git: a194eabe088b - stable/14 - arm64: Use store-pair to zero the kernel bss
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Mon, 02 Sep 2024 08:51:00 UTC
The branch stable/14 has been updated by andrew: URL: https://cgit.FreeBSD.org/src/commit/?id=a194eabe088bcc33ee07eefe3aace2aa3e2a3d25 commit a194eabe088bcc33ee07eefe3aace2aa3e2a3d25 Author: Andrew Turner <andrew@FreeBSD.org> AuthorDate: 2024-08-20 17:14:28 +0000 Commit: Andrew Turner <andrew@FreeBSD.org> CommitDate: 2024-09-02 08:49:38 +0000 arm64: Use store-pair to zero the kernel bss While this won't be noticed by most users the time to zero the bss while using instruction tracing in the Arm FVP models (simulators) is noticeable. Reduce this time by using a store-pair instruction to double the size of memory we zero on each iteration of the loop. Sponsored by: Arm Ltd Differential Revision: https://reviews.freebsd.org/D42733 (cherry picked from commit f1bc3750cf9a6623b0c0861984ef2a8ac966a4e3) --- sys/arm64/arm64/locore.S | 2 +- sys/conf/ldscript.arm64 | 3 ++- 2 files changed, 3 insertions(+), 2 deletions(-) diff --git a/sys/arm64/arm64/locore.S b/sys/arm64/arm64/locore.S index 3db1d3025967..d448f06b2580 100644 --- a/sys/arm64/arm64/locore.S +++ b/sys/arm64/arm64/locore.S @@ -109,7 +109,7 @@ virtdone: ldr x15, .Lbss ldr x14, .Lend 1: - str xzr, [x15], #8 + stp xzr, xzr, [x15], #16 cmp x15, x14 b.lo 1b diff --git a/sys/conf/ldscript.arm64 b/sys/conf/ldscript.arm64 index 185b4485e07a..21c7dc35d9b7 100644 --- a/sys/conf/ldscript.arm64 +++ b/sys/conf/ldscript.arm64 @@ -119,6 +119,7 @@ SECTIONS .sdata : { *(.sdata) } _edata = .; PROVIDE (edata = .); + . = ALIGN(16); __bss_start = .; .sbss : { *(.sbss) *(.scommon) } .bss : @@ -126,7 +127,7 @@ SECTIONS *(.dynbss) *(.bss) *(COMMON) - . = ALIGN(8); + . = ALIGN(16); __bss_end = .; /* A section for the initial page table, it doesn't need to be in the kernel file, however unlike normal .bss entries should not be zeroed