1 /* 2 * Copyright © 2016 Intel Corporation 3 * 4 * Permission is hereby granted, free of charge, to any person obtaining a 5 * copy of this software and associated documentation files (the "Software"), 6 * to deal in the Software without restriction, including without limitation 7 * the rights to use, copy, modify, merge, publish, distribute, sublicense, 8 * and/or sell copies of the Software, and to permit persons to whom the 9 * Software is furnished to do so, subject to the following conditions: 10 * 11 * The above copyright notice and this permission notice (including the next 12 * paragraph) shall be included in all copies or substantial portions of the 13 * Software. 14 * 15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 18 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING 20 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS 21 * IN THE SOFTWARE. 22 * 23 */ 24 25 #include <linux/kernel.h> 26 #include <asm/fpu/api.h> 27 28 #include "i915_memcpy.h" 29 30 #if IS_ENABLED(CONFIG_DRM_I915_DEBUG) 31 #define CI_BUG_ON(expr) BUG_ON(expr) 32 #else 33 #define CI_BUG_ON(expr) BUILD_BUG_ON_INVALID(expr) 34 #endif 35 36 static DEFINE_STATIC_KEY_FALSE(has_movntdqa); 37 38 #ifdef CONFIG_AS_MOVNTDQA 39 static void __memcpy_ntdqa(void *dst, const void *src, unsigned long len) 40 { 41 kernel_fpu_begin(); 42 43 while (len >= 4) { 44 asm("movntdqa (%0), %%xmm0\n" 45 "movntdqa 16(%0), %%xmm1\n" 46 "movntdqa 32(%0), %%xmm2\n" 47 "movntdqa 48(%0), %%xmm3\n" 48 "movaps %%xmm0, (%1)\n" 49 "movaps %%xmm1, 16(%1)\n" 50 "movaps %%xmm2, 32(%1)\n" 51 "movaps %%xmm3, 48(%1)\n" 52 :: "r" (src), "r" (dst) : "memory"); 53 src += 64; 54 dst += 64; 55 len -= 4; 56 } 57 while (len--) { 58 asm("movntdqa (%0), %%xmm0\n" 59 "movaps %%xmm0, (%1)\n" 60 :: "r" (src), "r" (dst) : "memory"); 61 src += 16; 62 dst += 16; 63 } 64 65 kernel_fpu_end(); 66 } 67 68 static void __memcpy_ntdqu(void *dst, const void *src, unsigned long len) 69 { 70 kernel_fpu_begin(); 71 72 while (len >= 4) { 73 asm("movntdqa (%0), %%xmm0\n" 74 "movntdqa 16(%0), %%xmm1\n" 75 "movntdqa 32(%0), %%xmm2\n" 76 "movntdqa 48(%0), %%xmm3\n" 77 "movups %%xmm0, (%1)\n" 78 "movups %%xmm1, 16(%1)\n" 79 "movups %%xmm2, 32(%1)\n" 80 "movups %%xmm3, 48(%1)\n" 81 :: "r" (src), "r" (dst) : "memory"); 82 src += 64; 83 dst += 64; 84 len -= 4; 85 } 86 while (len--) { 87 asm("movntdqa (%0), %%xmm0\n" 88 "movups %%xmm0, (%1)\n" 89 :: "r" (src), "r" (dst) : "memory"); 90 src += 16; 91 dst += 16; 92 } 93 94 kernel_fpu_end(); 95 } 96 #else 97 static void __memcpy_ntdqa(void *dst, const void *src, unsigned long len) {} 98 static void __memcpy_ntdqu(void *dst, const void *src, unsigned long len) {} 99 #endif 100 101 /** 102 * i915_memcpy_from_wc: perform an accelerated *aligned* read from WC 103 * @dst: destination pointer 104 * @src: source pointer 105 * @len: how many bytes to copy 106 * 107 * i915_memcpy_from_wc copies @len bytes from @src to @dst using 108 * non-temporal instructions where available. Note that all arguments 109 * (@src, @dst) must be aligned to 16 bytes and @len must be a multiple 110 * of 16. 111 * 112 * To test whether accelerated reads from WC are supported, use 113 * i915_memcpy_from_wc(NULL, NULL, 0); 114 * 115 * Returns true if the copy was successful, false if the preconditions 116 * are not met. 117 */ 118 bool i915_memcpy_from_wc(void *dst, const void *src, unsigned long len) 119 { 120 if (unlikely(((unsigned long)dst | (unsigned long)src | len) & 15)) 121 return false; 122 123 if (static_branch_likely(&has_movntdqa)) { 124 if (likely(len)) 125 __memcpy_ntdqa(dst, src, len >> 4); 126 return true; 127 } 128 129 return false; 130 } 131 132 /** 133 * i915_unaligned_memcpy_from_wc: perform a mostly accelerated read from WC 134 * @dst: destination pointer 135 * @src: source pointer 136 * @len: how many bytes to copy 137 * 138 * Like i915_memcpy_from_wc(), the unaligned variant copies @len bytes from 139 * @src to @dst using * non-temporal instructions where available, but 140 * accepts that its arguments may not be aligned, but are valid for the 141 * potential 16-byte read past the end. 142 */ 143 void i915_unaligned_memcpy_from_wc(void *dst, void *src, unsigned long len) 144 { 145 unsigned long addr; 146 147 CI_BUG_ON(!i915_has_memcpy_from_wc()); 148 149 addr = (unsigned long)src; 150 if (!IS_ALIGNED(addr, 16)) { 151 unsigned long x = min(ALIGN(addr, 16) - addr, len); 152 153 memcpy(dst, src, x); 154 155 len -= x; 156 dst += x; 157 src += x; 158 } 159 160 if (likely(len)) 161 __memcpy_ntdqu(dst, src, DIV_ROUND_UP(len, 16)); 162 } 163 164 void i915_memcpy_init_early(struct drm_i915_private *dev_priv) 165 { 166 /* 167 * Some hypervisors (e.g. KVM) don't support VEX-prefix instructions 168 * emulation. So don't enable movntdqa in hypervisor guest. 169 */ 170 if (static_cpu_has(X86_FEATURE_XMM4_1) && 171 !boot_cpu_has(X86_FEATURE_HYPERVISOR)) 172 static_branch_enable(&has_movntdqa); 173 } 174