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