| 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
|---|---|
| 2 | /* Written 2003 by Andi Kleen, based on a kernel by Evandro Menezes */ |
| 3 | |
| 4 | #include <linux/export.h> |
| 5 | #include <linux/linkage.h> |
| 6 | #include <linux/cfi_types.h> |
| 7 | #include <asm/cpufeatures.h> |
| 8 | #include <asm/alternative.h> |
| 9 | |
| 10 | /* |
| 11 | * Some CPUs run faster using the string copy instructions (sane microcode). |
| 12 | * It is also a lot simpler. Use this when possible. But, don't use streaming |
| 13 | * copy unless the CPU indicates X86_FEATURE_REP_GOOD. Could vary the |
| 14 | * prefetch distance based on SMP/UP. |
| 15 | */ |
| 16 | ALIGN |
| 17 | SYM_TYPED_FUNC_START(copy_page) |
| 18 | ALTERNATIVE "jmp copy_page_regs", "", X86_FEATURE_REP_GOOD |
| 19 | movl $4096/8, %ecx |
| 20 | rep movsq |
| 21 | RET |
| 22 | SYM_FUNC_END(copy_page) |
| 23 | EXPORT_SYMBOL(copy_page) |
| 24 | |
| 25 | SYM_FUNC_START_LOCAL(copy_page_regs) |
| 26 | subq $2*8, %rsp |
| 27 | movq %rbx, (%rsp) |
| 28 | movq %r12, 1*8(%rsp) |
| 29 | |
| 30 | movl $(4096/64)-5, %ecx |
| 31 | .p2align 4 |
| 32 | .Loop64: |
| 33 | dec %rcx |
| 34 | movq 0x8*0(%rsi), %rax |
| 35 | movq 0x8*1(%rsi), %rbx |
| 36 | movq 0x8*2(%rsi), %rdx |
| 37 | movq 0x8*3(%rsi), %r8 |
| 38 | movq 0x8*4(%rsi), %r9 |
| 39 | movq 0x8*5(%rsi), %r10 |
| 40 | movq 0x8*6(%rsi), %r11 |
| 41 | movq 0x8*7(%rsi), %r12 |
| 42 | |
| 43 | prefetcht0 5*64(%rsi) |
| 44 | |
| 45 | movq %rax, 0x8*0(%rdi) |
| 46 | movq %rbx, 0x8*1(%rdi) |
| 47 | movq %rdx, 0x8*2(%rdi) |
| 48 | movq %r8, 0x8*3(%rdi) |
| 49 | movq %r9, 0x8*4(%rdi) |
| 50 | movq %r10, 0x8*5(%rdi) |
| 51 | movq %r11, 0x8*6(%rdi) |
| 52 | movq %r12, 0x8*7(%rdi) |
| 53 | |
| 54 | leaq 64 (%rsi), %rsi |
| 55 | leaq 64 (%rdi), %rdi |
| 56 | |
| 57 | jnz .Loop64 |
| 58 | |
| 59 | movl $5, %ecx |
| 60 | .p2align 4 |
| 61 | .Loop2: |
| 62 | decl %ecx |
| 63 | |
| 64 | movq 0x8*0(%rsi), %rax |
| 65 | movq 0x8*1(%rsi), %rbx |
| 66 | movq 0x8*2(%rsi), %rdx |
| 67 | movq 0x8*3(%rsi), %r8 |
| 68 | movq 0x8*4(%rsi), %r9 |
| 69 | movq 0x8*5(%rsi), %r10 |
| 70 | movq 0x8*6(%rsi), %r11 |
| 71 | movq 0x8*7(%rsi), %r12 |
| 72 | |
| 73 | movq %rax, 0x8*0(%rdi) |
| 74 | movq %rbx, 0x8*1(%rdi) |
| 75 | movq %rdx, 0x8*2(%rdi) |
| 76 | movq %r8, 0x8*3(%rdi) |
| 77 | movq %r9, 0x8*4(%rdi) |
| 78 | movq %r10, 0x8*5(%rdi) |
| 79 | movq %r11, 0x8*6(%rdi) |
| 80 | movq %r12, 0x8*7(%rdi) |
| 81 | |
| 82 | leaq 64(%rdi), %rdi |
| 83 | leaq 64(%rsi), %rsi |
| 84 | jnz .Loop2 |
| 85 | |
| 86 | movq (%rsp), %rbx |
| 87 | movq 1*8(%rsp), %r12 |
| 88 | addq $2*8, %rsp |
| 89 | RET |
| 90 | SYM_FUNC_END(copy_page_regs) |
| 91 |