"ax\"\n" + ".pushsection .altinstr_aux,\"ax\"\n" "6:\n" - " testb %[bitnum],%[cap_byte]\n" + " testb %[bitnum]," _ASM_RIP(%P[cap_byte]) "\n" " jnz %l[t_yes]\n" " jmp %l[t_no]\n" - ".previous\n" + ".popsection\n" : : [feature] "i" (bit), [bitnum] "i" (1 << (bit & 7)), - [cap_byte] "m" (((const char *)boot_cpu_data.x86_capability)[bit >> 3]) + [cap_byte] "i" (&((const char *)boot_cpu_data.x86_capability)[bit >> 3]) : : t_yes, t_no); t_yes: return true; -- 2.31.1
| |