ceph/0001-src-common-crc32c_intel_fast.patch
Kaleb S. KEITHLEY 5c46ad7f7e ceph 15.2.2, CET enable src/common/crc32c_intel_*_asm.s; shstk, ibt
and other fixes
see https://github.com/intel/isa-l/blob/master/crc/crc32_iscsi_00.asm

Signed-off-by: Kaleb S. KEITHLEY <kkeithle@redhat.com>
2020-05-26 11:15:55 -04:00

301 lines
10 KiB
Diff

--- ceph-15.2.2/src/yasm-wrapper.orig 2020-03-09 12:57:37.603837466 -0400
+++ ceph-15.2.2/src/yasm-wrapper 2020-03-09 13:02:31.496796609 -0400
@@ -1,10 +1,11 @@
-#!/bin/sh -e
+#!/bin/sh
# libtool and yasm do not get along.
# filter out any crap that libtool feeds us that yasm does not understand.
#echo $0: got $*
new=""
touch=""
+object=""
while [ -n "$*" ]; do
case "$1" in
-f )
@@ -29,6 +30,12 @@
touch="$1"
shift
;;
+ -o )
+ shift
+ object="$1"
+ new="$new -o $1"
+ shift
+ ;;
* )
new="$new $1"
shift
@@ -39,6 +46,25 @@
#echo $0: yasm $new
yasm $new
+INDEXOFNOTEGNUPROPLINE=$(readelf -S -W ${object} | grep ".note.gnu.property")
+
+if [ -n "${INDEXOFNOTEGNUPROPLINE}" ]; then
+
+ STARTOFSECTHDRLINE=$(readelf -h ${object} | grep "Start of section headers:")
+ SIZEOFSECTHDRLINE=$(readelf -h ${object} | grep "Size of section headers:")
+
+ STARTOFSECTHDR=$(echo ${STARTOFSECTHDRLINE} | cut -d ' ' -f 5)
+ SIZEOFSECTHDR=$(echo ${SIZEOFSECTHDRLINE} | cut -d ' ' -f 5)
+ INDEXOFNOTEGNUPROPFRAG=$(echo ${INDEXOFNOTEGNUPROPLINE} | cut -d '.' -f 1)
+
+ INDEXOFNOTEGNUPROP=$(echo ${INDEXOFNOTEGNUPROPFRAG} | grep -Eo "[[:digit:]]+")
+
+ OFFSET=$(( ${STARTOFSECTHDR} + 4 + ( ${SIZEOFSECTHDR} * ${INDEXOFNOTEGNUPROP} ) ))
+
+ printf '\x07' | dd of=${object} bs=1 seek=${OFFSET} count=1 conv=notrunc &> /dev/null
+
+fi
+
[ -n "$touch" ] && touch $touch
true
--- ceph-15.2.2/src/common/crc32c_intel_fast_asm.s.orig 2020-05-26 08:34:32.226201974 -0400
+++ ceph-15.2.2/src/common/crc32c_intel_fast_asm.s 2020-05-26 09:25:55.528201974 -0400
@@ -1,5 +1,5 @@
;
-; Copyright 2012-2013 Intel Corporation All Rights Reserved.
+; Copyright 2012-2015 Intel Corporation All Rights Reserved.
; All rights reserved.
;
; http://opensource.org/licenses/BSD-3-Clause
@@ -59,16 +59,34 @@
xor rbx, rbx ;; rbx = crc1 = 0;
xor r10, r10 ;; r10 = crc2 = 0;
+ cmp len, %%bSize*3*2
+ jbe %%non_prefetch
+
%assign i 0
%rep %%bSize/8 - 1
- crc32 rax, [bufptmp+i + 0*%%bSize] ;; update crc0
- crc32 rbx, [bufptmp+i + 1*%%bSize] ;; update crc1
- crc32 r10, [bufptmp+i + 2*%%bSize] ;; update crc2
+ %if i < %%bSize*3/4
+ prefetchnta [bufptmp+ %%bSize*3 + i*4]
+ %endif
+ crc32 rax, qword [bufptmp+i + 0*%%bSize] ;; update crc0
+ crc32 rbx, qword [bufptmp+i + 1*%%bSize] ;; update crc1
+ crc32 r10, qword [bufptmp+i + 2*%%bSize] ;; update crc2
%assign i (i+8)
%endrep
- crc32 rax, [bufptmp+i + 0*%%bSize] ;; update crc0
- crc32 rbx, [bufptmp+i + 1*%%bSize] ;; update crc1
-; SKIP ;crc32 r10, [bufptmp+i + 2*%%bSize] ;; update crc2
+ jmp %%next %+ %1
+
+%%non_prefetch:
+ %assign i 0
+ %rep %%bSize/8 - 1
+ crc32 rax, qword [bufptmp+i + 0*%%bSize] ;; update crc0
+ crc32 rbx, qword [bufptmp+i + 1*%%bSize] ;; update crc1
+ crc32 r10, qword [bufptmp+i + 2*%%bSize] ;; update crc2
+ %assign i (i+8)
+ %endrep
+
+%%next %+ %1:
+ crc32 rax, qword [bufptmp+i + 0*%%bSize] ;; update crc0
+ crc32 rbx, qword [bufptmp+i + 1*%%bSize] ;; update crc1
+; SKIP ;crc32 r10, qword [bufptmp+i + 2*%%bSize] ;; update crc2
; merge in crc0
movzx bufp_dw, al
@@ -180,12 +198,15 @@
%define crc_init_dw r8d
%endif
-
+ endbranch
push rdi
push rbx
mov rax, crc_init ;; rax = crc_init;
+ cmp len, 8
+ jb less_than_8
+
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; 1) ALIGN: ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
@@ -195,9 +216,6 @@
;; amount of the address
je proc_block ;; Skip if aligned
- cmp len, 8
- jb less_than_8
-
;;;; Calculate CRC of unaligned bytes of the buffer (if any) ;;;;
mov rbx, [bufptmp] ;; load a quadword from the buffer
add bufptmp, bufp ;; align buffer pointer for
@@ -217,7 +235,7 @@
cmp len, 240
jb bit8
- lea crc_init, [mul_table_72 wrt rip] ;; load table base address
+ lea crc_init, [mul_table_72] ;; load table base address
crcB3 640, 0x1000, 0x0c00 ; 640*3 = 1920 (Tables 1280, 640)
crcB3 320, 0x0c00, 0x0800 ; 320*3 = 960 (Tables 640, 320)
@@ -233,7 +251,7 @@
jnc bit7 ;; jump to bit-6 if bit-7 == 0
%assign i 0
%rep 16
- crc32 rax, [bufptmp+i] ;; compute crc32 of 8-byte data
+ crc32 rax, qword [bufptmp+i] ;; compute crc32 of 8-byte data
%assign i (i+8)
%endrep
je do_return ;; return if remaining data is zero
@@ -244,7 +262,7 @@
jnc bit6 ;; jump to bit-6 if bit-7 == 0
%assign i 0
%rep 8
- crc32 rax, [bufptmp+i] ;; compute crc32 of 8-byte data
+ crc32 rax, qword [bufptmp+i] ;; compute crc32 of 8-byte data
%assign i (i+8)
%endrep
je do_return ;; return if remaining data is zero
@@ -254,7 +272,7 @@
jnc bit5 ;; jump to bit-5 if bit-6 == 0
%assign i 0
%rep 4
- crc32 rax, [bufptmp+i] ;; compute crc32 of 8-byte data
+ crc32 rax, qword [bufptmp+i] ;; compute crc32 of 8-byte data
%assign i (i+8)
%endrep
je do_return ;; return if remaining data is zero
@@ -264,7 +282,7 @@
jnc bit4 ;; jump to bit-4 if bit-5 == 0
%assign i 0
%rep 2
- crc32 rax, [bufptmp+i] ;; compute crc32 of 8-byte data
+ crc32 rax, qword [bufptmp+i] ;; compute crc32 of 8-byte data
%assign i (i+8)
%endrep
je do_return ;; return if remaining data is zero
@@ -272,11 +290,11 @@
bit4:
shl len_b, 1 ;; shift-out MSB (bit-4)
jnc bit3 ;; jump to bit-3 if bit-4 == 0
- crc32 rax, [bufptmp] ;; compute crc32 of 8-byte data
+ crc32 rax, qword [bufptmp] ;; compute crc32 of 8-byte data
je do_return ;; return if remaining data is zero
add bufptmp, 8 ;; buf +=8; (next 8 bytes)
bit3:
- mov rbx, [bufptmp] ;; load a 8-bytes from the buffer:
+ mov rbx, qword [bufptmp] ;; load a 8-bytes from the buffer:
shl len_b, 1 ;; shift-out MSB (bit-3)
jnc bit2 ;; jump to bit-2 if bit-3 == 0
crc32 eax, ebx ;; compute crc32 of 4-byte data
@@ -672,3 +690,22 @@
; inform linker that this doesn't require executable stack
section .note.GNU-stack noalloc noexec nowrite progbits
%endif
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+SECTION .note.gnu.property align=8 noexec ; section number 22, const
+
+ db 04H, 00H, 00H, 00H, 20H, 00H, 00H, 00H ; 0000 _ .... ...
+ db 05H, 00H, 00H, 00H, 47H, 4EH, 55H, 00H ; 0008 _ ....GNU.
+ db 00H, 00H, 00H, 0C0H, 04H, 00H, 00H, 00H ; 0010 _ ........
+ db 00H, 00H, 00H, 00H, 00H, 00H, 00H, 00H ; 0018 _ ........
+ db 01H, 00H, 00H, 0C0H, 04H, 00H, 00H, 00H ; 0020 _ ........
+ db 00H, 00H, 00H, 00H, 00H, 00H, 00H, 00H ; 0028 _ ........
+ db 04H, 00H, 00H, 00H, 10H, 00H, 00H, 00H ; 0030 _ ........
+ db 05H, 00H, 00H, 00H, 47H, 4EH, 55H, 00H ; 0038 _ ....GNU.
+ db 02H, 00H, 00H, 0C0H, 04H, 00H, 00H, 00H ; 0040 _ ........
+ db 03H, 00H, 00H, 00H, 00H, 00H, 00H, 00H ; 0048 _ ........
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
--- ceph-15.2.2/src/common/crc32c_intel_fast_zero_asm.s.orig 2020-05-26 08:34:32.226201974 -0400
+++ ceph-15.2.2/src/common/crc32c_intel_fast_zero_asm.s 2020-05-26 09:19:55.571201974 -0400
@@ -1,5 +1,5 @@
;
-; Copyright 2012-2013 Intel Corporation All Rights Reserved.
+; Copyright 2012-2015 Intel Corporation All Rights Reserved.
; All rights reserved.
;
; http://opensource.org/licenses/BSD-3-Clause
@@ -59,6 +59,19 @@
xor rbx, rbx ;; rbx = crc1 = 0;
xor r10, r10 ;; r10 = crc2 = 0;
+ cmp len, %%bSize*3*2
+ jbe %%non_prefetch
+
+ %assign i 0
+ %rep %%bSize/8 - 1
+ crc32 rax, bufptmp ;; update crc0
+ crc32 rbx, bufptmp ;; update crc1
+ crc32 r10, bufptmp ;; update crc2
+ %assign i (i+8)
+ %endrep
+ jmp %%next %+ %1
+
+%%non_prefetch:
%assign i 0
%rep %%bSize/8 - 1
crc32 rax, bufptmp ;; update crc0
@@ -66,6 +79,8 @@
crc32 r10, bufptmp ;; update crc2
%assign i (i+8)
%endrep
+
+%%next %+ %1:
crc32 rax, bufptmp ;; update crc0
crc32 rbx, bufptmp ;; update crc1
; SKIP ;crc32 r10, bufptmp ;; update crc2
@@ -180,12 +195,15 @@
%define crc_init_dw r8d
%endif
-
+ endbranch
push rdi
push rbx
mov rax, crc_init ;; rax = crc_init;
+ cmp len, 8
+ jb less_than_8
+
;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
;; 1) ALIGN: ;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
@@ -199,7 +217,7 @@
cmp len, 240
jb bit8
- lea crc_init, [mul_table_72 wrt rip] ;; load table base address
+ lea crc_init, [mul_table_72] ;; load table base address
crcB3 640, 0x1000, 0x0c00 ; 640*3 = 1920 (Tables 1280, 640)
crcB3 320, 0x0c00, 0x0800 ; 320*3 = 960 (Tables 640, 320)
@@ -654,3 +672,22 @@
; inform linker that this doesn't require executable stack
section .note.GNU-stack noalloc noexec nowrite progbits
%endif
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+
+SECTION .note.gnu.property align=8 noexec ; section number 22, const
+
+ db 04H, 00H, 00H, 00H, 20H, 00H, 00H, 00H ; 0000 _ .... ...
+ db 05H, 00H, 00H, 00H, 47H, 4EH, 55H, 00H ; 0008 _ ....GNU.
+ db 00H, 00H, 00H, 0C0H, 04H, 00H, 00H, 00H ; 0010 _ ........
+ db 00H, 00H, 00H, 00H, 00H, 00H, 00H, 00H ; 0018 _ ........
+ db 01H, 00H, 00H, 0C0H, 04H, 00H, 00H, 00H ; 0020 _ ........
+ db 00H, 00H, 00H, 00H, 00H, 00H, 00H, 00H ; 0028 _ ........
+ db 04H, 00H, 00H, 00H, 10H, 00H, 00H, 00H ; 0030 _ ........
+ db 05H, 00H, 00H, 00H, 47H, 4EH, 55H, 00H ; 0038 _ ....GNU.
+ db 02H, 00H, 00H, 0C0H, 04H, 00H, 00H, 00H ; 0040 _ ........
+ db 03H, 00H, 00H, 00H, 00H, 00H, 00H, 00H ; 0048 _ ........
+
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;
+;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;