diff --git a/tests/amdgpu/amdgpu_test.h b/tests/amdgpu/amdgpu_test.h index 068840c9..9f4453db 100644 --- a/tests/amdgpu/amdgpu_test.h +++ b/tests/amdgpu/amdgpu_test.h @@ -285,8 +285,8 @@ extern CU_TestInfo cp_dma_tests[]; void amdgpu_dispatch_hang_helper(amdgpu_device_handle device_handle, uint32_t ip_type); void amdgpu_dispatch_hang_slow_helper(amdgpu_device_handle device_handle, uint32_t ip_type); void amdgpu_memcpy_draw_test(amdgpu_device_handle device_handle, uint32_t ring, - int hang); -void amdgpu_memcpy_draw_hang_slow_test(amdgpu_device_handle device_handle, uint32_t ring); + int version, int hang); +void amdgpu_memcpy_draw_hang_slow_test(amdgpu_device_handle device_handle, uint32_t ring, int version); /** * Initialize security test suite diff --git a/tests/amdgpu/basic_tests.c b/tests/amdgpu/basic_tests.c index 8afd05c1..43b80c88 100644 --- a/tests/amdgpu/basic_tests.c +++ b/tests/amdgpu/basic_tests.c @@ -292,6 +292,8 @@ CU_TestInfo basic_tests[] = { #define PKT3_SET_SH_REG 0x76 #define PACKET3_SET_SH_REG_START 0x00002c00 +#define PKT3_SET_SH_REG_INDEX 0x9B + #define PACKET3_DISPATCH_DIRECT 0x15 #define PACKET3_EVENT_WRITE 0x46 #define PACKET3_ACQUIRE_MEM 0x58 @@ -390,6 +392,32 @@ static const uint32_t preamblecache_gfx9[] = { 0xc0017900, 0x24b, 0x0 }; +static const uint32_t preamblecache_gfx10[] = { + 0xc0026900, 0x81, 0x80000000, 0x40004000, 0xc0026900, 0x8c, 0xaa99aaaa, 0x0, + 0xc0026900, 0x90, 0x80000000, 0x40004000, 0xc0026900, 0x94, 0x80000000, 0x40004000, + 0xc0026900, 0xb4, 0x0, 0x3f800000, 0xc0016900, 0x103, 0x0, + 0xc0016900, 0x208, 0x0, 0xc0016900, 0x290, 0x0, + 0xc0016900, 0x2a1, 0x0, 0xc0026900, 0x2ad, 0x0, 0x0, + 0xc0016900, 0x2d5, 0x10000, 0xc0016900, 0x2dc, 0x0, + 0xc0066900, 0x2de, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0xc0026900, 0x2e5, 0x0, 0x0, + 0xc0056900, 0x2f9, 0x5, 0x3f800000, 0x3f800000, 0x3f800000, 0x3f800000, + 0xc0046900, 0x310, 0, 0x3, 0, 0x100000, 0xc0026900, 0x316, 0xe, 0x20, + 0xc0016900, 0x349, 0x0, 0xc0016900, 0x358, 0x0, 0xc0016900, 0x367, 0x0, + 0xc0016900, 0x376, 0x0, 0xc0016900, 0x385, 0x0, 0xc0016900, 0x6, 0x0, + 0xc0056900, 0xe8, 0x0, 0x0, 0x0, 0x0, 0x0, + 0xc0076900, 0x1e1, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, + 0xc0026900, 0x204, 0x90000, 0x4, 0xc0046900, 0x20c, 0x0, 0x0, 0x0, 0x0, + 0xc0016900, 0x2b2, 0x0, 0xc0026900, 0x30e, 0xffffffff, 0xffffffff, + 0xc0016900, 0x314, 0x0, 0xc0016900, 0x10a, 0, 0xc0016900, 0x2a6, 0, 0xc0016900, 0x210, 0, + 0xc0016900, 0x2db, 0, 0xc0016900, 0x1d4, 0, 0xc0002f00, 0x1, 0xc0016900, 0x1, 0x1, 0xc0016900, 0xe, 0x2, + 0xc0016900, 0x206, 0x300, 0xc0016900, 0x212, 0x200, 0xc0017900, 0x7b, 0x20, 0xc0017a00, 0x20000243, 0x0, + 0xc0017900, 0x249, 0, 0xc0017900, 0x24a, 0, 0xc0017900, 0x24b, 0, 0xc0017900, 0x259, 0xffffffff, + 0xc0017900, 0x25f, 0, 0xc0017900, 0x260, 0, 0xc0017900, 0x262, 0, + 0xc0017600, 0x45, 0x0, 0xc0017600, 0x6, 0x0, + 0xc0067600, 0x70, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, + 0xc0067600, 0x30, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0 +}; + enum ps_type { PS_CONST, PS_TEX, @@ -442,6 +470,39 @@ static const uint32_t ps_const_context_reg_gfx9[][2] = { {0xA1C5, 0x00000004}, //{ mmSPI_SHADER_COL_FORMAT, 0x00000004 } }; +static const uint32_t ps_const_shader_gfx10[] = { + 0x7E000200, 0x7E020201, 0x7E040202, 0x7E060203, + 0x5E000300, 0x5E020702, 0xBF800000, 0xBF800000, + 0xF8001C0F, 0x00000100, 0xBF810000 +}; + +static const uint32_t ps_const_shader_patchinfo_code_size_gfx10 = 6; + +static const uint32_t ps_const_shader_patchinfo_code_gfx10[][10][6] = { + {{ 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xF8001890, 0x00000000 }, + { 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xF8001801, 0x00000000 }, + { 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xF8001803, 0x00000100 }, + { 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xF8001803, 0x00000300 }, + { 0x5E000300, 0x5E020702, 0xBF800000, 0xBF800000, 0xF8001C0F, 0x00000100 }, + { 0xD7690000, 0x00020300, 0xD7690001, 0x00020702, 0xF8001C0F, 0x00000100 }, + { 0xD7680000, 0x00020300, 0xD7680001, 0x00020702, 0xF8001C0F, 0x00000100 }, + { 0xD76A0000, 0x00020300, 0xD76A0001, 0x00020702, 0xF8001C0F, 0x00000100 }, + { 0xD76B0000, 0x00020300, 0xD76B0001, 0x00020702, 0xF8001C0F, 0x00000100 }, + { 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xF800180F, 0x03020100 } + } +}; + +static const uint32_t ps_const_shader_patchinfo_offset_gfx10[] = { + 0x00000004 +}; + +static const uint32_t ps_num_sh_registers_gfx10 = 2; + +static const uint32_t ps_const_sh_registers_gfx10[][2] = { + {0x2C0A, 0x000C0000},//{ mmSPI_SHADER_PGM_RSRC1_PS, 0x000C0000 }, + {0x2C0B, 0x00000008}, //{ mmSPI_SHADER_PGM_RSRC2_PS, 0x00000008 } +}; + static const uint32_t ps_tex_shader_gfx9[] = { 0xBEFC000C, 0xBE8E017E, 0xBEFE077E, 0xD4180000, 0xD4190001, 0xD41C0100, 0xD41D0101, 0xF0800F00, @@ -485,6 +546,34 @@ static const uint32_t ps_tex_context_reg_gfx9[][2] = { {0xA1C5, 0x00000004}, //{ mmSPI_SHADER_COL_FORMAT, 0x00000004 } }; +static const uint32_t ps_tex_shader_gfx10[] = { + 0xBEFC030C, 0xBE8E047E, 0xBEFE0A7E, 0xC8080000, + 0xC80C0100, 0xC8090001, 0xC80D0101, 0xF0800F0A, + 0x00400402, 0x00000003, 0xBEFE040E, 0xBF8C0F70, + 0x5E000B04, 0x5E020F06, 0xBF800000, 0xBF800000, + 0xF8001C0F, 0x00000100, 0xBF810000 +}; + +static const uint32_t ps_tex_shader_patchinfo_offset_gfx10[] = { + 0x0000000C +}; + +static const uint32_t ps_tex_shader_patchinfo_code_size_gfx10 = 6; + +static const uint32_t ps_tex_shader_patchinfo_code_gfx10[][10][6] = { + {{ 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xF8001890, 0x00000000 }, + { 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xF8001801, 0x00000004 }, + { 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xF8001803, 0x00000504 }, + { 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xF8001803, 0x00000704 }, + { 0x5E000B04, 0x5E020F06, 0xBF800000, 0xBF800000, 0xF8001C0F, 0x00000100 }, + { 0xD7690000, 0x00020B04, 0xD7690001, 0x00020F06, 0xF8001C0F, 0x00000100 }, + { 0xD7680000, 0x00020B04, 0xD7680001, 0x00020F06, 0xF8001C0F, 0x00000100 }, + { 0xD76A0000, 0x00020B04, 0xD76A0001, 0x00020F06, 0xF8001C0F, 0x00000100 }, + { 0xD76B0000, 0x00020B04, 0xD76B0001, 0x00020F06, 0xF8001C0F, 0x00000100 }, + { 0xBF800000, 0xBF800000, 0xBF800000, 0xBF800000, 0xF800180F, 0x07060504 } + } +}; + static const uint32_t vs_RectPosTexFast_shader_gfx9[] = { 0x7E000B00, 0x020000F3, 0xD042000A, 0x00010100, 0x7E020202, 0x7E040200, 0x020000F3, 0x7E060206, @@ -496,6 +585,17 @@ static const uint32_t vs_RectPosTexFast_shader_gfx9[] = { 0xC400020F, 0x05060403, 0xBF810000 }; +static const uint32_t vs_RectPosTexFast_shader_gfx10[] = { + 0x7E000B00, 0x060000F3, 0x7E020202, 0x7E040206, + 0x7C040080, 0x060000F3, 0xD5010001, 0x01AA0200, + 0x7E060203, 0xD5010002, 0x01AA0404, 0x7E080207, + 0x7C040080, 0xD5010000, 0x01A80101, 0xD5010001, + 0x01AA0601, 0x7E060208, 0x7E0A02F2, 0xD5010002, + 0x01A80902, 0xD5010004, 0x01AA0805, 0x7E0C0209, + 0xF80008CF, 0x05030100, 0xF800020F, 0x05060402, + 0xBF810000 +}; + static const uint32_t cached_cmd_gfx9[] = { 0xc0016900, 0x0, 0x0, 0xc0026900, 0x3, 0x2a, 0x0, 0xc0046900, 0xa, 0x0, 0x0, 0x0, 0x200020, @@ -507,6 +607,17 @@ static const uint32_t cached_cmd_gfx9[] = { 0xc0026900, 0x2b0, 0x0, 0x0, 0xc0016900, 0x2f8, 0x0 }; +static const uint32_t cached_cmd_gfx10[] = { + 0xc0016900, 0x0, 0x0, 0xc0026900, 0x3, 0x2a, 0x0, + 0xc0046900, 0xa, 0x0, 0x0, 0x0, 0x200020, + 0xc0016900, 0x83, 0xffff, 0xc0026900, 0x8e, 0xf, 0xf, + 0xc0056900, 0x105, 0x0, 0x0, 0x0, 0x0, 0x18, + 0xc0026900, 0x10b, 0x0, 0x0, 0xc0016900, 0x1e0, 0x0, + 0xc0036900, 0x200, 0x0, 0x10000, 0xcc0011, + 0xc0026900, 0x292, 0x20, 0x6020000, + 0xc0026900, 0x2b0, 0x0, 0x0, 0xc0016900, 0x2f8, 0x0 +}; + unsigned int memcpy_ps_hang[] = { 0xFFFFFFFF, 0xBEFE0A7E, 0xBEFC0304, 0xC0C20100, 0xC0800300, 0xC8080000, 0xC80C0100, 0xC8090001, @@ -3004,7 +3115,7 @@ static int amdgpu_draw_load_ps_shader_hang_slow(uint32_t *ptr, int family) return 0; } -static int amdgpu_draw_load_ps_shader(uint8_t *ptr, int ps_type) +static int amdgpu_draw_load_ps_shader(uint8_t *ptr, int ps_type, uint32_t version) { int i; uint32_t shader_offset= 256; @@ -3016,18 +3127,34 @@ static int amdgpu_draw_load_ps_shader(uint8_t *ptr, int ps_type) switch (ps_type) { case PS_CONST: - shader = ps_const_shader_gfx9; - shader_size = sizeof(ps_const_shader_gfx9); - patchinfo_code = (const uint32_t *)ps_const_shader_patchinfo_code_gfx9; - patchinfo_code_size = ps_const_shader_patchinfo_code_size_gfx9; - patchcode_offset = ps_const_shader_patchinfo_offset_gfx9; + if (version == 9) { + shader = ps_const_shader_gfx9; + shader_size = sizeof(ps_const_shader_gfx9); + patchinfo_code = (const uint32_t *)ps_const_shader_patchinfo_code_gfx9; + patchinfo_code_size = ps_const_shader_patchinfo_code_size_gfx9; + patchcode_offset = ps_const_shader_patchinfo_offset_gfx9; + } else if (version == 10){ + shader = ps_const_shader_gfx10; + shader_size = sizeof(ps_const_shader_gfx10); + patchinfo_code = (const uint32_t *)ps_const_shader_patchinfo_code_gfx10; + patchinfo_code_size = ps_const_shader_patchinfo_code_size_gfx10; + patchcode_offset = ps_const_shader_patchinfo_offset_gfx10; + } break; case PS_TEX: - shader = ps_tex_shader_gfx9; - shader_size = sizeof(ps_tex_shader_gfx9); - patchinfo_code = (const uint32_t *)ps_tex_shader_patchinfo_code_gfx9; - patchinfo_code_size = ps_tex_shader_patchinfo_code_size_gfx9; - patchcode_offset = ps_tex_shader_patchinfo_offset_gfx9; + if (version == 9) { + shader = ps_tex_shader_gfx9; + shader_size = sizeof(ps_tex_shader_gfx9); + patchinfo_code = (const uint32_t *)ps_tex_shader_patchinfo_code_gfx9; + patchinfo_code_size = ps_tex_shader_patchinfo_code_size_gfx9; + patchcode_offset = ps_tex_shader_patchinfo_offset_gfx9; + } else if (version == 10) { + shader = ps_tex_shader_gfx10; + shader_size = sizeof(ps_tex_shader_gfx10); + patchinfo_code = (const uint32_t *)ps_tex_shader_patchinfo_code_gfx10; + patchinfo_code_size = ps_tex_shader_patchinfo_code_size_gfx10; + patchcode_offset = ps_tex_shader_patchinfo_offset_gfx10; + } break; case PS_HANG: shader = memcpy_ps_hang; @@ -3059,20 +3186,25 @@ static int amdgpu_draw_load_ps_shader(uint8_t *ptr, int ps_type) } /* load RectPosTexFast_VS */ -static int amdgpu_draw_load_vs_shader(uint8_t *ptr) +static int amdgpu_draw_load_vs_shader(uint8_t *ptr, uint32_t version) { const uint32_t *shader; uint32_t shader_size; - shader = vs_RectPosTexFast_shader_gfx9; - shader_size = sizeof(vs_RectPosTexFast_shader_gfx9); + if (version == 9) { + shader = vs_RectPosTexFast_shader_gfx9; + shader_size = sizeof(vs_RectPosTexFast_shader_gfx9); + } else if (version == 10) { + shader = vs_RectPosTexFast_shader_gfx10; + shader_size = sizeof(vs_RectPosTexFast_shader_gfx10); + } memcpy(ptr, shader, shader_size); return 0; } -static int amdgpu_draw_init(uint32_t *ptr) +static int amdgpu_draw_init(uint32_t *ptr, uint32_t version) { int i = 0; const uint32_t *preamblecache_ptr; @@ -3083,8 +3215,13 @@ static int amdgpu_draw_init(uint32_t *ptr) ptr[i++] = 0x80000000; ptr[i++] = 0x80000000; - preamblecache_ptr = preamblecache_gfx9; - preamblecache_size = sizeof(preamblecache_gfx9); + if (version == 9) { + preamblecache_ptr = preamblecache_gfx9; + preamblecache_size = sizeof(preamblecache_gfx9); + } else if (version == 10) { + preamblecache_ptr = preamblecache_gfx10; + preamblecache_size = sizeof(preamblecache_gfx10); + } memcpy(ptr + i, preamblecache_ptr, preamblecache_size); return i + preamblecache_size/sizeof(uint32_t); @@ -3092,41 +3229,95 @@ static int amdgpu_draw_init(uint32_t *ptr) static int amdgpu_draw_setup_and_write_drawblt_surf_info(uint32_t *ptr, uint64_t dst_addr, + uint32_t version, int hang_slow) { int i = 0; /* setup color buffer */ - /* offset reg - 0xA318 CB_COLOR0_BASE - 0xA319 CB_COLOR0_BASE_EXT - 0xA31A CB_COLOR0_ATTRIB2 - 0xA31B CB_COLOR0_VIEW - 0xA31C CB_COLOR0_INFO - 0xA31D CB_COLOR0_ATTRIB - 0xA31E CB_COLOR0_DCC_CONTROL - 0xA31F CB_COLOR0_CMASK - 0xA320 CB_COLOR0_CMASK_BASE_EXT - 0xA321 CB_COLOR0_FMASK - 0xA322 CB_COLOR0_FMASK_BASE_EXT - 0xA323 CB_COLOR0_CLEAR_WORD0 - 0xA324 CB_COLOR0_CLEAR_WORD1 - 0xA325 CB_COLOR0_DCC_BASE - 0xA326 CB_COLOR0_DCC_BASE_EXT */ - ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 15); - ptr[i++] = 0x318; - ptr[i++] = dst_addr >> 8; - ptr[i++] = dst_addr >> 40; - ptr[i++] = hang_slow ? 0x1ffc7ff : 0x7c01f; - ptr[i++] = 0; - ptr[i++] = 0x50438; - ptr[i++] = 0x10140000; - i += 9; + if (version == 9) { + /* offset reg + 0xA318 CB_COLOR0_BASE + 0xA319 CB_COLOR0_BASE_EXT + 0xA31A CB_COLOR0_ATTRIB2 + 0xA31B CB_COLOR0_VIEW + 0xA31C CB_COLOR0_INFO + 0xA31D CB_COLOR0_ATTRIB + 0xA31E CB_COLOR0_DCC_CONTROL + 0xA31F CB_COLOR0_CMASK + 0xA320 CB_COLOR0_CMASK_BASE_EXT + 0xA321 CB_COLOR0_FMASK + 0xA322 CB_COLOR0_FMASK_BASE_EXT + 0xA323 CB_COLOR0_CLEAR_WORD0 + 0xA324 CB_COLOR0_CLEAR_WORD1 + 0xA325 CB_COLOR0_DCC_BASE + 0xA326 CB_COLOR0_DCC_BASE_EXT */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 15); + ptr[i++] = 0x318; + ptr[i++] = dst_addr >> 8; + ptr[i++] = dst_addr >> 40; + ptr[i++] = hang_slow ? 0x3ffc7ff : 0x7c01f; + ptr[i++] = 0; + ptr[i++] = 0x50438; + ptr[i++] = 0x10140000; + i += 9; - /* mmCB_MRT0_EPITCH */ - ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); - ptr[i++] = 0x1e8; - ptr[i++] = hang_slow ? 0x7ff : 0x1f; + /* mmCB_MRT0_EPITCH */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); + ptr[i++] = 0x1e8; + ptr[i++] = hang_slow ? 0xfff : 0x1f; + } else if (version == 10) { + /* 0xA318 CB_COLOR0_BASE + 0xA319 CB_COLOR0_PITCH + 0xA31A CB_COLOR0_SLICE + 0xA31B CB_COLOR0_VIEW + 0xA31C CB_COLOR0_INFO + 0xA31D CB_COLOR0_ATTRIB + 0xA31E CB_COLOR0_DCC_CONTROL + 0xA31F CB_COLOR0_CMASK + 0xA320 CB_COLOR0_CMASK_SLICE + 0xA321 CB_COLOR0_FMASK + 0xA322 CB_COLOR0_FMASK_SLICE + 0xA323 CB_COLOR0_CLEAR_WORD0 + 0xA324 CB_COLOR0_CLEAR_WORD1 + 0xA325 CB_COLOR0_DCC_BASE */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 14); + ptr[i++] = 0x318; + ptr[i++] = dst_addr >> 8; + i += 3; + ptr[i++] = 0x50438; + i += 9; + + /* 0xA390 CB_COLOR0_BASE_EXT */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); + ptr[i++] = 0x390; + ptr[i++] = dst_addr >> 40; + + /* 0xA398 CB_COLOR0_CMASK_BASE_EXT */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); + ptr[i++] = 0x398; + ptr[i++] = 0; + + /* 0xA3A0 CB_COLOR0_FMASK_BASE_EXT */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); + ptr[i++] = 0x3a0; + ptr[i++] = 0; + + /* 0xA3A8 CB_COLOR0_DCC_BASE_EXT */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); + ptr[i++] = 0x3a8; + ptr[i++] = 0; + + /* 0xA3B0 CB_COLOR0_ATTRIB2 */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); + ptr[i++] = 0x3b0; + ptr[i++] = hang_slow ? 0x3ffc7ff : 0x7c01f; + + /* 0xA3B8 CB_COLOR0_ATTRIB3 */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); + ptr[i++] = 0x3b8; + ptr[i++] = 0x9014000; + } /* 0xA32B CB_COLOR1_BASE */ ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); @@ -3144,15 +3335,24 @@ static int amdgpu_draw_setup_and_write_drawblt_surf_info(uint32_t *ptr, ptr[i++] = 9; /* Setup depth buffer */ - /* mmDB_Z_INFO */ - ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 2); - ptr[i++] = 0xe; - i += 2; + if (version == 9) { + /* mmDB_Z_INFO */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 2); + ptr[i++] = 0xe; + i += 2; + } else if (version == 10) { + /* mmDB_Z_INFO */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 2); + ptr[i++] = 0x10; + i += 2; + } return i; } -static int amdgpu_draw_setup_and_write_drawblt_state(uint32_t *ptr, int hang_slow) +static int amdgpu_draw_setup_and_write_drawblt_state(uint32_t *ptr, + uint32_t version, + int hang_slow) { int i = 0; const uint32_t *cached_cmd_ptr; @@ -3168,7 +3368,10 @@ static int amdgpu_draw_setup_and_write_drawblt_state(uint32_t *ptr, int hang_slo ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); ptr[i++] = 0xd7; - ptr[i++] = 1; + if (version == 9) + ptr[i++] = 1; + else if (version == 10) + ptr[i++] = 0; /* mmPA_SC_AA_SAMPLE_LOCS_PIXEL_X0Y0_0 */ ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 16); @@ -3180,20 +3383,37 @@ static int amdgpu_draw_setup_and_write_drawblt_state(uint32_t *ptr, int hang_slo ptr[i++] = 0x2f5; i += 2; - cached_cmd_ptr = cached_cmd_gfx9; - cached_cmd_size = sizeof(cached_cmd_gfx9); + if (version == 9) { + cached_cmd_ptr = cached_cmd_gfx9; + cached_cmd_size = sizeof(cached_cmd_gfx9); + } else if (version == 10) { + cached_cmd_ptr = cached_cmd_gfx10; + cached_cmd_size = sizeof(cached_cmd_gfx10); + } memcpy(ptr + i, cached_cmd_ptr, cached_cmd_size); if (hang_slow) *(ptr + i + 12) = 0x8000800; i += cached_cmd_size/sizeof(uint32_t); + if (version == 10) { + /* mmCB_RMI_GL2_CACHE_CONTROL */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); + ptr[i++] = 0x104; + ptr[i++] = 0x40aa0055; + /* mmDB_RMI_L2_CACHE_CONTROL */ + ptr[i++] = PACKET3(PACKET3_SET_CONTEXT_REG, 1); + ptr[i++] = 0x1f; + ptr[i++] = 0x2a0055; + } + return i; } static int amdgpu_draw_vs_RectPosTexFast_write2hw(uint32_t *ptr, int ps_type, uint64_t shader_addr, + uint32_t version, int hang_slow) { int i = 0; @@ -3203,10 +3423,21 @@ static int amdgpu_draw_vs_RectPosTexFast_write2hw(uint32_t *ptr, ptr[i++] = 0x207; ptr[i++] = 0; - /* mmSPI_SHADER_PGM_RSRC3_VS */ - ptr[i++] = PACKET3(PKT3_SET_SH_REG, 1); - ptr[i++] = 0x46; - ptr[i++] = 0xffff; + if (version == 9) { + /* mmSPI_SHADER_PGM_RSRC3_VS */ + ptr[i++] = PACKET3(PKT3_SET_SH_REG, 1); + ptr[i++] = 0x46; + ptr[i++] = 0xffff; + } else if (version == 10) { + /* mmSPI_SHADER_PGM_RSRC3_VS */ + ptr[i++] = PACKET3(PKT3_SET_SH_REG_INDEX, 1); + ptr[i++] = 0x30000046; + ptr[i++] = 0xffff; + /* mmSPI_SHADER_PGM_RSRC4_VS */ + ptr[i++] = PACKET3(PKT3_SET_SH_REG_INDEX, 1); + ptr[i++] = 0x30000041; + ptr[i++] = 0xffff; + } /* mmSPI_SHADER_PGM_LO_VS...mmSPI_SHADER_PGM_HI_VS */ ptr[i++] = PACKET3(PKT3_SET_SH_REG, 2); @@ -3217,7 +3448,10 @@ static int amdgpu_draw_vs_RectPosTexFast_write2hw(uint32_t *ptr, /* mmSPI_SHADER_PGM_RSRC1_VS */ ptr[i++] = PACKET3(PKT3_SET_SH_REG, 1); ptr[i++] = 0x4a; - ptr[i++] = 0xc0081; + if (version == 9) + ptr[i++] = 0xc0081; + else if (version == 10) + ptr[i++] = 0xc0041; /* mmSPI_SHADER_PGM_RSRC2_VS */ ptr[i++] = PACKET3(PKT3_SET_SH_REG, 1); ptr[i++] = 0x4b; @@ -3258,7 +3492,8 @@ static int amdgpu_draw_vs_RectPosTexFast_write2hw(uint32_t *ptr, static int amdgpu_draw_ps_write2hw(uint32_t *ptr, int ps_type, - uint64_t shader_addr) + uint64_t shader_addr, + uint32_t version) { int i, j; const uint32_t *sh_registers; @@ -3266,9 +3501,14 @@ static int amdgpu_draw_ps_write2hw(uint32_t *ptr, uint32_t num_sh_reg, num_context_reg; if (ps_type == PS_CONST) { - sh_registers = (const uint32_t *)ps_const_sh_registers_gfx9; + if (version == 9) { + sh_registers = (const uint32_t *)ps_const_sh_registers_gfx9; + num_sh_reg = ps_num_sh_registers_gfx9; + } else if (version == 10) { + sh_registers = (const uint32_t *)ps_const_sh_registers_gfx10; + num_sh_reg = ps_num_sh_registers_gfx10; + } context_registers = (const uint32_t *)ps_const_context_reg_gfx9; - num_sh_reg = ps_num_sh_registers_gfx9; num_context_reg = ps_num_context_registers_gfx9; } else if (ps_type == PS_TEX) { sh_registers = (const uint32_t *)ps_tex_sh_registers_gfx9; @@ -3279,15 +3519,35 @@ static int amdgpu_draw_ps_write2hw(uint32_t *ptr, i = 0; - /* 0x2c07 SPI_SHADER_PGM_RSRC3_PS - 0x2c08 SPI_SHADER_PGM_LO_PS - 0x2c09 SPI_SHADER_PGM_HI_PS */ - shader_addr += 256 * 9; - ptr[i++] = PACKET3(PKT3_SET_SH_REG, 3); - ptr[i++] = 0x7; - ptr[i++] = 0xffff; - ptr[i++] = shader_addr >> 8; - ptr[i++] = shader_addr >> 40; + if (version == 9) { + /* 0x2c07 SPI_SHADER_PGM_RSRC3_PS + 0x2c08 SPI_SHADER_PGM_LO_PS + 0x2c09 SPI_SHADER_PGM_HI_PS */ + /* multiplicator 9 is from SPI_SHADER_COL_FORMAT */ + shader_addr += 256 * 9; + ptr[i++] = PACKET3(PKT3_SET_SH_REG, 3); + ptr[i++] = 0x7; + ptr[i++] = 0xffff; + ptr[i++] = shader_addr >> 8; + ptr[i++] = shader_addr >> 40; + } else if (version == 10) { + shader_addr += 256 * 9; + /* 0x2c08 SPI_SHADER_PGM_LO_PS + 0x2c09 SPI_SHADER_PGM_HI_PS */ + ptr[i++] = PACKET3(PKT3_SET_SH_REG, 2); + ptr[i++] = 0x8; + ptr[i++] = shader_addr >> 8; + ptr[i++] = shader_addr >> 40; + + /* mmSPI_SHADER_PGM_RSRC3_PS */ + ptr[i++] = PACKET3(PKT3_SET_SH_REG_INDEX, 1); + ptr[i++] = 0x30000007; + ptr[i++] = 0xffff; + /* mmSPI_SHADER_PGM_RSRC4_PS */ + ptr[i++] = PACKET3(PKT3_SET_SH_REG_INDEX, 1); + ptr[i++] = 0x30000001; + ptr[i++] = 0xffff; + } for (j = 0; j < num_sh_reg; j++) { ptr[i++] = PACKET3(PKT3_SET_SH_REG, 1); @@ -3312,19 +3572,29 @@ static int amdgpu_draw_ps_write2hw(uint32_t *ptr, return i; } -static int amdgpu_draw_draw(uint32_t *ptr) +static int amdgpu_draw_draw(uint32_t *ptr, uint32_t version) { int i = 0; - /* mmIA_MULTI_VGT_PARAM */ - ptr[i++] = PACKET3(PACKET3_SET_UCONFIG_REG, 1); - ptr[i++] = 0x40000258; - ptr[i++] = 0xd00ff; - - /* mmVGT_PRIMITIVE_TYPE */ - ptr[i++] = PACKET3(PACKET3_SET_UCONFIG_REG, 1); - ptr[i++] = 0x10000242; - ptr[i++] = 0x11; + if (version == 9) { + /* mmIA_MULTI_VGT_PARAM */ + ptr[i++] = PACKET3(PACKET3_SET_UCONFIG_REG, 1); + ptr[i++] = 0x40000258; + ptr[i++] = 0xd00ff; + /* mmVGT_PRIMITIVE_TYPE */ + ptr[i++] = PACKET3(PACKET3_SET_UCONFIG_REG, 1); + ptr[i++] = 0x10000242; + ptr[i++] = 0x11; + } else if (version == 10) { + /* mmGE_CNTL */ + ptr[i++] = PACKET3(PACKET3_SET_UCONFIG_REG, 1); + ptr[i++] = 0x25b; + ptr[i++] = 0xff; + /* mmVGT_PRIMITIVE_TYPE */ + ptr[i++] = PACKET3(PACKET3_SET_UCONFIG_REG, 1); + ptr[i++] = 0x242; + ptr[i++] = 0x11; + } ptr[i++] = PACKET3(PACKET3_DRAW_INDEX_AUTO, 1); ptr[i++] = 3; @@ -3338,7 +3608,7 @@ void amdgpu_memset_draw(amdgpu_device_handle device_handle, amdgpu_bo_handle bo_shader_vs, uint64_t mc_address_shader_ps, uint64_t mc_address_shader_vs, - uint32_t ring_id) + uint32_t ring_id, uint32_t version) { amdgpu_context_handle context_handle; amdgpu_bo_handle bo_dst, bo_cmd, resources[4]; @@ -3372,15 +3642,16 @@ void amdgpu_memset_draw(amdgpu_device_handle device_handle, CU_ASSERT_EQUAL(r, 0); i = 0; - i += amdgpu_draw_init(ptr_cmd + i); + i += amdgpu_draw_init(ptr_cmd + i, version); - i += amdgpu_draw_setup_and_write_drawblt_surf_info(ptr_cmd + i, mc_address_dst, 0); + i += amdgpu_draw_setup_and_write_drawblt_surf_info(ptr_cmd + i, mc_address_dst, version, 0); - i += amdgpu_draw_setup_and_write_drawblt_state(ptr_cmd + i, 0); + i += amdgpu_draw_setup_and_write_drawblt_state(ptr_cmd + i, version, 0); - i += amdgpu_draw_vs_RectPosTexFast_write2hw(ptr_cmd + i, PS_CONST, mc_address_shader_vs, 0); + i += amdgpu_draw_vs_RectPosTexFast_write2hw(ptr_cmd + i, PS_CONST, mc_address_shader_vs, + version, 0); - i += amdgpu_draw_ps_write2hw(ptr_cmd + i, PS_CONST, mc_address_shader_ps); + i += amdgpu_draw_ps_write2hw(ptr_cmd + i, PS_CONST, mc_address_shader_ps, version); ptr_cmd[i++] = PACKET3(PKT3_SET_SH_REG, 4); ptr_cmd[i++] = 0xc; @@ -3389,7 +3660,7 @@ void amdgpu_memset_draw(amdgpu_device_handle device_handle, ptr_cmd[i++] = 0x33333333; ptr_cmd[i++] = 0x33333333; - i += amdgpu_draw_draw(ptr_cmd + i); + i += amdgpu_draw_draw(ptr_cmd + i, version); while (i & 7) ptr_cmd[i++] = 0xffff1000; /* type3 nop packet */ @@ -3447,7 +3718,7 @@ void amdgpu_memset_draw(amdgpu_device_handle device_handle, } static void amdgpu_memset_draw_test(amdgpu_device_handle device_handle, - uint32_t ring) + uint32_t ring, int version) { amdgpu_bo_handle bo_shader_ps, bo_shader_vs; void *ptr_shader_ps; @@ -3471,14 +3742,15 @@ static void amdgpu_memset_draw_test(amdgpu_device_handle device_handle, CU_ASSERT_EQUAL(r, 0); memset(ptr_shader_vs, 0, bo_shader_size); - r = amdgpu_draw_load_ps_shader(ptr_shader_ps, PS_CONST); + r = amdgpu_draw_load_ps_shader(ptr_shader_ps, PS_CONST, version); CU_ASSERT_EQUAL(r, 0); - r = amdgpu_draw_load_vs_shader(ptr_shader_vs); + r = amdgpu_draw_load_vs_shader(ptr_shader_vs, version); CU_ASSERT_EQUAL(r, 0); amdgpu_memset_draw(device_handle, bo_shader_ps, bo_shader_vs, - mc_address_shader_ps, mc_address_shader_vs, ring); + mc_address_shader_ps, mc_address_shader_vs, + ring, version); r = amdgpu_bo_unmap_and_free(bo_shader_ps, va_shader_ps, mc_address_shader_ps, bo_shader_size); CU_ASSERT_EQUAL(r, 0); @@ -3492,7 +3764,7 @@ static void amdgpu_memcpy_draw(amdgpu_device_handle device_handle, amdgpu_bo_handle bo_shader_vs, uint64_t mc_address_shader_ps, uint64_t mc_address_shader_vs, - uint32_t ring, int hang) + uint32_t ring, int version, int hang) { amdgpu_context_handle context_handle; amdgpu_bo_handle bo_dst, bo_src, bo_cmd, resources[5]; @@ -3536,24 +3808,36 @@ static void amdgpu_memcpy_draw(amdgpu_device_handle device_handle, memset(ptr_src, 0x55, bo_size); i = 0; - i += amdgpu_draw_init(ptr_cmd + i); + i += amdgpu_draw_init(ptr_cmd + i, version); - i += amdgpu_draw_setup_and_write_drawblt_surf_info(ptr_cmd + i, mc_address_dst, 0); + i += amdgpu_draw_setup_and_write_drawblt_surf_info(ptr_cmd + i, mc_address_dst, version, 0); - i += amdgpu_draw_setup_and_write_drawblt_state(ptr_cmd + i, 0); + i += amdgpu_draw_setup_and_write_drawblt_state(ptr_cmd + i, version, 0); - i += amdgpu_draw_vs_RectPosTexFast_write2hw(ptr_cmd + i, PS_TEX, mc_address_shader_vs, 0); + i += amdgpu_draw_vs_RectPosTexFast_write2hw(ptr_cmd + i, PS_TEX, mc_address_shader_vs, + version, 0); - i += amdgpu_draw_ps_write2hw(ptr_cmd + i, PS_TEX, mc_address_shader_ps); + i += amdgpu_draw_ps_write2hw(ptr_cmd + i, PS_TEX, mc_address_shader_ps, version); ptr_cmd[i++] = PACKET3(PKT3_SET_SH_REG, 8); - ptr_cmd[i++] = 0xc; - ptr_cmd[i++] = mc_address_src >> 8; - ptr_cmd[i++] = mc_address_src >> 40 | 0x10e00000; - ptr_cmd[i++] = 0x7c01f; - ptr_cmd[i++] = 0x90500fac; - ptr_cmd[i++] = 0x3e000; - i += 3; + if (version == 9) { + ptr_cmd[i++] = 0xc; + ptr_cmd[i++] = mc_address_src >> 8; + ptr_cmd[i++] = mc_address_src >> 40 | 0x10e00000; + ptr_cmd[i++] = 0x7c01f; + ptr_cmd[i++] = 0x90500fac; + ptr_cmd[i++] = 0x3e000; + i += 3; + } else if (version == 10) { + ptr_cmd[i++] = 0xc; + ptr_cmd[i++] = mc_address_src >> 8; + ptr_cmd[i++] = mc_address_src >> 40 | 0xc4b00000; + ptr_cmd[i++] = 0x8007c007; + ptr_cmd[i++] = 0x90500fac; + i += 2; + ptr_cmd[i++] = 0x400; + i++; + } ptr_cmd[i++] = PACKET3(PKT3_SET_SH_REG, 4); ptr_cmd[i++] = 0x14; @@ -3564,7 +3848,7 @@ static void amdgpu_memcpy_draw(amdgpu_device_handle device_handle, ptr_cmd[i++] = 0x191; ptr_cmd[i++] = 0; - i += amdgpu_draw_draw(ptr_cmd + i); + i += amdgpu_draw_draw(ptr_cmd + i, version); while (i & 7) ptr_cmd[i++] = 0xffff1000; /* type3 nop packet */ @@ -3630,7 +3914,7 @@ static void amdgpu_memcpy_draw(amdgpu_device_handle device_handle, } void amdgpu_memcpy_draw_test(amdgpu_device_handle device_handle, uint32_t ring, - int hang) + int version, int hang) { amdgpu_bo_handle bo_shader_ps, bo_shader_vs; void *ptr_shader_ps; @@ -3655,14 +3939,15 @@ void amdgpu_memcpy_draw_test(amdgpu_device_handle device_handle, uint32_t ring, CU_ASSERT_EQUAL(r, 0); memset(ptr_shader_vs, 0, bo_shader_size); - r = amdgpu_draw_load_ps_shader(ptr_shader_ps, ps_type); + r = amdgpu_draw_load_ps_shader(ptr_shader_ps, ps_type, version); CU_ASSERT_EQUAL(r, 0); - r = amdgpu_draw_load_vs_shader(ptr_shader_vs); + r = amdgpu_draw_load_vs_shader(ptr_shader_vs, version); CU_ASSERT_EQUAL(r, 0); amdgpu_memcpy_draw(device_handle, bo_shader_ps, bo_shader_vs, - mc_address_shader_ps, mc_address_shader_vs, ring, hang); + mc_address_shader_ps, mc_address_shader_vs, + ring, version, hang); r = amdgpu_bo_unmap_and_free(bo_shader_ps, va_shader_ps, mc_address_shader_ps, bo_shader_size); CU_ASSERT_EQUAL(r, 0); @@ -3675,20 +3960,26 @@ static void amdgpu_draw_test(void) { int r; struct drm_amdgpu_info_hw_ip info; - uint32_t ring_id; + uint32_t ring_id, version; r = amdgpu_query_hw_ip_info(device_handle, AMDGPU_HW_IP_GFX, 0, &info); CU_ASSERT_EQUAL(r, 0); if (!info.available_rings) printf("SKIP ... as there's no graphics ring\n"); + version = info.hw_ip_version_major; + if (version != 9 && version != 10) { + printf("SKIP ... unsupported gfx version %d\n", version); + return; + } + for (ring_id = 0; (1 << ring_id) & info.available_rings; ring_id++) { - amdgpu_memset_draw_test(device_handle, ring_id); - amdgpu_memcpy_draw_test(device_handle, ring_id, 0); + amdgpu_memset_draw_test(device_handle, ring_id, version); + amdgpu_memcpy_draw_test(device_handle, ring_id, version, 0); } } -void amdgpu_memcpy_draw_hang_slow_test(amdgpu_device_handle device_handle, uint32_t ring) +void amdgpu_memcpy_draw_hang_slow_test(amdgpu_device_handle device_handle, uint32_t ring, int version) { amdgpu_context_handle context_handle; amdgpu_bo_handle bo_shader_ps, bo_shader_vs; @@ -3744,7 +4035,7 @@ void amdgpu_memcpy_draw_hang_slow_test(amdgpu_device_handle device_handle, uint3 r = amdgpu_draw_load_ps_shader_hang_slow(ptr_shader_ps, gpu_info.family_id); CU_ASSERT_EQUAL(r, 0); - r = amdgpu_draw_load_vs_shader(ptr_shader_vs); + r = amdgpu_draw_load_vs_shader(ptr_shader_vs, version); CU_ASSERT_EQUAL(r, 0); r = amdgpu_bo_alloc_and_map(device_handle, bo_size, 4096, @@ -3762,25 +4053,35 @@ void amdgpu_memcpy_draw_hang_slow_test(amdgpu_device_handle device_handle, uint3 memset(ptr_src, 0x55, bo_size); i = 0; - i += amdgpu_draw_init(ptr_cmd + i); + i += amdgpu_draw_init(ptr_cmd + i, version); - i += amdgpu_draw_setup_and_write_drawblt_surf_info(ptr_cmd + i, mc_address_dst, 1); + i += amdgpu_draw_setup_and_write_drawblt_surf_info(ptr_cmd + i, mc_address_dst, version, 1); - i += amdgpu_draw_setup_and_write_drawblt_state(ptr_cmd + i, 1); + i += amdgpu_draw_setup_and_write_drawblt_state(ptr_cmd + i, version, 1); i += amdgpu_draw_vs_RectPosTexFast_write2hw(ptr_cmd + i, PS_TEX, - mc_address_shader_vs, 1); + mc_address_shader_vs, version, 1); - i += amdgpu_draw_ps_write2hw(ptr_cmd + i, PS_TEX, mc_address_shader_ps); + i += amdgpu_draw_ps_write2hw(ptr_cmd + i, PS_TEX, mc_address_shader_ps, version); ptr_cmd[i++] = PACKET3(PKT3_SET_SH_REG, 8); - ptr_cmd[i++] = 0xc; - ptr_cmd[i++] = mc_address_src >> 8; - ptr_cmd[i++] = mc_address_src >> 40 | 0x10e00000; - ptr_cmd[i++] = 0x1ffc7ff; - ptr_cmd[i++] = 0x90500fac; - ptr_cmd[i++] = 0xffe000; - i += 3; + + if (version == 9) { + ptr_cmd[i++] = 0xc; + ptr_cmd[i++] = mc_address_src >> 8; + ptr_cmd[i++] = mc_address_src >> 40 | 0x10e00000; + ptr_cmd[i++] = 0x1ffcfff; + ptr_cmd[i++] = 0x90500fac; + ptr_cmd[i++] = 0x1ffe000; + i += 3; + } else if (version == 10) { + ptr_cmd[i++] = 0xc; + ptr_cmd[i++] = mc_address_src >> 8; + ptr_cmd[i++] = mc_address_src >> 40 | 0xc4b00000; + ptr_cmd[i++] = 0x81ffc1ff; + ptr_cmd[i++] = 0x90500fac; + i += 4; + } ptr_cmd[i++] = PACKET3(PKT3_SET_SH_REG, 4); ptr_cmd[i++] = 0x14; @@ -3791,7 +4092,7 @@ void amdgpu_memcpy_draw_hang_slow_test(amdgpu_device_handle device_handle, uint3 ptr_cmd[i++] = 0x191; ptr_cmd[i++] = 0; - i += amdgpu_draw_draw(ptr_cmd + i); + i += amdgpu_draw_draw(ptr_cmd + i, version); while (i & 7) ptr_cmd[i++] = 0xffff1000; /* type3 nop packet */ diff --git a/tests/amdgpu/deadlock_tests.c b/tests/amdgpu/deadlock_tests.c index 0f780d48..f29a83ab 100644 --- a/tests/amdgpu/deadlock_tests.c +++ b/tests/amdgpu/deadlock_tests.c @@ -533,32 +533,44 @@ static void amdgpu_draw_hang_gfx(void) { int r; struct drm_amdgpu_info_hw_ip info; - uint32_t ring_id; + uint32_t ring_id, version; r = amdgpu_query_hw_ip_info(device_handle, AMDGPU_HW_IP_GFX, 0, &info); CU_ASSERT_EQUAL(r, 0); if (!info.available_rings) printf("SKIP ... as there's no graphic ring\n"); + version = info.hw_ip_version_major; + if (version != 9 && version != 10) { + printf("SKIP ... unsupported gfx version %d\n", version); + return; + } + for (ring_id = 0; (1 << ring_id) & info.available_rings; ring_id++) { - amdgpu_memcpy_draw_test(device_handle, ring_id, 0); - amdgpu_memcpy_draw_test(device_handle, ring_id, 1); - amdgpu_memcpy_draw_test(device_handle, ring_id, 0); + amdgpu_memcpy_draw_test(device_handle, ring_id, version, 0); + amdgpu_memcpy_draw_test(device_handle, ring_id, version, 1); + amdgpu_memcpy_draw_test(device_handle, ring_id, version, 0); } } static void amdgpu_draw_hang_slow_gfx(void) { struct drm_amdgpu_info_hw_ip info; - uint32_t ring_id; + uint32_t ring_id, version; int r; r = amdgpu_query_hw_ip_info(device_handle, AMDGPU_HW_IP_GFX, 0, &info); CU_ASSERT_EQUAL(r, 0); + version = info.hw_ip_version_major; + if (version != 9 && version != 10) { + printf("SKIP ... unsupported gfx version %d\n", version); + return; + } + for (ring_id = 0; (1 << ring_id) & info.available_rings; ring_id++) { - amdgpu_memcpy_draw_test(device_handle, ring_id, 0); - amdgpu_memcpy_draw_hang_slow_test(device_handle, ring_id); - amdgpu_memcpy_draw_test(device_handle, ring_id, 0); + amdgpu_memcpy_draw_test(device_handle, ring_id, version, 0); + amdgpu_memcpy_draw_hang_slow_test(device_handle, ring_id, version); + amdgpu_memcpy_draw_test(device_handle, ring_id, version, 0); } }