{"id":75371,"date":"2026-02-12T07:33:19","date_gmt":"2026-02-11T23:33:19","guid":{"rendered":"https:\/\/www.wsisp.com\/helps\/75371.html"},"modified":"2026-02-12T07:33:19","modified_gmt":"2026-02-11T23:33:19","slug":"07-svm%e5%86%85%e5%ad%98%e8%bf%81%e7%a7%bb%e6%9c%ba%e5%88%b6","status":"publish","type":"post","link":"https:\/\/www.wsisp.com\/helps\/75371.html","title":{"rendered":"07 - SVM\u5185\u5b58\u8fc1\u79fb\u673a\u5236"},"content":{"rendered":"<p>\u96be\u5ea6: &#x1f534;&#x1f534; \u9ad8\u7ea7 \u9884\u8ba1\u5b66\u4e60\u65f6\u95f4: 2.5-3\u5c0f\u65f6 \u524d\u7f6e\u77e5\u8bc6: \u7b2c6\u7ae0\u3001DMA\u57fa\u7840\u3001SDMA\/TTM\u6982\u5ff5<\/p>\n<hr \/>\n<h3>&#x1f4cb; \u6982\u8ff0<\/h3>\n<p>\u5185\u5b58\u8fc1\u79fb\u662fSVM\u7684\u6838\u5fc3\u7279\u6027&#xff0c;\u8d1f\u8d23\u5728\u7cfb\u7edfRAM\u548cGPU VRAM\u4e4b\u95f4\u81ea\u52a8\u79fb\u52a8\u9875\u9762\u3002\u6253\u4e2a\u6bd4\u65b9&#xff0c;\u8fc1\u79fb\u5c31\u50cf\u5728\u4e24\u4e2a\u4ed3\u5e93\u95f4\u642c\u8d27\u7269&#xff1a;<\/p>\n<ul>\n<li>&#x1f3e2; \u7cfb\u7edfRAM: CPU\u7684\u672c\u5730\u4ed3\u5e93&#xff0c;\u8bbf\u95ee\u5feb\u4f46GPU\u8bbf\u95ee\u6162<\/li>\n<li>&#x1f3ed; GPU VRAM: GPU\u7684\u672c\u5730\u4ed3\u5e93&#xff0c;GPU\u8bbf\u95ee\u5feb\u4f46CPU\u8bbf\u95ee\u6162&#xff08;\u6216\u65e0\u6cd5\u8bbf\u95ee&#xff09;<\/li>\n<li>&#x1f69a; SDMA\u5f15\u64ce: GPU\u7684\u642c\u8fd0\u5de5&#xff0c;\u4e13\u95e8\u8d1f\u8d23\u6570\u636e\u642c\u8fd0<\/li>\n<li>&#x1f4cb; GART\u8868: \u5730\u5740\u7ffb\u8bd1\u6e05\u5355&#xff0c;\u8ba9SDMA\u627e\u5230\u6b63\u786e\u5730\u5740<\/li>\n<\/ul>\n<p>\u672c\u7ae0\u6df1\u5165\u8fc1\u79fb\u673a\u5236\u7684\u5b9e\u73b0\u7ec6\u8282\u3002<\/p>\n<hr \/>\n<h3>7.1 \u8fc1\u79fb\u673a\u5236\u6982\u89c8<\/h3>\n<h4>\u8fc1\u79fb\u7684\u4e24\u4e2a\u65b9\u5411<\/h4>\n<p>RAM \u2192 VRAM (\u4e0a\u884c\u8fc1\u79fb)<br \/>\n    \u89e6\u53d1\u573a\u666f&#xff1a;<br \/>\n    &#8211; GPU\u9891\u7e41\u8bbf\u95ee\u67d0\u6bb5\u5185\u5b58<br \/>\n    &#8211; \u7528\u6237\u663e\u5f0f\u8bf7\u6c42 (prefetch)<br \/>\n    &#8211; GPU\u9875\u9762\u9519\u8bef\u89e6\u53d1<\/p>\n<p>    \u6d41\u7a0b&#xff1a;<br \/>\n    \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2510     \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2510     \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2510<br \/>\n    \u2502 RAM  \u2502 \u2192\u2192\u2192 \u2502 SDMA \u2502 \u2192\u2192\u2192 \u2502 VRAM \u2502<br \/>\n    \u2502pages \u2502     \u2502 copy \u2502     \u2502pages \u2502<br \/>\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2518     \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2518     \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2518<\/p>\n<p>VRAM \u2192 RAM (\u4e0b\u884c\u8fc1\u79fb)<br \/>\n    \u89e6\u53d1\u573a\u666f&#xff1a;<br \/>\n    &#8211; CPU\u9700\u8981\u8bbf\u95eeVRAM\u9875\u9762<br \/>\n    &#8211; \u5185\u5b58\u4e0d\u8db3\u9700\u8981\u56de\u6536VRAM<br \/>\n    &#8211; \u8fdb\u7a0b\u9000\u51fa\u6e05\u7406<\/p>\n<p>    \u6d41\u7a0b&#xff1a;<br \/>\n    \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2510     \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2510     \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2510<br \/>\n    \u2502 VRAM \u2502 \u2192\u2192\u2192 \u2502 SDMA \u2502 \u2192\u2192\u2192 \u2502 RAM  \u2502<br \/>\n    \u2502pages \u2502     \u2502 copy \u2502     \u2502pages \u2502<br \/>\n    \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2518     \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2518     \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2518<\/p>\n<h4>\u8fc1\u79fb\u5b50\u7cfb\u7edf\u7ec4\u4ef6<\/h4>\n<p><span class=\"token comment\">\/\/ \u4e3b\u8981\u7ec4\u4ef6<\/span><br \/>\n<span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">migrate_vma<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n    <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">vm_area_struct<\/span> <span class=\"token operator\">*<\/span>vma<span class=\"token punctuation\">;<\/span>  <span class=\"token comment\">\/\/ \u865a\u62df\u5185\u5b58\u533a\u57df<\/span><br \/>\n    <span class=\"token keyword\">unsigned<\/span> <span class=\"token keyword\">long<\/span> start<span class=\"token punctuation\">;<\/span>         <span class=\"token comment\">\/\/ \u8d77\u59cb\u5730\u5740<\/span><br \/>\n    <span class=\"token keyword\">unsigned<\/span> <span class=\"token keyword\">long<\/span> end<span class=\"token punctuation\">;<\/span>           <span class=\"token comment\">\/\/ \u7ed3\u675f\u5730\u5740<\/span><br \/>\n    <span class=\"token keyword\">unsigned<\/span> <span class=\"token keyword\">long<\/span> <span class=\"token operator\">*<\/span>src<span class=\"token punctuation\">;<\/span>          <span class=\"token comment\">\/\/ \u6e90\u9875\u9762\u6570\u7ec4<\/span><br \/>\n    <span class=\"token keyword\">unsigned<\/span> <span class=\"token keyword\">long<\/span> <span class=\"token operator\">*<\/span>dst<span class=\"token punctuation\">;<\/span>          <span class=\"token comment\">\/\/ \u76ee\u6807\u9875\u9762\u6570\u7ec4<\/span><br \/>\n    <span class=\"token keyword\">unsigned<\/span> <span class=\"token keyword\">long<\/span> cpages<span class=\"token punctuation\">;<\/span>        <span class=\"token comment\">\/\/ \u6536\u96c6\u7684\u9875\u9762\u6570<\/span><br \/>\n    <span class=\"token comment\">\/\/ &#8230;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>\u6838\u5fc3API&#xff1a;<\/p>\n<li>migrate_vma_setup(): \u51c6\u5907\u8fc1\u79fb&#xff0c;\u9501\u5b9a\u9875\u9762<\/li>\n<li>migrate_vma_pages(): \u6267\u884c\u8fc1\u79fb&#xff0c;\u66f4\u65b0\u9875\u8868<\/li>\n<li>migrate_vma_finalize(): \u5b8c\u6210\u8fc1\u79fb&#xff0c;\u6e05\u7406\u8d44\u6e90<\/li>\n<hr \/>\n<h3>7.2 \u4eceRAM\u5230VRAM\u7684\u8fc1\u79fb<\/h3>\n<h4>\u4e3b\u51fd\u6570&#xff1a;svm_migrate_ram_to_vram<\/h4>\n<p><span class=\"token keyword\">int<\/span> <span class=\"token function\">svm_migrate_ram_to_vram<\/span><span class=\"token punctuation\">(<\/span><span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">svm_range<\/span> <span class=\"token operator\">*<\/span>prange<span class=\"token punctuation\">,<\/span><br \/>\n                            <span class=\"token class-name\">uint32_t<\/span> best_loc<span class=\"token punctuation\">,<\/span><br \/>\n                            <span class=\"token keyword\">unsigned<\/span> <span class=\"token keyword\">long<\/span> start_mgr<span class=\"token punctuation\">,<\/span><br \/>\n                            <span class=\"token keyword\">unsigned<\/span> <span class=\"token keyword\">long<\/span> last_mgr<span class=\"token punctuation\">,<\/span><br \/>\n                            <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">mm_struct<\/span> <span class=\"token operator\">*<\/span>mm<span class=\"token punctuation\">,<\/span><br \/>\n                            <span class=\"token class-name\">uint32_t<\/span> trigger<span class=\"token punctuation\">)<\/span><\/p>\n<p>\u53c2\u6570\u8be6\u89e3&#xff1a;<\/p>\n<ul>\n<li>best_loc: \u76ee\u6807GPU\u8282\u70b9ID<\/li>\n<li>start_mgr\/last_mgr: \u8fc1\u79fb\u8303\u56f4&#xff08;\u9875\u53f7&#xff09;<\/li>\n<li>mm: \u8fdb\u7a0b\u7684mm_struct<\/li>\n<li>trigger: \u89e6\u53d1\u539f\u56e0&#xff08;KFD_MIGRATE_TRIGGER_xxx&#xff09;<\/li>\n<\/ul>\n<h4>\u8fc1\u79fb\u6d41\u7a0b\u8be6\u89e3<\/h4>\n<p>\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510<br \/>\n\u2502 1. \u9884\u68c0\u67e5                                \u2502<br \/>\n\u2502    &#8211; \u9a8c\u8bc1\u8303\u56f4\u6709\u6548\u6027                      \u2502<br \/>\n\u2502    &#8211; \u83b7\u53d6\u76ee\u6807GPU\u8282\u70b9                     \u2502<br \/>\n\u2502    &#8211; \u9884\u7559VRAM\u7a7a\u95f4                        \u2502<br \/>\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<br \/>\n                \u2193<br \/>\n\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510<br \/>\n\u2502 2. \u5206\u914dVRAM                             \u2502<br \/>\n\u2502    svm_range_vram_node_new()            \u2502<br \/>\n\u2502    &#8211; \u901a\u8fc7TTM\u5206\u914dVRAM                     \u2502<br \/>\n\u2502    &#8211; \u8bb0\u5f55\u5728prange-&gt;ttm_res               \u2502<br \/>\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<br \/>\n                \u2193<br \/>\n\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510<br \/>\n\u2502 3. \u6309VMA\u904d\u5386\u8fc1\u79fb                         \u2502<br \/>\n\u2502    for each VMA in [start, end]:        \u2502<br \/>\n\u2502      svm_migrate_vma_to_vram()          \u2502<br \/>\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<br \/>\n                \u2193<br \/>\n\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510<br \/>\n\u2502 4. \u66f4\u65b0\u7edf\u8ba1\u4fe1\u606f                          \u2502<br \/>\n\u2502    pdd-&gt;page_in &#043;&#061; mpages               \u2502<br \/>\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<\/p>\n<h4>svm_migrate_vma_to_vram \u8be6\u89e3<\/h4>\n<p><span class=\"token comment\">\/**<br \/>\n * svm_migrate_ram_to_vram &#8211; migrate svm range from system to device<br \/>\n * &#064;prange: range structure<br \/>\n * &#064;best_loc: the device to migrate to<br \/>\n * &#064;start_mgr: start page to migrate<br \/>\n * &#064;last_mgr: last page to migrate<br \/>\n * &#064;mm: the process mm structure<br \/>\n * &#064;trigger: reason of migration<br \/>\n *<br \/>\n * Context: Process context, caller hold mmap read lock, svms lock, prange lock<br \/>\n *<br \/>\n * Return:<br \/>\n * 0 &#8211; OK, otherwise error code<br \/>\n *\/<\/span><br \/>\n<span class=\"token keyword\">static<\/span> <span class=\"token keyword\">long<\/span> <span class=\"token function\">svm_migrate_vma_to_vram<\/span><span class=\"token punctuation\">(<\/span><span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">kfd_node<\/span> <span class=\"token operator\">*<\/span>node<span class=\"token punctuation\">,<\/span><br \/>\n                                   <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">svm_range<\/span> <span class=\"token operator\">*<\/span>prange<span class=\"token punctuation\">,<\/span><br \/>\n                                   <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">vm_area_struct<\/span> <span class=\"token operator\">*<\/span>vma<span class=\"token punctuation\">,<\/span><br \/>\n                                   <span class=\"token class-name\">uint64_t<\/span> start<span class=\"token punctuation\">,<\/span> <span class=\"token class-name\">uint64_t<\/span> end<span class=\"token punctuation\">,<\/span><br \/>\n                                   <span class=\"token class-name\">uint32_t<\/span> trigger<span class=\"token punctuation\">,<\/span> u64 ttm_res_offset<span class=\"token punctuation\">)<\/span><br \/>\n<span class=\"token punctuation\">{<\/span><br \/>\n    <span class=\"token comment\">\/\/ 1. \u51c6\u5907\u8fc1\u79fb\u4e0a\u4e0b\u6587<\/span><br \/>\n    <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">migrate_vma<\/span> migrate<span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token function\">memset<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">,<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token keyword\">sizeof<\/span><span class=\"token punctuation\">(<\/span>migrate<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>vma <span class=\"token operator\">&#061;<\/span> vma<span class=\"token punctuation\">;<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>start <span class=\"token operator\">&#061;<\/span> start<span class=\"token punctuation\">;<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>end <span class=\"token operator\">&#061;<\/span> end<span class=\"token punctuation\">;<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>flags <span class=\"token operator\">&#061;<\/span> MIGRATE_VMA_SELECT_SYSTEM<span class=\"token punctuation\">;<\/span>  <span class=\"token comment\">\/\/ \u53ea\u8fc1\u79fb\u7cfb\u7edf\u5185\u5b58<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>pgmap_owner <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">SVM_ADEV_PGMAP_OWNER<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 2. \u5206\u914d\u4e34\u65f6\u7f13\u51b2\u533a<\/span><br \/>\n    <span class=\"token comment\">\/\/ src\u6570\u7ec4 &#043; dst\u6570\u7ec4 &#043; DMA\u5730\u5740\u6570\u7ec4<\/span><br \/>\n    buf <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">kvcalloc<\/span><span class=\"token punctuation\">(<\/span>npages<span class=\"token punctuation\">,<\/span><br \/>\n                  <span class=\"token number\">2<\/span> <span class=\"token operator\">*<\/span> <span class=\"token keyword\">sizeof<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">*<\/span>migrate<span class=\"token punctuation\">.<\/span>src<span class=\"token punctuation\">)<\/span> <span class=\"token operator\">&#043;<\/span> <span class=\"token keyword\">sizeof<\/span><span class=\"token punctuation\">(<\/span>u64<span class=\"token punctuation\">)<\/span> <span class=\"token operator\">&#043;<\/span> <span class=\"token keyword\">sizeof<\/span><span class=\"token punctuation\">(<\/span><span class=\"token class-name\">dma_addr_t<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">,<\/span><br \/>\n                  GFP_KERNEL<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>src <span class=\"token operator\">&#061;<\/span> buf<span class=\"token punctuation\">;<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>dst <span class=\"token operator\">&#061;<\/span> migrate<span class=\"token punctuation\">.<\/span>src <span class=\"token operator\">&#043;<\/span> npages<span class=\"token punctuation\">;<\/span><br \/>\n    scratch <span class=\"token operator\">&#061;<\/span> <span class=\"token punctuation\">(<\/span><span class=\"token class-name\">dma_addr_t<\/span> <span class=\"token operator\">*<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">(<\/span>migrate<span class=\"token punctuation\">.<\/span>dst <span class=\"token operator\">&#043;<\/span> npages<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 3. \u6536\u96c6\u9700\u8981\u8fc1\u79fb\u7684\u9875\u9762<\/span><br \/>\n    r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">migrate_vma_setup<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token keyword\">if<\/span> <span class=\"token punctuation\">(<\/span>r<span class=\"token punctuation\">)<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n        <span class=\"token comment\">\/\/ \u5931\u8d25\u53ef\u80fd\u662f\u9875\u9762\u5df2\u88ab\u9501\u5b9a\u6216\u6b63\u5728\u4f7f\u7528<\/span><br \/>\n    <span class=\"token punctuation\">}<\/span><\/p>\n<p>    cpages <span class=\"token operator\">&#061;<\/span> migrate<span class=\"token punctuation\">.<\/span>cpages<span class=\"token punctuation\">;<\/span>  <span class=\"token comment\">\/\/ \u5b9e\u9645\u6536\u96c6\u7684\u9875\u9762\u6570<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 4. \u6267\u884cSDMA\u62f7\u8d1d<\/span><br \/>\n    r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_copy_to_vram<\/span><span class=\"token punctuation\">(<\/span>node<span class=\"token punctuation\">,<\/span> prange<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">,<\/span><br \/>\n                                <span class=\"token operator\">&amp;<\/span>mfence<span class=\"token punctuation\">,<\/span> scratch<span class=\"token punctuation\">,<\/span> ttm_res_offset<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 5. \u66f4\u65b0\u9875\u8868<\/span><br \/>\n    <span class=\"token function\">migrate_vma_pages<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 6. \u7b49\u5f85SDMA\u5b8c\u6210<\/span><br \/>\n    <span class=\"token function\">svm_migrate_copy_done<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span> mfence<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 7. \u5b8c\u6210\u8fc1\u79fb<\/span><br \/>\n    <span class=\"token function\">migrate_vma_finalize<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 8. \u6e05\u7406DMA\u6620\u5c04<\/span><br \/>\n    <span class=\"token function\">svm_range_dma_unmap_dev<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token operator\">-&gt;<\/span>dev<span class=\"token punctuation\">,<\/span> scratch<span class=\"token punctuation\">,<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">,<\/span> npages<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token keyword\">return<\/span> mpages<span class=\"token punctuation\">;<\/span>  <span class=\"token comment\">\/\/ \u6210\u529f\u8fc1\u79fb\u7684\u9875\u9762\u6570<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><\/p>\n<h4>migrate_vma_setup \u505a\u4e86\u4ec0\u4e48&#xff1f;<\/h4>\n<p>\u8fd9\u662fLinux\u5185\u6838HMM\u6846\u67b6\u63d0\u4f9b\u7684API&#xff1a;<\/p>\n<p>1. \u9501\u5b9aVMA\u548c\u9875\u8868<br \/>\n2. \u904d\u5386\u9875\u8868&#xff0c;\u627e\u5230\u6ee1\u8db3\u6761\u4ef6\u7684\u9875\u9762&#xff1a;<br \/>\n   &#8211; \u9875\u9762\u5728\u7cfb\u7edfRAM (MIGRATE_VMA_SELECT_SYSTEM)<br \/>\n   &#8211; \u9875\u9762\u672a\u88ab\u9501\u5b9a (pin_user_pages)<br \/>\n   &#8211; \u9875\u9762\u4e0d\u662f\u7279\u6b8a\u9875 (VM_SPECIAL)<br \/>\n3. \u589e\u52a0\u9875\u9762\u5f15\u7528\u8ba1\u6570&#xff0c;\u9632\u6b62\u88ab\u91ca\u653e<br \/>\n4. \u8bb0\u5f55\u5230migrate.src\u6570\u7ec4&#xff1a;<br \/>\n   src[i] &#061; page_to_pfn(page) | MIGRATE_PFN_VALID<\/p>\n<p>\u793a\u4f8b&#xff1a;<\/p>\n<p>VMA: [0x1000-0x3000] 8\u9875<br \/>\n    \u98750: \u5728RAM     \u2192 src[0] &#061; PFN(\u98750) | MIGRATE_PFN_VALID<br \/>\n    \u98751: \u5728RAM     \u2192 src[1] &#061; PFN(\u98751) | MIGRATE_PFN_VALID<br \/>\n    \u98752: \u5df2\u9501\u5b9a    \u2192 src[2] &#061; 0 (\u8df3\u8fc7)<br \/>\n    \u98753: \u5728RAM     \u2192 src[3] &#061; PFN(\u98753) | MIGRATE_PFN_VALID<br \/>\n    &#8230;<\/p>\n<p>cpages &#061; 3  (\u5b9e\u9645\u53ea\u67093\u9875\u53ef\u8fc1\u79fb)<\/p>\n<hr \/>\n<h3>7.3 SDMA\u62f7\u8d1d\u5b9e\u73b0<\/h3>\n<h4>svm_migrate_copy_to_vram<\/h4>\n<p><span class=\"token keyword\">static<\/span> <span class=\"token keyword\">int<\/span> <span class=\"token function\">svm_migrate_copy_to_vram<\/span><span class=\"token punctuation\">(<\/span><span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">kfd_node<\/span> <span class=\"token operator\">*<\/span>node<span class=\"token punctuation\">,<\/span><br \/>\n                                   <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">svm_range<\/span> <span class=\"token operator\">*<\/span>prange<span class=\"token punctuation\">,<\/span><br \/>\n                                   <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">migrate_vma<\/span> <span class=\"token operator\">*<\/span>migrate<span class=\"token punctuation\">,<\/span><br \/>\n                                   <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">dma_fence<\/span> <span class=\"token operator\">*<\/span><span class=\"token operator\">*<\/span>mfence<span class=\"token punctuation\">,<\/span><br \/>\n                                   <span class=\"token class-name\">dma_addr_t<\/span> <span class=\"token operator\">*<\/span>scratch<span class=\"token punctuation\">,<\/span><br \/>\n                                   u64 ttm_res_offset<span class=\"token punctuation\">)<\/span><br \/>\n<span class=\"token punctuation\">{<\/span><br \/>\n    u64 npages <span class=\"token operator\">&#061;<\/span> migrate<span class=\"token operator\">-&gt;<\/span>npages<span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">amdgpu_device<\/span> <span class=\"token operator\">*<\/span>adev <span class=\"token operator\">&#061;<\/span> node<span class=\"token operator\">-&gt;<\/span>adev<span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">device<\/span> <span class=\"token operator\">*<\/span>dev <span class=\"token operator\">&#061;<\/span> adev<span class=\"token operator\">-&gt;<\/span>dev<span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token class-name\">dma_addr_t<\/span> <span class=\"token operator\">*<\/span>src<span class=\"token punctuation\">;<\/span><br \/>\n    u64 <span class=\"token operator\">*<\/span>dst<span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token keyword\">int<\/span> i<span class=\"token punctuation\">,<\/span> j<span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 1. \u4e3a\u6e90\u548c\u76ee\u6807\u5206\u914d\u5730\u5740\u6570\u7ec4<\/span><br \/>\n    src <span class=\"token operator\">&#061;<\/span> scratch<span class=\"token punctuation\">;<\/span><br \/>\n    dst <span class=\"token operator\">&#061;<\/span> <span class=\"token punctuation\">(<\/span>u64 <span class=\"token operator\">*<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">(<\/span>src <span class=\"token operator\">&#043;<\/span> npages<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 2. \u6620\u5c04\u6e90\u9875\u9762\u5230\u8bbe\u5907\u53ef\u89c1\u5730\u5740<\/span><br \/>\n    r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_copy_to_vram_map_src<\/span><span class=\"token punctuation\">(<\/span>dev<span class=\"token punctuation\">,<\/span> migrate<span class=\"token punctuation\">,<\/span> scratch<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">&amp;<\/span>npages<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 3. \u51c6\u5907\u76ee\u6807VRAM\u5730\u5740<\/span><br \/>\n    r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_copy_to_vram_map_dst<\/span><span class=\"token punctuation\">(<\/span>node<span class=\"token punctuation\">,<\/span> prange<span class=\"token punctuation\">,<\/span> migrate<span class=\"token punctuation\">,<\/span><br \/>\n                                        dst<span class=\"token punctuation\">,<\/span> ttm_res_offset<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 4. \u901a\u8fc7GART\u6267\u884cSDMA\u62f7\u8d1d<\/span><br \/>\n    r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_copy_memory_gart<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span> src<span class=\"token punctuation\">,<\/span> dst<span class=\"token punctuation\">,<\/span> npages<span class=\"token punctuation\">,<\/span><br \/>\n                                    FROM_RAM_TO_VRAM<span class=\"token punctuation\">,<\/span> mfence<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token keyword\">return<\/span> r<span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><\/p>\n<h4>GART\u6620\u5c04\u539f\u7406<\/h4>\n<p>\u4ec0\u4e48\u662fGART&#xff1f;<\/p>\n<p>GART (Graphics Address Remapping Table) \u662fGPU\u7684MMU&#xff0c;\u7c7b\u4f3cCPU\u7684\u9875\u8868\u3002<\/p>\n<p>\u95ee\u9898&#xff1a;SDMA\u53ea\u80fd\u8bbf\u95eeGPU\u5730\u5740\u7a7a\u95f4<br \/>\n    \u7cfb\u7edfRAM\u4e0d\u5728GPU\u5730\u5740\u7a7a\u95f4\u5185<\/p>\n<p>\u89e3\u51b3&#xff1a;GART\u6620\u5c04<br \/>\n    1. \u5c06\u7cfb\u7edfRAM\u7684\u7269\u7406\u5730\u5740\u6620\u5c04\u5230GART\u8868<br \/>\n    2. GART\u8868\u9879\u6307\u5411\u7cfb\u7edfRAM<br \/>\n    3. SDMA\u901a\u8fc7GART\u5730\u5740\u8bbf\u95ee\u7cfb\u7edfRAM<\/p>\n<p>\u793a\u4f8b&#xff1a;<br \/>\n    RAM\u7269\u7406\u5730\u5740: 0x123456000<br \/>\n    GART\u8868\u5165\u53e3:  entry[0] &#061; 0x123456000 &#043; flags<br \/>\n    GART\u865a\u62df\u5730\u5740: 0xA00000000 (GPU\u5730\u5740\u7a7a\u95f4)<\/p>\n<p>    SDMA\u8bfb\u53d6 0xA00000000 \u2192 \u786c\u4ef6\u67e5GART\u8868 \u2192 \u8bbf\u95ee 0x123456000<\/p>\n<h4>svm_migrate_copy_memory_gart<\/h4>\n<p><span class=\"token keyword\">static<\/span> <span class=\"token keyword\">int<\/span> <span class=\"token function\">svm_migrate_copy_memory_gart<\/span><span class=\"token punctuation\">(<\/span><span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">amdgpu_device<\/span> <span class=\"token operator\">*<\/span>adev<span class=\"token punctuation\">,<\/span><br \/>\n                                       <span class=\"token class-name\">dma_addr_t<\/span> <span class=\"token operator\">*<\/span>sys<span class=\"token punctuation\">,<\/span>  <span class=\"token comment\">\/\/ \u7cfb\u7edf\u5185\u5b58DMA\u5730\u5740<\/span><br \/>\n                                       u64 <span class=\"token operator\">*<\/span>vram<span class=\"token punctuation\">,<\/span>        <span class=\"token comment\">\/\/ VRAM\u5730\u5740<\/span><br \/>\n                                       u64 npages<span class=\"token punctuation\">,<\/span><br \/>\n                                       <span class=\"token keyword\">enum<\/span> <span class=\"token class-name\">MIGRATION_COPY_DIR<\/span> direction<span class=\"token punctuation\">,<\/span><br \/>\n                                       <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">dma_fence<\/span> <span class=\"token operator\">*<\/span><span class=\"token operator\">*<\/span>mfence<span class=\"token punctuation\">)<\/span><br \/>\n<span class=\"token punctuation\">{<\/span><br \/>\n    <span class=\"token keyword\">const<\/span> u64 GTT_MAX_PAGES <span class=\"token operator\">&#061;<\/span> AMDGPU_GTT_MAX_TRANSFER_SIZE<span class=\"token punctuation\">;<\/span>  <span class=\"token comment\">\/\/ 256MB<\/span><br \/>\n    <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">amdgpu_ring<\/span> <span class=\"token operator\">*<\/span>ring <span class=\"token operator\">&#061;<\/span> adev<span class=\"token operator\">-&gt;<\/span>mman<span class=\"token punctuation\">.<\/span>buffer_funcs_ring<span class=\"token punctuation\">;<\/span><br \/>\n    u64 gart_s<span class=\"token punctuation\">,<\/span> gart_d<span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token function\">mutex_lock<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>adev<span class=\"token operator\">-&gt;<\/span>mman<span class=\"token punctuation\">.<\/span>gtt_window_lock<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ \u5206\u6279\u5904\u7406&#xff08;\u6bcf\u6b21\u6700\u591a256MB&#xff09;<\/span><br \/>\n    <span class=\"token keyword\">while<\/span> <span class=\"token punctuation\">(<\/span>npages<span class=\"token punctuation\">)<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n        size <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">min<\/span><span class=\"token punctuation\">(<\/span>GTT_MAX_PAGES<span class=\"token punctuation\">,<\/span> npages<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>        <span class=\"token keyword\">if<\/span> <span class=\"token punctuation\">(<\/span>direction <span class=\"token operator\">&#061;&#061;<\/span> FROM_RAM_TO_VRAM<span class=\"token punctuation\">)<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n            <span class=\"token comment\">\/\/ 1. \u6620\u5c04\u6e90\u7cfb\u7edf\u5185\u5b58\u5230GART<\/span><br \/>\n            r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_gart_map<\/span><span class=\"token punctuation\">(<\/span>ring<span class=\"token punctuation\">,<\/span> size<span class=\"token punctuation\">,<\/span> sys<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">&amp;<\/span>gart_s<span class=\"token punctuation\">,<\/span><br \/>\n                                    KFD_IOCTL_SVM_FLAG_GPU_RO<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>            <span class=\"token comment\">\/\/ 2. \u76ee\u6807VRAM\u76f4\u63a5\u6620\u5c04<\/span><br \/>\n            gart_d <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_direct_mapping_addr<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">*<\/span>vram<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>            <span class=\"token comment\">\/\/ 3. SDMA\u62f7\u8d1d: GART\u5730\u5740 \u2192 VRAM\u5730\u5740<\/span><br \/>\n            r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">amdgpu_copy_buffer<\/span><span class=\"token punctuation\">(<\/span>ring<span class=\"token punctuation\">,<\/span> gart_s<span class=\"token punctuation\">,<\/span> gart_d<span class=\"token punctuation\">,<\/span><br \/>\n                                  size <span class=\"token operator\">*<\/span> PAGE_SIZE<span class=\"token punctuation\">,<\/span><br \/>\n                                  <span class=\"token constant\">NULL<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token operator\">&amp;<\/span>next<span class=\"token punctuation\">,<\/span> false<span class=\"token punctuation\">,<\/span> true<span class=\"token punctuation\">,<\/span> false<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n        <span class=\"token punctuation\">}<\/span> <span class=\"token keyword\">else<\/span> <span class=\"token keyword\">if<\/span> <span class=\"token punctuation\">(<\/span>direction <span class=\"token operator\">&#061;&#061;<\/span> FROM_VRAM_TO_RAM<span class=\"token punctuation\">)<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n            <span class=\"token comment\">\/\/ \u53cd\u5411\u64cd\u4f5c<\/span><br \/>\n        <span class=\"token punctuation\">}<\/span><\/p>\n<p>        <span class=\"token comment\">\/\/ \u66f4\u65b0fence\u94fe<\/span><br \/>\n        <span class=\"token function\">dma_fence_put<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">*<\/span>mfence<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n        <span class=\"token operator\">*<\/span>mfence <span class=\"token operator\">&#061;<\/span> next<span class=\"token punctuation\">;<\/span><\/p>\n<p>        <span class=\"token comment\">\/\/ \u66f4\u65b0\u6307\u9488<\/span><br \/>\n        npages <span class=\"token operator\">-&#061;<\/span> size<span class=\"token punctuation\">;<\/span><br \/>\n        sys <span class=\"token operator\">&#043;&#061;<\/span> size<span class=\"token punctuation\">;<\/span><br \/>\n        <span class=\"token operator\">*<\/span>vram <span class=\"token operator\">&#043;&#061;<\/span> size <span class=\"token operator\">*<\/span> PAGE_SIZE<span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token punctuation\">}<\/span><\/p>\n<p>    <span class=\"token function\">mutex_unlock<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>adev<span class=\"token operator\">-&gt;<\/span>mman<span class=\"token punctuation\">.<\/span>gtt_window_lock<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token keyword\">return<\/span> r<span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><\/p>\n<p>GART\u6620\u5c04\u8be6\u7ec6\u6d41\u7a0b&#xff1a;<\/p>\n<p><span class=\"token keyword\">static<\/span> <span class=\"token keyword\">int<\/span> <span class=\"token function\">svm_migrate_gart_map<\/span><span class=\"token punctuation\">(<\/span><span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">amdgpu_ring<\/span> <span class=\"token operator\">*<\/span>ring<span class=\"token punctuation\">,<\/span><br \/>\n                               u64 npages<span class=\"token punctuation\">,<\/span><br \/>\n                               <span class=\"token class-name\">dma_addr_t<\/span> <span class=\"token operator\">*<\/span>addr<span class=\"token punctuation\">,<\/span>  <span class=\"token comment\">\/\/ \u7cfb\u7edf\u5185\u5b58DMA\u5730\u5740<\/span><br \/>\n                               u64 <span class=\"token operator\">*<\/span>gart_addr<span class=\"token punctuation\">,<\/span>    <span class=\"token comment\">\/\/ \u8f93\u51fa&#xff1a;GART\u865a\u62df\u5730\u5740<\/span><br \/>\n                               u64 flags<span class=\"token punctuation\">)<\/span><br \/>\n<span class=\"token punctuation\">{<\/span><br \/>\n    <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">amdgpu_device<\/span> <span class=\"token operator\">*<\/span>adev <span class=\"token operator\">&#061;<\/span> ring<span class=\"token operator\">-&gt;<\/span>adev<span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 1. \u4f7f\u7528GART\u7a97\u53e30&#xff08;\u9884\u7559\u7684\u6620\u5c04\u7a7a\u95f4&#xff09;<\/span><br \/>\n    <span class=\"token operator\">*<\/span>gart_addr <span class=\"token operator\">&#061;<\/span> adev<span class=\"token operator\">-&gt;<\/span>gmc<span class=\"token punctuation\">.<\/span>gart_start<span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 2. \u51c6\u5907PTE&#xff08;\u9875\u8868\u9879&#xff09;\u6807\u5fd7<\/span><br \/>\n    pte_flags <span class=\"token operator\">&#061;<\/span> AMDGPU_PTE_VALID <span class=\"token operator\">|<\/span> AMDGPU_PTE_READABLE<span class=\"token punctuation\">;<\/span><br \/>\n    pte_flags <span class=\"token operator\">|&#061;<\/span> AMDGPU_PTE_SYSTEM <span class=\"token operator\">|<\/span> AMDGPU_PTE_SNOOPED<span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token keyword\">if<\/span> <span class=\"token punctuation\">(<\/span><span class=\"token operator\">!<\/span><span class=\"token punctuation\">(<\/span>flags <span class=\"token operator\">&amp;<\/span> KFD_IOCTL_SVM_FLAG_GPU_RO<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><br \/>\n        pte_flags <span class=\"token operator\">|&#061;<\/span> AMDGPU_PTE_WRITEABLE<span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 3. \u5206\u914dJob&#xff08;GPU\u547d\u4ee4\u7f13\u51b2\u533a&#xff09;<\/span><br \/>\n    r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">amdgpu_job_alloc_with_ib<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">&amp;<\/span>adev<span class=\"token operator\">-&gt;<\/span>mman<span class=\"token punctuation\">.<\/span>high_pr<span class=\"token punctuation\">,<\/span><br \/>\n                                <span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token operator\">&amp;<\/span>job<span class=\"token punctuation\">,<\/span> <span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 4. \u751f\u6210GART\u66f4\u65b0\u547d\u4ee4<\/span><br \/>\n    <span class=\"token comment\">\/\/ \u5c06DMA\u5730\u5740\u6570\u7ec4\u8f6c\u6362\u4e3aPTE\u6761\u76ee<\/span><br \/>\n    cpu_addr <span class=\"token operator\">&#061;<\/span> <span class=\"token operator\">&amp;<\/span>job<span class=\"token operator\">-&gt;<\/span>ibs<span class=\"token punctuation\">[<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">.<\/span>ptr<span class=\"token punctuation\">[<\/span>num_dw<span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token function\">amdgpu_gart_map<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">,<\/span> npages<span class=\"token punctuation\">,<\/span> addr<span class=\"token punctuation\">,<\/span> pte_flags<span class=\"token punctuation\">,<\/span> cpu_addr<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 5. \u4f7f\u7528SDMA\u5c06PTE\u5199\u5165GART\u8868<\/span><br \/>\n    src_addr <span class=\"token operator\">&#061;<\/span> job<span class=\"token operator\">-&gt;<\/span>ibs<span class=\"token punctuation\">[<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">.<\/span>gpu_addr <span class=\"token operator\">&#043;<\/span> offset<span class=\"token punctuation\">;<\/span>  <span class=\"token comment\">\/\/ PTE\u6570\u636e\u7684GPU\u5730\u5740<\/span><br \/>\n    dst_addr <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">amdgpu_bo_gpu_offset<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token operator\">-&gt;<\/span>gart<span class=\"token punctuation\">.<\/span>bo<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span>  <span class=\"token comment\">\/\/ GART\u8868\u7684GPU\u5730\u5740<\/span><br \/>\n    <span class=\"token function\">amdgpu_emit_copy_buffer<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">&amp;<\/span>job<span class=\"token operator\">-&gt;<\/span>ibs<span class=\"token punctuation\">[<\/span><span class=\"token number\">0<\/span><span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">,<\/span><br \/>\n                           src_addr<span class=\"token punctuation\">,<\/span> dst_addr<span class=\"token punctuation\">,<\/span> num_bytes<span class=\"token punctuation\">,<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 6. \u63d0\u4ea4\u5e76\u83b7\u53d6fence<\/span><br \/>\n    fence <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">amdgpu_job_submit<\/span><span class=\"token punctuation\">(<\/span>job<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token function\">dma_fence_put<\/span><span class=\"token punctuation\">(<\/span>fence<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token keyword\">return<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><\/p>\n<p>\u56fe\u89e3GART\u6620\u5c04\u8fc7\u7a0b&#xff1a;<\/p>\n<p>\u6b65\u9aa41: \u51c6\u5907PTE\u6570\u636e<br \/>\n\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510<br \/>\n\u2502 CPU\u751f\u6210PTE\u6570\u7ec4          \u2502<br \/>\n\u2502 PTE[0] &#061; sys_addr[0]   \u2502<br \/>\n\u2502 PTE[1] &#061; sys_addr[1]   \u2502<br \/>\n\u2502 &#8230;                    \u2502<br \/>\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<br \/>\n            \u2193<br \/>\n\u6b65\u9aa42: SDMA\u5199\u5165GART\u8868<br \/>\n\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510  SDMA\u62f7\u8d1d  \u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510<br \/>\n\u2502 GPU\u5185\u5b58\u4e2d\u7684PTE\u6570\u636e      \u2502 \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2192  \u2502 GART\u8868&#xff08;\u5728VRAM&#xff09;        \u2502<br \/>\n\u2502 &#064; job-&gt;ibs[0].gpu_addr \u2502            \u2502 &#064; adev-&gt;gart.bo        \u2502<br \/>\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518            \u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<br \/>\n                                               \u2193<br \/>\n\u6b65\u9aa43: \u6620\u5c04\u751f\u6548<br \/>\n\u250c\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2510<br \/>\n\u2502 GART\u5730\u5740\u7a7a\u95f4                                             \u2502<br \/>\n\u2502 [0xA00000000-0xA00001000] \u2192 \u7cfb\u7edfRAM &#064; 0x123456000       \u2502<br \/>\n\u2502 [0xA00001000-0xA00002000] \u2192 \u7cfb\u7edfRAM &#064; 0x123457000       \u2502<br \/>\n\u2502 &#8230;                                                     \u2502<br \/>\n\u2514\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2518<\/p>\n<hr \/>\n<h3>7.4 \u4eceVRAM\u5230RAM\u7684\u8fc1\u79fb<\/h3>\n<h4>\u4e3b\u51fd\u6570&#xff1a;svm_migrate_vram_to_ram<\/h4>\n<p><span class=\"token comment\">\/\/ \u6587\u4ef6: kfd_migrate.c, \u884c: 800-920<\/span><br \/>\n<span class=\"token keyword\">int<\/span> <span class=\"token function\">svm_migrate_vram_to_ram<\/span><span class=\"token punctuation\">(<\/span><span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">svm_range<\/span> <span class=\"token operator\">*<\/span>prange<span class=\"token punctuation\">,<\/span><br \/>\n                           <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">mm_struct<\/span> <span class=\"token operator\">*<\/span>mm<span class=\"token punctuation\">,<\/span><br \/>\n                           <span class=\"token keyword\">unsigned<\/span> <span class=\"token keyword\">long<\/span> start_mgr<span class=\"token punctuation\">,<\/span><br \/>\n                           <span class=\"token keyword\">unsigned<\/span> <span class=\"token keyword\">long<\/span> last_mgr<span class=\"token punctuation\">,<\/span><br \/>\n                           <span class=\"token class-name\">uint32_t<\/span> trigger<span class=\"token punctuation\">,<\/span><br \/>\n                           <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">page<\/span> <span class=\"token operator\">*<\/span>fault_page<span class=\"token punctuation\">)<\/span><\/p>\n<p>\u5173\u952e\u7279\u6027&#xff1a;<\/p>\n<ul>\n<li>fault_page: CPU\u9875\u9762\u9519\u8bef\u65f6\u4f20\u5165&#xff0c;\u6307\u5b9a\u5177\u4f53\u7684\u6545\u969c\u9875<\/li>\n<li>\u4e0b\u884c\u8fc1\u79fb\u901a\u5e38\u7531CPU\u9875\u9762\u9519\u8bef\u89e6\u53d1<\/li>\n<\/ul>\n<h4>\u8fc1\u79fb\u6d41\u7a0b<\/h4>\n<p>1. \u68c0\u67e5actual_loc<br \/>\n   if (!prange-&gt;actual_loc)<br \/>\n       return 0;  \/\/ \u5df2\u7ecf\u5728RAM&#xff0c;\u65e0\u9700\u8fc1\u79fb<\/p>\n<p>2. \u6309VMA\u904d\u5386<br \/>\n   for each VMA:<br \/>\n       svm_migrate_vma_to_ram()<\/p>\n<p>3. \u66f4\u65b0\u9875\u9762\u8ba1\u6570<br \/>\n   prange-&gt;vram_pages -&#061; mpages<\/p>\n<p>4. \u91ca\u653eVRAM&#xff08;\u5982\u679c\u5168\u90e8\u8fc1\u79fb\u5b8c&#xff09;<br \/>\n   if (prange-&gt;vram_pages &#061;&#061; 0)<br \/>\n       svm_range_vram_node_free()<\/p>\n<h4>svm_migrate_vma_to_ram \u8be6\u89e3<\/h4>\n<p><span class=\"token keyword\">static<\/span> <span class=\"token keyword\">long<\/span> <span class=\"token function\">svm_migrate_vma_to_ram<\/span><span class=\"token punctuation\">(<\/span><span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">kfd_node<\/span> <span class=\"token operator\">*<\/span>node<span class=\"token punctuation\">,<\/span><br \/>\n                                  <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">svm_range<\/span> <span class=\"token operator\">*<\/span>prange<span class=\"token punctuation\">,<\/span><br \/>\n                                  <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">vm_area_struct<\/span> <span class=\"token operator\">*<\/span>vma<span class=\"token punctuation\">,<\/span><br \/>\n                                  <span class=\"token class-name\">uint64_t<\/span> start<span class=\"token punctuation\">,<\/span> <span class=\"token class-name\">uint64_t<\/span> end<span class=\"token punctuation\">,<\/span><br \/>\n                                  <span class=\"token class-name\">uint32_t<\/span> trigger<span class=\"token punctuation\">,<\/span><br \/>\n                                  <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">page<\/span> <span class=\"token operator\">*<\/span>fault_page<span class=\"token punctuation\">)<\/span><br \/>\n<span class=\"token punctuation\">{<\/span><br \/>\n    <span class=\"token comment\">\/\/ 1. \u51c6\u5907\u8fc1\u79fb\u4e0a\u4e0b\u6587<\/span><br \/>\n    <span class=\"token function\">memset<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">,<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token keyword\">sizeof<\/span><span class=\"token punctuation\">(<\/span>migrate<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>vma <span class=\"token operator\">&#061;<\/span> vma<span class=\"token punctuation\">;<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>start <span class=\"token operator\">&#061;<\/span> start<span class=\"token punctuation\">;<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>end <span class=\"token operator\">&#061;<\/span> end<span class=\"token punctuation\">;<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>pgmap_owner <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">SVM_ADEV_PGMAP_OWNER<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ \u9009\u62e9\u8bbe\u5907\u9875\u9762&#xff08;\u5728VRAM\u7684\u9875\u9762&#xff09;<\/span><br \/>\n    <span class=\"token keyword\">if<\/span> <span class=\"token punctuation\">(<\/span>adev<span class=\"token operator\">-&gt;<\/span>gmc<span class=\"token punctuation\">.<\/span>xgmi<span class=\"token punctuation\">.<\/span>connected_to_cpu<span class=\"token punctuation\">)<\/span><br \/>\n        migrate<span class=\"token punctuation\">.<\/span>flags <span class=\"token operator\">&#061;<\/span> MIGRATE_VMA_SELECT_DEVICE_COHERENT<span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token keyword\">else<\/span><br \/>\n        migrate<span class=\"token punctuation\">.<\/span>flags <span class=\"token operator\">&#061;<\/span> MIGRATE_VMA_SELECT_DEVICE_PRIVATE<span class=\"token punctuation\">;<\/span><\/p>\n<p>    migrate<span class=\"token punctuation\">.<\/span>fault_page <span class=\"token operator\">&#061;<\/span> fault_page<span class=\"token punctuation\">;<\/span>  <span class=\"token comment\">\/\/ \u6307\u5b9a\u6545\u969c\u9875<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 2. \u6536\u96c6VRAM\u9875\u9762<\/span><br \/>\n    r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">migrate_vma_setup<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    cpages <span class=\"token operator\">&#061;<\/span> migrate<span class=\"token punctuation\">.<\/span>cpages<span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 3. \u5206\u914dRAM\u9875\u9762\u5e76\u6267\u884cSDMA\u62f7\u8d1d<\/span><br \/>\n    r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_copy_to_ram<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span> prange<span class=\"token punctuation\">,<\/span> <span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">,<\/span><br \/>\n                               <span class=\"token operator\">&amp;<\/span>mfence<span class=\"token punctuation\">,<\/span> scratch<span class=\"token punctuation\">,<\/span> npages<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 4. \u66f4\u65b0\u9875\u8868<\/span><br \/>\n    <span class=\"token function\">migrate_vma_pages<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 5. \u7b49\u5f85\u5b8c\u6210<\/span><br \/>\n    <span class=\"token function\">svm_migrate_copy_done<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span> mfence<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token function\">migrate_vma_finalize<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token keyword\">return<\/span> mpages<span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><\/p>\n<h4>svm_migrate_copy_to_ram<\/h4>\n<p><span class=\"token keyword\">static<\/span> <span class=\"token keyword\">int<\/span> <span class=\"token function\">svm_migrate_copy_to_ram<\/span><span class=\"token punctuation\">(<\/span><span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">amdgpu_device<\/span> <span class=\"token operator\">*<\/span>adev<span class=\"token punctuation\">,<\/span><br \/>\n                                  <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">svm_range<\/span> <span class=\"token operator\">*<\/span>prange<span class=\"token punctuation\">,<\/span><br \/>\n                                  <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">migrate_vma<\/span> <span class=\"token operator\">*<\/span>migrate<span class=\"token punctuation\">,<\/span><br \/>\n                                  <span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">dma_fence<\/span> <span class=\"token operator\">*<\/span><span class=\"token operator\">*<\/span>mfence<span class=\"token punctuation\">,<\/span><br \/>\n                                  <span class=\"token class-name\">dma_addr_t<\/span> <span class=\"token operator\">*<\/span>scratch<span class=\"token punctuation\">,<\/span> u64 npages<span class=\"token punctuation\">)<\/span><br \/>\n<span class=\"token punctuation\">{<\/span><br \/>\n    <span class=\"token comment\">\/\/ 1. \u5206\u914d\u76ee\u6807RAM\u9875\u9762<\/span><br \/>\n    <span class=\"token keyword\">for<\/span> <span class=\"token punctuation\">(<\/span>i <span class=\"token operator\">&#061;<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">,<\/span> j <span class=\"token operator\">&#061;<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">;<\/span> i <span class=\"token operator\">&lt;<\/span> npages<span class=\"token punctuation\">;<\/span> i<span class=\"token operator\">&#043;&#043;<\/span><span class=\"token punctuation\">)<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n        <span class=\"token keyword\">if<\/span> <span class=\"token punctuation\">(<\/span>migrate<span class=\"token operator\">-&gt;<\/span>src<span class=\"token punctuation\">[<\/span>i<span class=\"token punctuation\">]<\/span> <span class=\"token operator\">&amp;<\/span> MIGRATE_PFN_VALID<span class=\"token punctuation\">)<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n            <span class=\"token comment\">\/\/ \u5206\u914d\u65b0\u9875\u9762<\/span><br \/>\n            dst_page <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">alloc_page_vma<\/span><span class=\"token punctuation\">(<\/span>GFP_HIGHUSER<span class=\"token punctuation\">,<\/span> vma<span class=\"token punctuation\">,<\/span> addr<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>            <span class=\"token comment\">\/\/ \u6620\u5c04\u5230\u8bbe\u5907\u53ef\u89c1\u5730\u5740<\/span><br \/>\n            r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_copy_to_ram_map_page<\/span><span class=\"token punctuation\">(<\/span>dev<span class=\"token punctuation\">,<\/span> dst_page<span class=\"token punctuation\">,<\/span><br \/>\n                                                 <span class=\"token operator\">&amp;<\/span>scratch<span class=\"token punctuation\">[<\/span>j<span class=\"token punctuation\">]<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>            <span class=\"token comment\">\/\/ \u8bb0\u5f55\u76ee\u6807\u9875\u9762<\/span><br \/>\n            migrate<span class=\"token operator\">-&gt;<\/span>dst<span class=\"token punctuation\">[<\/span>i<span class=\"token punctuation\">]<\/span> <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">migrate_pfn<\/span><span class=\"token punctuation\">(<\/span><span class=\"token function\">page_to_pfn<\/span><span class=\"token punctuation\">(<\/span>dst_page<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>            <span class=\"token comment\">\/\/ \u51c6\u5907\u6e90VRAM\u5730\u5740<\/span><br \/>\n            src<span class=\"token punctuation\">[<\/span>j<span class=\"token punctuation\">]<\/span> <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_direct_mapping_addr<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span><br \/>\n                        prange<span class=\"token operator\">-&gt;<\/span>start <span class=\"token operator\">&#043;<\/span> i <span class=\"token operator\">&#8211;<\/span> prange<span class=\"token operator\">-&gt;<\/span>offset<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n            j<span class=\"token operator\">&#043;&#043;<\/span><span class=\"token punctuation\">;<\/span><br \/>\n        <span class=\"token punctuation\">}<\/span><br \/>\n    <span class=\"token punctuation\">}<\/span><\/p>\n<p>    <span class=\"token comment\">\/\/ 2. SDMA\u62f7\u8d1d: VRAM \u2192 RAM<\/span><br \/>\n    r <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_copy_memory_gart<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span> scratch<span class=\"token punctuation\">,<\/span> src<span class=\"token punctuation\">,<\/span> npages<span class=\"token punctuation\">,<\/span><br \/>\n                                    FROM_VRAM_TO_RAM<span class=\"token punctuation\">,<\/span> mfence<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>    <span class=\"token keyword\">return<\/span> r<span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><\/p>\n<hr \/>\n<h3>7.5 \u8fc1\u79fb\u4f18\u5316\u6280\u672f<\/h3>\n<h4>7.5.1 \u5206\u6279\u5904\u7406<\/h4>\n<p><span class=\"token comment\">\/\/ \u907f\u514d\u4e00\u6b21\u8fc1\u79fb\u8fc7\u5927\u7684\u8303\u56f4<\/span><br \/>\n<span class=\"token keyword\">const<\/span> u64 GTT_MAX_PAGES <span class=\"token operator\">&#061;<\/span> AMDGPU_GTT_MAX_TRANSFER_SIZE<span class=\"token punctuation\">;<\/span>  <span class=\"token comment\">\/\/ 256MB<\/span><\/p>\n<p><span class=\"token keyword\">while<\/span> <span class=\"token punctuation\">(<\/span>npages<span class=\"token punctuation\">)<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n    size <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">min<\/span><span class=\"token punctuation\">(<\/span>GTT_MAX_PAGES<span class=\"token punctuation\">,<\/span> npages<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    <span class=\"token comment\">\/\/ \u8fc1\u79fb size \u9875<\/span><br \/>\n    npages <span class=\"token operator\">-&#061;<\/span> size<span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><\/p>\n<p>\u539f\u56e0&#xff1a;<\/p>\n<ul>\n<li>GART\u7a97\u53e3\u6709\u9650&#xff08;\u901a\u5e38256MB&#xff09;<\/li>\n<li>\u907f\u514d\u957f\u65f6\u95f4\u9501\u5b9agtt_window_lock<\/li>\n<li>\u5141\u8bb8\u5176\u4ed6\u64cd\u4f5c\u7a7f\u63d2\u6267\u884c<\/li>\n<\/ul>\n<h4>7.5.2 \u90e8\u5206\u8fc1\u79fb<\/h4>\n<p><span class=\"token comment\">\/\/ migrate_vma_setup \u53ef\u80fd\u53ea\u6536\u96c6\u90e8\u5206\u9875\u9762<\/span><br \/>\n<span class=\"token keyword\">if<\/span> <span class=\"token punctuation\">(<\/span>cpages <span class=\"token operator\">!&#061;<\/span> npages<span class=\"token punctuation\">)<\/span><br \/>\n    <span class=\"token function\">pr_debug<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">&#034;partial migration, 0x%lx\/0x%llx pages\\\\n&#034;<\/span><span class=\"token punctuation\">,<\/span><br \/>\n             cpages<span class=\"token punctuation\">,<\/span> npages<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p>\u5e38\u89c1\u539f\u56e0&#xff1a;<\/p>\n<ul>\n<li>\u9875\u9762\u88abpin_user_pages\u9501\u5b9a&#xff08;RDMA\u3001GPU\u76f4\u63a5\u8bbf\u95ee&#xff09;<\/li>\n<li>\u9875\u9762\u6b63\u5728\u88ab\u5176\u4ed6\u64cd\u4f5c\u4f7f\u7528<\/li>\n<li>\u9875\u9762\u662f\u7279\u6b8a\u9875&#xff08;huge pages, VM_SPECIAL&#xff09;<\/li>\n<\/ul>\n<h4>7.5.3 \u5f02\u6b65\u6267\u884c<\/h4>\n<p><span class=\"token comment\">\/\/ SDMA\u64cd\u4f5c\u8fd4\u56defence&#xff0c;\u5141\u8bb8\u5f02\u6b65\u6267\u884c<\/span><br \/>\n<span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">dma_fence<\/span> <span class=\"token operator\">*<\/span>mfence <span class=\"token operator\">&#061;<\/span> <span class=\"token constant\">NULL<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p><span class=\"token comment\">\/\/ \u63d0\u4ea4SDMA\u4efb\u52a1&#xff08;\u7acb\u5373\u8fd4\u56de&#xff09;<\/span><br \/>\n<span class=\"token function\">svm_migrate_copy_memory_gart<\/span><span class=\"token punctuation\">(<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">,<\/span> <span class=\"token operator\">&amp;<\/span>mfence<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p><span class=\"token comment\">\/\/ \u7ee7\u7eed\u5176\u4ed6\u5de5\u4f5c&#8230;<\/span><br \/>\n<span class=\"token function\">migrate_vma_pages<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>migrate<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p><span class=\"token comment\">\/\/ \u9700\u8981\u65f6\u7b49\u5f85\u5b8c\u6210<\/span><br \/>\n<span class=\"token function\">svm_migrate_copy_done<\/span><span class=\"token punctuation\">(<\/span>adev<span class=\"token punctuation\">,<\/span> mfence<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<h4>7.5.4 XGMI\u4f18\u5316<\/h4>\n<p><span class=\"token comment\">\/\/ \u5bf9\u4e8eXGMI\u8fde\u63a5\u7684GPU&#xff0c;\u53ef\u4ee5\u76f4\u63a5\u8bbf\u95ee<\/span><br \/>\n<span class=\"token keyword\">if<\/span> <span class=\"token punctuation\">(<\/span>adev<span class=\"token operator\">-&gt;<\/span>gmc<span class=\"token punctuation\">.<\/span>xgmi<span class=\"token punctuation\">.<\/span>connected_to_cpu<span class=\"token punctuation\">)<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n    <span class=\"token comment\">\/\/ \u4f7f\u7528\u8bbe\u5907\u4e00\u81f4\u6027\u5185\u5b58<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>flags <span class=\"token operator\">&#061;<\/span> MIGRATE_VMA_SELECT_DEVICE_COHERENT<span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span> <span class=\"token keyword\">else<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n    <span class=\"token comment\">\/\/ \u4f7f\u7528\u8bbe\u5907\u79c1\u6709\u5185\u5b58<\/span><br \/>\n    migrate<span class=\"token punctuation\">.<\/span>flags <span class=\"token operator\">&#061;<\/span> MIGRATE_VMA_SELECT_DEVICE_PRIVATE<span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><\/p>\n<p>XGMI&#xff08;AMD Infinity Fabric&#xff09;&#xff1a;<\/p>\n<ul>\n<li>CPU\u53ef\u4ee5\u76f4\u63a5\u8bbf\u95eeVRAM<\/li>\n<li>\u65e0\u9700\u8fc1\u79fb\u5373\u53ef\u8bbf\u95ee<\/li>\n<li>\u6027\u80fd\u63a5\u8fd1\u672c\u5730\u5185\u5b58<\/li>\n<\/ul>\n<hr \/>\n<h3>7.6 \u8fc1\u79fb\u89e6\u53d1\u573a\u666f<\/h3>\n<h4>\u89e6\u53d1\u539f\u56e0&#xff08;trigger\u53c2\u6570&#xff09;<\/h4>\n<p><span class=\"token keyword\">enum<\/span> <span class=\"token class-name\">KFD_MIGRATE_TRIGGERS<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n    KFD_MIGRATE_TRIGGER_PREFETCH<span class=\"token punctuation\">,<\/span>     <span class=\"token comment\">\/\/ \u7528\u6237\u9884\u53d6<\/span><br \/>\n    KFD_MIGRATE_TRIGGER_PAGEFAULT_GPU<span class=\"token punctuation\">,<\/span><span class=\"token comment\">\/\/ GPU\u9875\u9762\u9519\u8bef<\/span><br \/>\n    KFD_MIGRATE_TRIGGER_PAGEFAULT_CPU<span class=\"token punctuation\">,<\/span><span class=\"token comment\">\/\/ CPU\u9875\u9762\u9519\u8bef<\/span><br \/>\n    KFD_MIGRATE_TRIGGER_TTM_EVICTION<span class=\"token punctuation\">,<\/span> <span class=\"token comment\">\/\/ TTM\u9a71\u9010&#xff08;\u5185\u5b58\u4e0d\u8db3&#xff09;<\/span><br \/>\n    KFD_MIGRATE_TRIGGER_SUSPEND<span class=\"token punctuation\">,<\/span>      <span class=\"token comment\">\/\/ \u7cfb\u7edf\u6302\u8d77<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<h4>\u89e6\u53d1\u6d41\u7a0b\u793a\u4f8b<\/h4>\n<p>\u573a\u666f1&#xff1a;GPU\u9875\u9762\u9519\u8bef<\/p>\n<p>1. GPU\u8bbf\u95ee\u5730\u57400x1234<br \/>\n   \u2193<br \/>\n2. \u5730\u5740\u672a\u6620\u5c04&#xff0c;\u89e6\u53d1GPU\u9875\u9762\u9519\u8bef<br \/>\n   \u2193<br \/>\n3. XNACK\u91cd\u8bd5\u673a\u5236\u6682\u505cGPU\u6267\u884c<br \/>\n   \u2193<br \/>\n4. GPU\u4e2d\u65adCPU<br \/>\n   \u2193<br \/>\n5. \u4e2d\u65ad\u5904\u7406\u7a0b\u5e8f\u8c03\u7528&#xff1a;<br \/>\n   svm_range_restore_pages()<br \/>\n   \u2193<br \/>\n6. \u68c0\u67e5\u9875\u9762\u4f4d\u7f6e&#xff1a;<br \/>\n   if (\u5728RAM)<br \/>\n       \u2192 \u53ea\u9700\u5efa\u7acb\u6620\u5c04<br \/>\n   else<br \/>\n       \u2192 \u8c03\u7528svm_migrate_ram_to_vram()<br \/>\n   \u2193<br \/>\n7. \u8fc1\u79fb\u5b8c\u6210&#xff0c;\u66f4\u65b0GPU\u9875\u8868<br \/>\n   \u2193<br \/>\n8. XNACK\u6062\u590dGPU\u6267\u884c<\/p>\n<p>\u573a\u666f2&#xff1a;CPU\u9875\u9762\u9519\u8bef<\/p>\n<p>1. CPU\u8bbf\u95eeVRAM\u5730\u57400x5678<br \/>\n   \u2193<br \/>\n2. \u9875\u8868\u9879\u6807\u8bb0\u4e3a\u8bbe\u5907\u79c1\u6709&#xff0c;\u89e6\u53d1\u9875\u9762\u9519\u8bef<br \/>\n   \u2193<br \/>\n3. \u5185\u6838\u9875\u9762\u9519\u8bef\u5904\u7406\u7a0b\u5e8f\u8c03\u7528&#xff1a;<br \/>\n   vm_ops-&gt;fault()<br \/>\n   \u2193<br \/>\n4. AMDGPU\u9a71\u52a8\u6ce8\u518c\u7684\u5904\u7406\u51fd\u6570&#xff1a;<br \/>\n   svm_migrate_to_ram()<br \/>\n   \u2193<br \/>\n5. \u8c03\u7528svm_migrate_vram_to_ram()<br \/>\n   \u53c2\u6570: fault_page &#061; vmf-&gt;page<br \/>\n   \u2193<br \/>\n6. \u8fc1\u79fb\u5b8c\u6210&#xff0c;\u66f4\u65b0CPU\u9875\u8868<br \/>\n   \u2193<br \/>\n7. CPU\u91cd\u65b0\u6267\u884c\u8bbf\u95ee\u6307\u4ee4<\/p>\n<p>\u573a\u666f3&#xff1a;\u7528\u6237\u9884\u53d6<\/p>\n<p>\u7528\u6237\u7a0b\u5e8f:<br \/>\n    hipMemPrefetchAsync(ptr, size, deviceId);<br \/>\n    \u2193<br \/>\n\u5185\u6838\u5904\u7406:<br \/>\n    svm_range_prefault()<br \/>\n    \u2193<br \/>\n    svm_migrate_ram_to_vram(trigger&#061;PREFETCH)<\/p>\n<hr \/>\n<h3>7.7 \u4ee3\u7801\u793a\u4f8b<\/h3>\n<h4>\u5b8c\u6574\u8fc1\u79fb\u793a\u4f8b<\/h4>\n<p><span class=\"token comment\">\/\/ \u5c06\u4e00\u4e2a\u8303\u56f4\u8fc1\u79fb\u5230GPU 0<\/span><br \/>\n<span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">kfd_process<\/span> <span class=\"token operator\">*<\/span>p <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">kfd_get_process<\/span><span class=\"token punctuation\">(<\/span>current<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">svm_range<\/span> <span class=\"token operator\">*<\/span>prange <span class=\"token operator\">&#061;<\/span> <span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token class-name\">uint32_t<\/span> gpu_id <span class=\"token operator\">&#061;<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p><span class=\"token comment\">\/\/ \u9501\u5b9a\u5fc5\u8981\u7684\u9501<\/span><br \/>\n<span class=\"token function\">mmap_read_lock<\/span><span class=\"token punctuation\">(<\/span>p<span class=\"token operator\">-&gt;<\/span>mm<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token function\">mutex_lock<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>prange<span class=\"token operator\">-&gt;<\/span>migrate_mutex<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p><span class=\"token comment\">\/\/ \u6267\u884c\u8fc1\u79fb<\/span><br \/>\nr <span class=\"token operator\">&#061;<\/span> <span class=\"token function\">svm_migrate_ram_to_vram<\/span><span class=\"token punctuation\">(<\/span>prange<span class=\"token punctuation\">,<\/span><br \/>\n                           gpu_id<span class=\"token punctuation\">,<\/span><br \/>\n                           prange<span class=\"token operator\">-&gt;<\/span>start<span class=\"token punctuation\">,<\/span>  <span class=\"token comment\">\/\/ \u6574\u4e2a\u8303\u56f4<\/span><br \/>\n                           prange<span class=\"token operator\">-&gt;<\/span>last<span class=\"token punctuation\">,<\/span><br \/>\n                           p<span class=\"token operator\">-&gt;<\/span>mm<span class=\"token punctuation\">,<\/span><br \/>\n                           KFD_MIGRATE_TRIGGER_PREFETCH<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p><span class=\"token keyword\">if<\/span> <span class=\"token punctuation\">(<\/span>r <span class=\"token operator\">&gt;<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">)<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n    <span class=\"token function\">pr_info<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">&#034;migrated %d pages to GPU %u\\\\n&#034;<\/span><span class=\"token punctuation\">,<\/span> r<span class=\"token punctuation\">,<\/span> gpu_id<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n    prange<span class=\"token operator\">-&gt;<\/span>actual_loc <span class=\"token operator\">&#061;<\/span> gpu_id<span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span> <span class=\"token keyword\">else<\/span> <span class=\"token keyword\">if<\/span> <span class=\"token punctuation\">(<\/span>r <span class=\"token operator\">&lt;<\/span> <span class=\"token number\">0<\/span><span class=\"token punctuation\">)<\/span> <span class=\"token punctuation\">{<\/span><br \/>\n    <span class=\"token function\">pr_err<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">&#034;migration failed: %d\\\\n&#034;<\/span><span class=\"token punctuation\">,<\/span> r<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token punctuation\">}<\/span><\/p>\n<p><span class=\"token function\">mutex_unlock<\/span><span class=\"token punctuation\">(<\/span><span class=\"token operator\">&amp;<\/span>prange<span class=\"token operator\">-&gt;<\/span>migrate_mutex<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token function\">mmap_read_unlock<\/span><span class=\"token punctuation\">(<\/span>p<span class=\"token operator\">-&gt;<\/span>mm<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<h4>\u76d1\u63a7\u8fc1\u79fb\u7edf\u8ba1<\/h4>\n<p><span class=\"token comment\">\/\/ \u6bcf\u4e2aGPU\u6709\u72ec\u7acb\u7684\u7edf\u8ba1<\/span><br \/>\n<span class=\"token keyword\">struct<\/span> <span class=\"token class-name\">kfd_process_device<\/span> <span class=\"token operator\">*<\/span>pdd <span class=\"token operator\">&#061;<\/span> <span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">.<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<p><span class=\"token function\">pr_info<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">&#034;GPU %u statistics:\\\\n&#034;<\/span><span class=\"token punctuation\">,<\/span> pdd<span class=\"token operator\">-&gt;<\/span>dev<span class=\"token operator\">-&gt;<\/span>id<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token function\">pr_info<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">&#034;  Pages migrated in:  %llu\\\\n&#034;<\/span><span class=\"token punctuation\">,<\/span> pdd<span class=\"token operator\">-&gt;<\/span>page_in<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><br \/>\n<span class=\"token function\">pr_info<\/span><span class=\"token punctuation\">(<\/span><span class=\"token string\">&#034;  Pages migrated out: %llu\\\\n&#034;<\/span><span class=\"token punctuation\">,<\/span> pdd<span class=\"token operator\">-&gt;<\/span>page_out<span class=\"token punctuation\">)<\/span><span class=\"token punctuation\">;<\/span><\/p>\n<hr \/>\n<h3>&#x1f4a1; \u91cd\u70b9\u63d0\u793a<\/h3>\n<li>\n<p>GART\u662f\u5173\u952e&#xff1a;\u7406\u89e3GART\u6620\u5c04\u673a\u5236\u662f\u7406\u89e3\u8fc1\u79fb\u7684\u57fa\u7840\u3002<\/p>\n<\/li>\n<li>\n<p>\u4e24\u9636\u6bb5\u63d0\u4ea4&#xff1a;<\/p>\n<ul>\n<li>SDMA\u62f7\u8d1d\u6570\u636e&#xff08;\u5f02\u6b65&#xff09;<\/li>\n<li>migrate_vma_pages \u66f4\u65b0\u9875\u8868&#xff08;\u540c\u6b65&#xff09;<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u90e8\u5206\u8fc1\u79fb\u662f\u5e38\u6001&#xff1a;\u4e0d\u8981\u671f\u671b\u6240\u6709\u9875\u9762\u90fd\u80fd\u8fc1\u79fb\u3002<\/p>\n<\/li>\n<li>\n<p>\u9501\u987a\u5e8f&#xff1a;mmap_lock \u2192 migrate_mutex&#xff0c;\u907f\u514d\u6b7b\u9501\u3002<\/p>\n<\/li>\n<li>\n<p>\u5f02\u6b65\u64cd\u4f5c&#xff1a;SDMA\u8fd4\u56defence&#xff0c;\u5141\u8bb8\u5e76\u53d1\u6267\u884c\u3002<\/p>\n<\/li>\n<hr \/>\n<h3>\u26a0\ufe0f \u5e38\u89c1\u9677\u9631<\/h3>\n<p>\u274c \u9677\u96311&#xff1a;\u201c\u5fd8\u8bb0\u7b49\u5f85SDMA\u5b8c\u6210\u201d<\/p>\n<ul>\n<li>\u2705 \u6b63\u786e&#xff1a;\u8c03\u7528svm_migrate_copy_done()\u7b49\u5f85fence\u3002<\/li>\n<\/ul>\n<p>\u274c \u9677\u96312&#xff1a;\u201c\u5927\u8303\u56f4\u8fc1\u79fb\u4e0d\u5206\u6279\u201d<\/p>\n<ul>\n<li>\u2705 \u6b63\u786e&#xff1a;\u4f7f\u7528GTT_MAX_PAGES\u5206\u6279\u5904\u7406\u3002<\/li>\n<\/ul>\n<p>\u274c \u9677\u96313&#xff1a;\u201c\u5047\u8bbe\u6240\u6709\u9875\u9762\u90fd\u80fd\u8fc1\u79fb\u201d<\/p>\n<ul>\n<li>\u2705 \u6b63\u786e&#xff1a;\u68c0\u67e5cpages\u548cmpages&#xff0c;\u5904\u7406\u90e8\u5206\u5931\u8d25\u3002<\/li>\n<\/ul>\n<p>\u274c \u9677\u96314&#xff1a;\u201c\u5728\u4e2d\u65ad\u4e0a\u4e0b\u6587\u8c03\u7528\u8fc1\u79fb\u201d<\/p>\n<ul>\n<li>\u2705 \u6b63\u786e&#xff1a;\u8fc1\u79fb\u9700\u8981\u8fdb\u7a0b\u4e0a\u4e0b\u6587&#xff08;\u53ef\u80fd\u7761\u7720&#xff09;\u3002<\/li>\n<\/ul>\n<hr \/>\n<h3>&#x1f4dd; \u5b9e\u8df5\u7ec3\u4e60<\/h3>\n<li>\n<p>\u8ffd\u8e2a\u8fc1\u79fb\u8def\u5f84&#xff1a;<\/p>\n<p> <span class=\"token comment\"># \u542f\u7528\u8fc1\u79fb\u8c03\u8bd5<\/span><br \/>\n<span class=\"token builtin class-name\">echo<\/span> <span class=\"token string\">&#039;file kfd_migrate.c &#043;p&#039;<\/span> <span class=\"token operator\">&gt;<\/span> \/sys\/kernel\/debug\/dynamic_debug\/control<\/p>\n<p><span class=\"token comment\"># \u8fd0\u884cGPU\u7a0b\u5e8f\u5e76\u89c2\u5bdf\u8fc1\u79fb<\/span><br \/>\n<span class=\"token function\">dmesg<\/span> <span class=\"token operator\">|<\/span> <span class=\"token function\">grep<\/span> <span class=\"token string\">&#034;svm_migrate&#034;<\/span>\n <\/li>\n<li>\n<p>\u67e5\u770b\u8fc1\u79fb\u7edf\u8ba1&#xff1a;<\/p>\n<p> <span class=\"token comment\"># \u67e5\u770b\u8fdb\u7a0b\u7684SVM\u7edf\u8ba1<\/span><br \/>\n<span class=\"token function\">cat<\/span> \/sys\/kernel\/debug\/kfd\/proc\/<span class=\"token operator\">&lt;<\/span>pid<span class=\"token operator\">&gt;<\/span>\/svm_ranges\n <\/li>\n<li>\n<p>\u601d\u8003\u9898&#xff1a;<\/p>\n<ul>\n<li>\u4e3a\u4ec0\u4e48\u9700\u8981GART\u8868&#xff1f;\u4e0d\u80fd\u76f4\u63a5\u62f7\u8d1d\u5417&#xff1f;<\/li>\n<li>CPU\u9875\u9762\u9519\u8bef\u65f6\u4e3a\u4ec0\u4e48\u8981\u4f20\u5165fault_page&#xff1f;<\/li>\n<li>\u5982\u679c\u8fc1\u79fb\u8fc7\u7a0b\u4e2d\u9875\u9762\u88ab\u4fee\u6539\u4f1a\u600e\u6837&#xff1f;<\/li>\n<li>XGMI\u8fde\u63a5\u7684\u4f18\u52bf\u662f\u4ec0\u4e48&#xff1f;<\/li>\n<\/ul>\n<\/li>\n<li>\n<p>\u4ee3\u7801\u63a2\u7d22&#xff1a;<\/p>\n<p> <span class=\"token comment\"># \u67e5\u770bGART\u6620\u5c04\u5b9e\u73b0<\/span><br \/>\n<span class=\"token function\">grep<\/span> -A <span class=\"token number\">50<\/span> <span class=\"token string\">&#034;svm_migrate_gart_map&#034;<\/span> drivers\/gpu\/drm\/amd\/amdkfd\/kfd_migrate.c<\/p>\n<p><span class=\"token comment\"># \u67e5\u770bSDMA\u62f7\u8d1d\u547d\u4ee4<\/span><br \/>\n<span class=\"token function\">grep<\/span> -A <span class=\"token number\">20<\/span> <span class=\"token string\">&#034;amdgpu_emit_copy_buffer&#034;<\/span> drivers\/gpu\/drm\/amd\/amdgpu\/amdgpu_job.c\n <\/li>\n<hr \/>\n<h3>&#x1f4da; \u672c\u7ae0\u5c0f\u7ed3<\/h3>\n<ul>\n<li>\u4e24\u4e2a\u65b9\u5411&#xff1a;RAM\u2192VRAM&#xff08;\u4e0a\u884c&#xff09;\u548cVRAM\u2192RAM&#xff08;\u4e0b\u884c&#xff09;<\/li>\n<li>\u6838\u5fc3API&#xff1a;HMM\u6846\u67b6\u7684migrate_vma_xxx\u7cfb\u5217\u51fd\u6570<\/li>\n<li>GART\u6620\u5c04&#xff1a;\u8ba9SDMA\u80fd\u591f\u8bbf\u95ee\u7cfb\u7edfRAM<\/li>\n<li>SDMA\u62f7\u8d1d&#xff1a;GPU\u7684DMA\u5f15\u64ce\u6267\u884c\u5b9e\u9645\u6570\u636e\u642c\u8fd0<\/li>\n<li>\u5f02\u6b65\u6267\u884c&#xff1a;\u4f7f\u7528fence\u673a\u5236\u5141\u8bb8\u5e76\u53d1<\/li>\n<li>\u5206\u6279\u5904\u7406&#xff1a;\u907f\u514d\u957f\u65f6\u95f4\u5360\u7528\u8d44\u6e90<\/li>\n<li>\u90e8\u5206\u8fc1\u79fb&#xff1a;\u7075\u6d3b\u5904\u7406\u4e0d\u53ef\u8fc1\u79fb\u7684\u9875\u9762<\/li>\n<\/ul>\n<hr \/>\n<h3>&#x1f4d6; \u6269\u5c55\u9605\u8bfb<\/h3>\n<ul>\n<li>AMDGPU\u7684GART\u8bbe\u8ba1\u4e0e\u5b9e\u73b0\u5206\u6790<\/li>\n<li>AMD KFD\u7684BO\u8bbe\u8ba1\u5206\u6790<\/li>\n<\/ul>\n<hr \/>\n<h3>\u27a1\ufe0f \u4e0b\u4e00\u6b65<\/h3>\n<p>\u638c\u63e1\u4e86\u8fc1\u79fb\u673a\u5236\u540e&#xff0c;\u4e0b\u4e00\u7ae0\u6211\u4eec\u5c06\u5b66\u4e60\u9875\u9762\u6620\u5c04\u4e0eGPU\u9875\u8868\u2014\u2014\u5982\u4f55\u8ba9GPU\u80fd\u591f\u8bbf\u95ee\u8fd9\u4e9b\u9875\u9762\u3002<\/p>\n<hr \/>\n<p>&#x1f517; \u5bfc\u822a<\/p>\n<ul>\n<li>\u4e0a\u4e00\u7ae0&#xff1a;06 &#8211; SVM\u8303\u56f4\u7ba1\u7406<\/li>\n<li>\u4e0b\u4e00\u7ae0&#xff1a;08 &#8211; \u9875\u9762\u6620\u5c04\u4e0eGPU\u9875\u8868<\/li>\n<li>\u8fd4\u56de\u76ee\u5f55: AMD ROCm-SVM\u6280\u672f\u7684\u5b9e\u73b0\u4e0e\u5e94\u7528\u6df1\u5ea6\u5206\u6790\u76ee\u5f55<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>\u96be\u5ea6: &#x1f534;&#x1f534; \u9ad8\u7ea7 \u9884\u8ba1\u5b66\u4e60\u65f6\u95f4: 2.5-3\u5c0f\u65f6 \u524d\u7f6e\u77e5\u8bc6: \u7b2c6\u7ae0\u3001DMA\u57fa\u7840\u3001SDMA\/TTM\u6982\u5ff5 &#x1f4cb; \u6982\u8ff0<br \/>\n\u5185\u5b58\u8fc1\u79fb\u662fSVM\u7684\u6838\u5fc3\u7279\u6027&#xff0c;\u8d1f\u8d23\u5728\u7cfb\u7edfRAM\u548cGPU VRAM\u4e4b\u95f4\u81ea\u52a8\u79fb\u52a8\u9875\u9762\u3002\u6253\u4e2a\u6bd4\u65b9&#xff0c;\u8fc1\u79fb\u5c31\u50cf\u5728\u4e24\u4e2a\u4ed3\u5e93\u95f4\u642c\u8d27\u7269&#xff1a;<br \/>\n&#x1f3e2; \u7cfb\u7edfRAM: CPU\u7684\u672c\u5730\u4ed3\u5e93&#xff0c;\u8bbf\u95ee\u5feb\u4f46GPU\u8bbf\u95ee\u6162&#x1f3ed; GPU VRAM: GPU\u7684\u672c\u5730\u4ed3\u5e93&amp;#xf<\/p>\n","protected":false},"author":2,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[1753,7641,7642,8007,7640],"topic":[],"class_list":["post-75371","post","type-post","status-publish","format-standard","hentry","category-server","tag-rocm","tag-amdgpu","tag-kfd","tag-rocr","tag-svm"],"yoast_head":"<!-- This site is optimized with the Yoast SEO plugin v20.3 - https:\/\/yoast.com\/wordpress\/plugins\/seo\/ -->\n<title>07 - SVM\u5185\u5b58\u8fc1\u79fb\u673a\u5236 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3<\/title>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/www.wsisp.com\/helps\/75371.html\" \/>\n<meta property=\"og:locale\" content=\"zh_CN\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"07 - SVM\u5185\u5b58\u8fc1\u79fb\u673a\u5236 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"og:description\" content=\"\u96be\u5ea6: &#x1f534;&#x1f534; \u9ad8\u7ea7 \u9884\u8ba1\u5b66\u4e60\u65f6\u95f4: 2.5-3\u5c0f\u65f6 \u524d\u7f6e\u77e5\u8bc6: \u7b2c6\u7ae0\u3001DMA\u57fa\u7840\u3001SDMA\/TTM\u6982\u5ff5 &#x1f4cb; \u6982\u8ff0 \u5185\u5b58\u8fc1\u79fb\u662fSVM\u7684\u6838\u5fc3\u7279\u6027&#xff0c;\u8d1f\u8d23\u5728\u7cfb\u7edfRAM\u548cGPU VRAM\u4e4b\u95f4\u81ea\u52a8\u79fb\u52a8\u9875\u9762\u3002\u6253\u4e2a\u6bd4\u65b9&#xff0c;\u8fc1\u79fb\u5c31\u50cf\u5728\u4e24\u4e2a\u4ed3\u5e93\u95f4\u642c\u8d27\u7269&#xff1a; &#x1f3e2; \u7cfb\u7edfRAM: CPU\u7684\u672c\u5730\u4ed3\u5e93&#xff0c;\u8bbf\u95ee\u5feb\u4f46GPU\u8bbf\u95ee\u6162&#x1f3ed; GPU VRAM: GPU\u7684\u672c\u5730\u4ed3\u5e93&amp;#xf\" \/>\n<meta property=\"og:url\" content=\"https:\/\/www.wsisp.com\/helps\/75371.html\" \/>\n<meta property=\"og:site_name\" content=\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\" \/>\n<meta property=\"article:published_time\" content=\"2026-02-11T23:33:19+00:00\" \/>\n<meta name=\"author\" content=\"admin\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:label1\" content=\"\u4f5c\u8005\" \/>\n\t<meta name=\"twitter:data1\" content=\"admin\" \/>\n\t<meta name=\"twitter:label2\" content=\"\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4\" \/>\n\t<meta name=\"twitter:data2\" content=\"10 \u5206\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\/\/schema.org\",\"@graph\":[{\"@type\":\"WebPage\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/75371.html\",\"url\":\"https:\/\/www.wsisp.com\/helps\/75371.html\",\"name\":\"07 - SVM\u5185\u5b58\u8fc1\u79fb\u673a\u5236 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"isPartOf\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\"},\"datePublished\":\"2026-02-11T23:33:19+00:00\",\"dateModified\":\"2026-02-11T23:33:19+00:00\",\"author\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\"},\"breadcrumb\":{\"@id\":\"https:\/\/www.wsisp.com\/helps\/75371.html#breadcrumb\"},\"inLanguage\":\"zh-Hans\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\/\/www.wsisp.com\/helps\/75371.html\"]}]},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/75371.html#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"\u9996\u9875\",\"item\":\"https:\/\/www.wsisp.com\/helps\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"07 - SVM\u5185\u5b58\u8fc1\u79fb\u673a\u5236\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#website\",\"url\":\"https:\/\/www.wsisp.com\/helps\/\",\"name\":\"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3\",\"description\":\"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}\"},\"query-input\":\"required name=search_term_string\"}],\"inLanguage\":\"zh-Hans\"},{\"@type\":\"Person\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41\",\"name\":\"admin\",\"image\":{\"@type\":\"ImageObject\",\"inLanguage\":\"zh-Hans\",\"@id\":\"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/\",\"url\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"contentUrl\":\"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery\",\"caption\":\"admin\"},\"sameAs\":[\"http:\/\/wp.wsisp.com\"],\"url\":\"https:\/\/www.wsisp.com\/helps\/author\/admin\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"07 - SVM\u5185\u5b58\u8fc1\u79fb\u673a\u5236 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/www.wsisp.com\/helps\/75371.html","og_locale":"zh_CN","og_type":"article","og_title":"07 - SVM\u5185\u5b58\u8fc1\u79fb\u673a\u5236 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","og_description":"\u96be\u5ea6: &#x1f534;&#x1f534; \u9ad8\u7ea7 \u9884\u8ba1\u5b66\u4e60\u65f6\u95f4: 2.5-3\u5c0f\u65f6 \u524d\u7f6e\u77e5\u8bc6: \u7b2c6\u7ae0\u3001DMA\u57fa\u7840\u3001SDMA\/TTM\u6982\u5ff5 &#x1f4cb; \u6982\u8ff0 \u5185\u5b58\u8fc1\u79fb\u662fSVM\u7684\u6838\u5fc3\u7279\u6027&#xff0c;\u8d1f\u8d23\u5728\u7cfb\u7edfRAM\u548cGPU VRAM\u4e4b\u95f4\u81ea\u52a8\u79fb\u52a8\u9875\u9762\u3002\u6253\u4e2a\u6bd4\u65b9&#xff0c;\u8fc1\u79fb\u5c31\u50cf\u5728\u4e24\u4e2a\u4ed3\u5e93\u95f4\u642c\u8d27\u7269&#xff1a; &#x1f3e2; \u7cfb\u7edfRAM: CPU\u7684\u672c\u5730\u4ed3\u5e93&#xff0c;\u8bbf\u95ee\u5feb\u4f46GPU\u8bbf\u95ee\u6162&#x1f3ed; GPU VRAM: GPU\u7684\u672c\u5730\u4ed3\u5e93&amp;#xf","og_url":"https:\/\/www.wsisp.com\/helps\/75371.html","og_site_name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","article_published_time":"2026-02-11T23:33:19+00:00","author":"admin","twitter_card":"summary_large_image","twitter_misc":{"\u4f5c\u8005":"admin","\u9884\u8ba1\u9605\u8bfb\u65f6\u95f4":"10 \u5206"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"WebPage","@id":"https:\/\/www.wsisp.com\/helps\/75371.html","url":"https:\/\/www.wsisp.com\/helps\/75371.html","name":"07 - SVM\u5185\u5b58\u8fc1\u79fb\u673a\u5236 - \u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","isPartOf":{"@id":"https:\/\/www.wsisp.com\/helps\/#website"},"datePublished":"2026-02-11T23:33:19+00:00","dateModified":"2026-02-11T23:33:19+00:00","author":{"@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41"},"breadcrumb":{"@id":"https:\/\/www.wsisp.com\/helps\/75371.html#breadcrumb"},"inLanguage":"zh-Hans","potentialAction":[{"@type":"ReadAction","target":["https:\/\/www.wsisp.com\/helps\/75371.html"]}]},{"@type":"BreadcrumbList","@id":"https:\/\/www.wsisp.com\/helps\/75371.html#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"\u9996\u9875","item":"https:\/\/www.wsisp.com\/helps"},{"@type":"ListItem","position":2,"name":"07 - SVM\u5185\u5b58\u8fc1\u79fb\u673a\u5236"}]},{"@type":"WebSite","@id":"https:\/\/www.wsisp.com\/helps\/#website","url":"https:\/\/www.wsisp.com\/helps\/","name":"\u7f51\u7855\u4e92\u8054\u5e2e\u52a9\u4e2d\u5fc3","description":"\u9999\u6e2f\u670d\u52a1\u5668_\u9999\u6e2f\u4e91\u670d\u52a1\u5668\u8d44\u8baf_\u670d\u52a1\u5668\u5e2e\u52a9\u6587\u6863_\u670d\u52a1\u5668\u6559\u7a0b","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/www.wsisp.com\/helps\/?s={search_term_string}"},"query-input":"required name=search_term_string"}],"inLanguage":"zh-Hans"},{"@type":"Person","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/358e386c577a3ab51c4493330a20ad41","name":"admin","image":{"@type":"ImageObject","inLanguage":"zh-Hans","@id":"https:\/\/www.wsisp.com\/helps\/#\/schema\/person\/image\/","url":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","contentUrl":"https:\/\/gravatar.wp-china-yes.net\/avatar\/?s=96&d=mystery","caption":"admin"},"sameAs":["http:\/\/wp.wsisp.com"],"url":"https:\/\/www.wsisp.com\/helps\/author\/admin"}]}},"_links":{"self":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/75371","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/comments?post=75371"}],"version-history":[{"count":0,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/posts\/75371\/revisions"}],"wp:attachment":[{"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/media?parent=75371"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/categories?post=75371"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/tags?post=75371"},{"taxonomy":"topic","embeddable":true,"href":"https:\/\/www.wsisp.com\/helps\/wp-json\/wp\/v2\/topic?post=75371"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}