xref: /linux-6.15/drivers/gpu/drm/drm_cache.c (revision 3abc6670)
1 /**************************************************************************
2  *
3  * Copyright (c) 2006-2007 Tungsten Graphics, Inc., Cedar Park, TX., USA
4  * All Rights Reserved.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the
8  * "Software"), to deal in the Software without restriction, including
9  * without limitation the rights to use, copy, modify, merge, publish,
10  * distribute, sub license, and/or sell copies of the Software, and to
11  * permit persons to whom the Software is furnished to do so, subject to
12  * the following conditions:
13  *
14  * The above copyright notice and this permission notice (including the
15  * next paragraph) shall be included in all copies or substantial portions
16  * of the Software.
17  *
18  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
19  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
20  * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
21  * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
22  * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
23  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
24  * USE OR OTHER DEALINGS IN THE SOFTWARE.
25  *
26  **************************************************************************/
27 /*
28  * Authors: Thomas Hellström <thomas-at-tungstengraphics-dot-com>
29  */
30 
31 #include <linux/export.h>
32 #include <linux/highmem.h>
33 #include <xen/xen.h>
34 
35 #include <drm/drm_cache.h>
36 
37 #if defined(CONFIG_X86)
38 #include <asm/smp.h>
39 
40 /*
41  * clflushopt is an unordered instruction which needs fencing with mfence or
42  * sfence to avoid ordering issues.  For drm_clflush_page this fencing happens
43  * in the caller.
44  */
45 static void
46 drm_clflush_page(struct page *page)
47 {
48 	uint8_t *page_virtual;
49 	unsigned int i;
50 	const int size = boot_cpu_data.x86_clflush_size;
51 
52 	if (unlikely(page == NULL))
53 		return;
54 
55 	page_virtual = kmap_atomic(page);
56 	for (i = 0; i < PAGE_SIZE; i += size)
57 		clflushopt(page_virtual + i);
58 	kunmap_atomic(page_virtual);
59 }
60 
61 static void drm_cache_flush_clflush(struct page *pages[],
62 				    unsigned long num_pages)
63 {
64 	unsigned long i;
65 
66 	mb(); /*Full memory barrier used before so that CLFLUSH is ordered*/
67 	for (i = 0; i < num_pages; i++)
68 		drm_clflush_page(*pages++);
69 	mb(); /*Also used after CLFLUSH so that all cache is flushed*/
70 }
71 #endif
72 
73 /**
74  * drm_clflush_pages - Flush dcache lines of a set of pages.
75  * @pages: List of pages to be flushed.
76  * @num_pages: Number of pages in the array.
77  *
78  * Flush every data cache line entry that points to an address belonging
79  * to a page in the array.
80  */
81 void
82 drm_clflush_pages(struct page *pages[], unsigned long num_pages)
83 {
84 
85 #if defined(CONFIG_X86)
86 	if (static_cpu_has(X86_FEATURE_CLFLUSH)) {
87 		drm_cache_flush_clflush(pages, num_pages);
88 		return;
89 	}
90 
91 	if (wbinvd_on_all_cpus())
92 		pr_err("Timed out waiting for cache flush\n");
93 
94 #elif defined(__powerpc__)
95 	unsigned long i;
96 
97 	for (i = 0; i < num_pages; i++) {
98 		struct page *page = pages[i];
99 		void *page_virtual;
100 
101 		if (unlikely(page == NULL))
102 			continue;
103 
104 		page_virtual = kmap_atomic(page);
105 		flush_dcache_range((unsigned long)page_virtual,
106 				   (unsigned long)page_virtual + PAGE_SIZE);
107 		kunmap_atomic(page_virtual);
108 	}
109 #else
110 	pr_err("Architecture has no drm_cache.c support\n");
111 	WARN_ON_ONCE(1);
112 #endif
113 }
114 EXPORT_SYMBOL(drm_clflush_pages);
115 
116 /**
117  * drm_clflush_sg - Flush dcache lines pointing to a scather-gather.
118  * @st: struct sg_table.
119  *
120  * Flush every data cache line entry that points to an address in the
121  * sg.
122  */
123 void
124 drm_clflush_sg(struct sg_table *st)
125 {
126 #if defined(CONFIG_X86)
127 	if (static_cpu_has(X86_FEATURE_CLFLUSH)) {
128 		struct sg_page_iter sg_iter;
129 
130 		mb(); /*CLFLUSH is ordered only by using memory barriers*/
131 		for_each_sgtable_page(st, &sg_iter, 0)
132 			drm_clflush_page(sg_page_iter_page(&sg_iter));
133 		mb(); /*Make sure that all cache line entry is flushed*/
134 
135 		return;
136 	}
137 
138 	if (wbinvd_on_all_cpus())
139 		pr_err("Timed out waiting for cache flush\n");
140 #else
141 	pr_err("Architecture has no drm_cache.c support\n");
142 	WARN_ON_ONCE(1);
143 #endif
144 }
145 EXPORT_SYMBOL(drm_clflush_sg);
146 
147 /**
148  * drm_clflush_virt_range - Flush dcache lines of a region
149  * @addr: Initial kernel memory address.
150  * @length: Region size.
151  *
152  * Flush every data cache line entry that points to an address in the
153  * region requested.
154  */
155 void
156 drm_clflush_virt_range(void *addr, unsigned long length)
157 {
158 #if defined(CONFIG_X86)
159 	if (static_cpu_has(X86_FEATURE_CLFLUSH)) {
160 		const int size = boot_cpu_data.x86_clflush_size;
161 		void *end = addr + length;
162 
163 		addr = (void *)(((unsigned long)addr) & -size);
164 		mb(); /*CLFLUSH is only ordered with a full memory barrier*/
165 		for (; addr < end; addr += size)
166 			clflushopt(addr);
167 		clflushopt(end - 1); /* force serialisation */
168 		mb(); /*Ensure that evry data cache line entry is flushed*/
169 		return;
170 	}
171 
172 	if (wbinvd_on_all_cpus())
173 		pr_err("Timed out waiting for cache flush\n");
174 #else
175 	pr_err("Architecture has no drm_cache.c support\n");
176 	WARN_ON_ONCE(1);
177 #endif
178 }
179 EXPORT_SYMBOL(drm_clflush_virt_range);
180 
181 bool drm_need_swiotlb(int dma_bits)
182 {
183 	struct resource *tmp;
184 	resource_size_t max_iomem = 0;
185 
186 	/*
187 	 * Xen paravirtual hosts require swiotlb regardless of requested dma
188 	 * transfer size.
189 	 *
190 	 * NOTE: Really, what it requires is use of the dma_alloc_coherent
191 	 *       allocator used in ttm_dma_populate() instead of
192 	 *       ttm_populate_and_map_pages(), which bounce buffers so much in
193 	 *       Xen it leads to swiotlb buffer exhaustion.
194 	 */
195 	if (xen_pv_domain())
196 		return true;
197 
198 	/*
199 	 * Enforce dma_alloc_coherent when memory encryption is active as well
200 	 * for the same reasons as for Xen paravirtual hosts.
201 	 */
202 	if (mem_encrypt_active())
203 		return true;
204 
205 	for (tmp = iomem_resource.child; tmp; tmp = tmp->sibling)
206 		max_iomem = max(max_iomem,  tmp->end);
207 
208 	return max_iomem > ((u64)1 << dma_bits);
209 }
210 EXPORT_SYMBOL(drm_need_swiotlb);
211