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 <linux/string.h>
27#include <linux/cpufeature.h>
28#include <linux/bug.h>
29#include <linux/build_bug.h>
30#include <asm/fpu/api.h>
31
32#include "i915_memcpy.h"
33
34#if IS_ENABLED(CONFIG_DRM_I915_DEBUG)
35#define CI_BUG_ON(expr) BUG_ON(expr)
36#else
37#define CI_BUG_ON(expr) BUILD_BUG_ON_INVALID(expr)
38#endif
39
40static DEFINE_STATIC_KEY_FALSE(has_movntdqa);
41
42static void __memcpy_ntdqa(void *dst, const void *src, unsigned long len)
43{
44 kernel_fpu_begin();
45
46 while (len >= 4) {
47 asm("movntdqa (%0), %%xmm0\n"
48 "movntdqa 16(%0), %%xmm1\n"
49 "movntdqa 32(%0), %%xmm2\n"
50 "movntdqa 48(%0), %%xmm3\n"
51 "movaps %%xmm0, (%1)\n"
52 "movaps %%xmm1, 16(%1)\n"
53 "movaps %%xmm2, 32(%1)\n"
54 "movaps %%xmm3, 48(%1)\n"
55 :: "r" (src), "r" (dst) : "memory");
56 src += 64;
57 dst += 64;
58 len -= 4;
59 }
60 while (len--) {
61 asm("movntdqa (%0), %%xmm0\n"
62 "movaps %%xmm0, (%1)\n"
63 :: "r" (src), "r" (dst) : "memory");
64 src += 16;
65 dst += 16;
66 }
67
68 kernel_fpu_end();
69}
70
71static void __memcpy_ntdqu(void *dst, const void *src, unsigned long len)
72{
73 kernel_fpu_begin();
74
75 while (len >= 4) {
76 asm("movntdqa (%0), %%xmm0\n"
77 "movntdqa 16(%0), %%xmm1\n"
78 "movntdqa 32(%0), %%xmm2\n"
79 "movntdqa 48(%0), %%xmm3\n"
80 "movups %%xmm0, (%1)\n"
81 "movups %%xmm1, 16(%1)\n"
82 "movups %%xmm2, 32(%1)\n"
83 "movups %%xmm3, 48(%1)\n"
84 :: "r" (src), "r" (dst) : "memory");
85 src += 64;
86 dst += 64;
87 len -= 4;
88 }
89 while (len--) {
90 asm("movntdqa (%0), %%xmm0\n"
91 "movups %%xmm0, (%1)\n"
92 :: "r" (src), "r" (dst) : "memory");
93 src += 16;
94 dst += 16;
95 }
96
97 kernel_fpu_end();
98}
99
100/**
101 * i915_memcpy_from_wc: perform an accelerated *aligned* read from WC
102 * @dst: destination pointer
103 * @src: source pointer
104 * @len: how many bytes to copy
105 *
106 * i915_memcpy_from_wc copies @len bytes from @src to @dst using
107 * non-temporal instructions where available. Note that all arguments
108 * (@src, @dst) must be aligned to 16 bytes and @len must be a multiple
109 * of 16.
110 *
111 * To test whether accelerated reads from WC are supported, use
112 * i915_memcpy_from_wc(NULL, NULL, 0);
113 *
114 * Returns true if the copy was successful, false if the preconditions
115 * are not met.
116 */
117bool i915_memcpy_from_wc(void *dst, const void *src, unsigned long len)
118{
119 if (unlikely(((unsigned long)dst | (unsigned long)src | len) & 15))
120 return false;
121
122 if (static_branch_likely(&has_movntdqa)) {
123 if (likely(len))
124 __memcpy_ntdqa(dst, src, len: len >> 4);
125 return true;
126 }
127
128 return false;
129}
130
131/**
132 * i915_unaligned_memcpy_from_wc: perform a mostly accelerated read from WC
133 * @dst: destination pointer
134 * @src: source pointer
135 * @len: how many bytes to copy
136 *
137 * Like i915_memcpy_from_wc(), the unaligned variant copies @len bytes from
138 * @src to @dst using * non-temporal instructions where available, but
139 * accepts that its arguments may not be aligned, but are valid for the
140 * potential 16-byte read past the end.
141 */
142void i915_unaligned_memcpy_from_wc(void *dst, const void *src, unsigned long len)
143{
144 unsigned long addr;
145
146 CI_BUG_ON(!i915_has_memcpy_from_wc());
147
148 addr = (unsigned long)src;
149 if (!IS_ALIGNED(addr, 16)) {
150 unsigned long x = min(ALIGN(addr, 16) - addr, len);
151
152 memcpy(dst, src, x);
153
154 len -= x;
155 dst += x;
156 src += x;
157 }
158
159 if (likely(len))
160 __memcpy_ntdqu(dst, src, DIV_ROUND_UP(len, 16));
161}
162
163void i915_memcpy_init_early(struct drm_i915_private *dev_priv)
164{
165 /*
166 * Some hypervisors (e.g. KVM) don't support VEX-prefix instructions
167 * emulation. So don't enable movntdqa in hypervisor guest.
168 */
169 if (static_cpu_has(X86_FEATURE_XMM4_1) &&
170 !boot_cpu_has(X86_FEATURE_HYPERVISOR))
171 static_branch_enable(&has_movntdqa);
172}
173

source code of linux/drivers/gpu/drm/i915/i915_memcpy.c