c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2026-06-01 19:54:10 +03:00
Code Issues Packages Projects Releases Wiki Activity

[icp] fpu and asm cleanup for linux

Browse Source 
Properly annotate functions and data section so that objtool does not complain
when CONFIG_STACK_VALIDATION and CONFIG_FRAME_POINTER are enabled.

Pass KERNELCPPFLAGS to assembler.

Use kfpu_begin()/kfpu_end() to protect SIMD regions in Linux kernel.

Reviewed-by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Gvozden Neskovic <neskovic@gmail.com>
Closes #5872 
Closes #5041
This commit is contained in:
Gvozden Neskovic
2017-03-07 21:59:31 +01:00
committed by Brian Behlendorf
parent 5fc73c46f9
commit 650383f283
12 changed files with 155 additions and 307 deletions
Download Patch File Download Diff File Expand all files Collapse all files
module/icp/asm-x86_64/modes/gcm_intel.S
+2 -86
View File
@@ -95,99 +95,18 @@ gcm_mul_pclmulqdq(uint64_t *x_in, uint64_t *y, uint64_t *res) {
#define _ASM
#include <sys/asm_linkage.h>
#ifdef _KERNEL
/*
* Note: the CLTS macro clobbers P2 (%rsi) under i86xpv. That is,
* it calls HYPERVISOR_fpu_taskswitch() which modifies %rsi when it
* uses it to pass P2 to syscall.
* This also occurs with the STTS macro, but we dont care if
* P2 (%rsi) is modified just before function exit.
* The CLTS and STTS macros push and pop P1 (%rdi) already.
*/
#ifdef __xpv
#define PROTECTED_CLTS \
push %rsi; \
CLTS; \
pop %rsi
#else
#define PROTECTED_CLTS \
CLTS
#endif /* __xpv */
/*
* If CR0_TS is not set, align stack (with push %rbp) and push
* %xmm0 - %xmm10 on stack, otherwise clear CR0_TS
*/
#define CLEAR_TS_OR_PUSH_XMM_REGISTERS(tmpreg) \
push %rbp; \
mov %rsp, %rbp; \
movq %cr0, tmpreg; \
testq $CR0_TS, tmpreg; \
jnz 1f; \
and $-XMM_ALIGN, %rsp; \
sub $[XMM_SIZE * 11], %rsp; \
movaps %xmm0, 160(%rsp); \
movaps %xmm1, 144(%rsp); \
movaps %xmm2, 128(%rsp); \
movaps %xmm3, 112(%rsp); \
movaps %xmm4, 96(%rsp); \
movaps %xmm5, 80(%rsp); \
movaps %xmm6, 64(%rsp); \
movaps %xmm7, 48(%rsp); \
movaps %xmm8, 32(%rsp); \
movaps %xmm9, 16(%rsp); \
movaps %xmm10, (%rsp); \
jmp 2f; \
1: \
PROTECTED_CLTS; \
2:
/*
* If CR0_TS was not set above, pop %xmm0 - %xmm10 off stack,
* otherwise set CR0_TS.
*/
#define SET_TS_OR_POP_XMM_REGISTERS(tmpreg) \
testq $CR0_TS, tmpreg; \
jnz 1f; \
movaps (%rsp), %xmm10; \
movaps 16(%rsp), %xmm9; \
movaps 32(%rsp), %xmm8; \
movaps 48(%rsp), %xmm7; \
movaps 64(%rsp), %xmm6; \
movaps 80(%rsp), %xmm5; \
movaps 96(%rsp), %xmm4; \
movaps 112(%rsp), %xmm3; \
movaps 128(%rsp), %xmm2; \
movaps 144(%rsp), %xmm1; \
movaps 160(%rsp), %xmm0; \
jmp 2f; \
1: \
STTS(tmpreg); \
2: \
mov %rbp, %rsp; \
pop %rbp
#else
#define PROTECTED_CLTS
#define CLEAR_TS_OR_PUSH_XMM_REGISTERS(tmpreg)
#define SET_TS_OR_POP_XMM_REGISTERS(tmpreg)
#endif /* _KERNEL */
/*
* Use this mask to byte-swap a 16-byte integer with the pshufb instruction
*/
// static uint8_t byte_swap16_mask[] = {
// 15, 14, 13, 12, 11, 10, 9, 8, 7, 6 ,5, 4, 3, 2, 1, 0 };
.text
.data
.align XMM_ALIGN
.Lbyte_swap16_mask:
.byte 15, 14, 13, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0
/*
* void gcm_mul_pclmulqdq(uint64_t *x_in, uint64_t *y, uint64_t *res);
*
@@ -221,8 +140,6 @@ gcm_mul_pclmulqdq(uint64_t *x_in, uint64_t *y, uint64_t *res) {
*/
ENTRY_NP(gcm_mul_pclmulqdq)
CLEAR_TS_OR_PUSH_XMM_REGISTERS(%r10)
//
// Copy Parameters
//
@@ -325,9 +242,8 @@ ENTRY_NP(gcm_mul_pclmulqdq)
//
// Cleanup and Return
// Return
//
SET_TS_OR_POP_XMM_REGISTERS(%r10)
ret
SET_SIZE(gcm_mul_pclmulqdq)