From f3f4335f739ced7a8c0545b5455893a8a85387f4 Mon Sep 17 00:00:00 2001 From: Raveendra Talabattula Date: Thu, 10 Jan 2019 16:23:30 +0530 Subject: [PATCH 5752/5758] Revert drm/amdgpu: abstract the function of enter/exit safe mode Signed-off-by: Raveendra Talabattula --- drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.c | 229 +-------------------- drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.h | 33 ++- drivers/gpu/drm/amd/amdgpu/ci_dpm.c | 6 +- drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c | 24 ++- drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c | 148 ++++++++++--- drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c | 201 ++++++++++++------ drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c | 183 ++++++++++++---- drivers/gpu/drm/amd/amdgpu/kv_dpm.c | 6 +- .../gpu/drm/amd/powerplay/hwmgr/smu7_powertune.c | 12 +- .../gpu/drm/amd/powerplay/hwmgr/vega10_powertune.c | 36 ++-- 10 files changed, 470 insertions(+), 408 deletions(-) diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.c index c8793e6..c5459ab 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.c +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.c @@ -1,3 +1,4 @@ + /* * Copyright 2014 Advanced Micro Devices, Inc. * Copyright 2008 Red Hat Inc. @@ -22,238 +23,12 @@ * OTHER DEALINGS IN THE SOFTWARE. * */ -#include + #include "amdgpu.h" #include "amdgpu_gfx.h" #include "amdgpu_rlc.h" /** - * amdgpu_gfx_rlc_enter_safe_mode - Set RLC into safe mode - * - * @adev: amdgpu_device pointer - * - * Set RLC enter into safe mode if RLC is enabled and haven't in safe mode. - */ -void amdgpu_gfx_rlc_enter_safe_mode(struct amdgpu_device *adev) -{ - if (adev->gfx.rlc.in_safe_mode) - return; - - /* if RLC is not enabled, do nothing */ - if (!adev->gfx.rlc.funcs->is_rlc_enabled(adev)) - return; - - if (adev->cg_flags & - (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG | - AMD_CG_SUPPORT_GFX_3D_CGCG)) { - adev->gfx.rlc.funcs->set_safe_mode(adev); - adev->gfx.rlc.in_safe_mode = true; - } -} - -/** - * amdgpu_gfx_rlc_exit_safe_mode - Set RLC out of safe mode - * - * @adev: amdgpu_device pointer - * - * Set RLC exit safe mode if RLC is enabled and have entered into safe mode. - */ -void amdgpu_gfx_rlc_exit_safe_mode(struct amdgpu_device *adev) -{ - if (!(adev->gfx.rlc.in_safe_mode)) - return; - - /* if RLC is not enabled, do nothing */ - if (!adev->gfx.rlc.funcs->is_rlc_enabled(adev)) - return; - - if (adev->cg_flags & - (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG | - AMD_CG_SUPPORT_GFX_3D_CGCG)) { - adev->gfx.rlc.funcs->unset_safe_mode(adev); - adev->gfx.rlc.in_safe_mode = false; - } -} - -/** - * amdgpu_gfx_rlc_init_sr - Init save restore block - * - * @adev: amdgpu_device pointer - * @dws: the size of save restore block - * - * Allocate and setup value to save restore block of rlc. - * Returns 0 on succeess or negative error code if allocate failed. - */ -int amdgpu_gfx_rlc_init_sr(struct amdgpu_device *adev, u32 dws) -{ - const u32 *src_ptr; - volatile u32 *dst_ptr; - u32 i; - int r; - - /* allocate save restore block */ - r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.save_restore_obj, - &adev->gfx.rlc.save_restore_gpu_addr, - (void **)&adev->gfx.rlc.sr_ptr); - if (r) { - dev_warn(adev->dev, "(%d) create RLC sr bo failed\n", r); - amdgpu_gfx_rlc_fini(adev); - return r; - } - - /* write the sr buffer */ - src_ptr = adev->gfx.rlc.reg_list; - dst_ptr = adev->gfx.rlc.sr_ptr; - for (i = 0; i < adev->gfx.rlc.reg_list_size; i++) - dst_ptr[i] = cpu_to_le32(src_ptr[i]); - amdgpu_bo_kunmap(adev->gfx.rlc.save_restore_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj); - - return 0; -} - -/** - * amdgpu_gfx_rlc_init_csb - Init clear state block - * - * @adev: amdgpu_device pointer - * - * Allocate and setup value to clear state block of rlc. - * Returns 0 on succeess or negative error code if allocate failed. - */ -int amdgpu_gfx_rlc_init_csb(struct amdgpu_device *adev) -{ - volatile u32 *dst_ptr; - u32 dws; - int r; - - /* allocate clear state block */ - adev->gfx.rlc.clear_state_size = dws = adev->gfx.rlc.funcs->get_csb_size(adev); - r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, - AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.clear_state_obj, - &adev->gfx.rlc.clear_state_gpu_addr, - (void **)&adev->gfx.rlc.cs_ptr); - if (r) { - dev_err(adev->dev, "(%d) failed to create rlc csb bo\n", r); - amdgpu_gfx_rlc_fini(adev); - return r; - } - - /* set up the cs buffer */ - dst_ptr = adev->gfx.rlc.cs_ptr; - adev->gfx.rlc.funcs->get_csb_buffer(adev, dst_ptr); - amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj); - amdgpu_bo_unpin(adev->gfx.rlc.clear_state_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj); - - return 0; -} - -/** - * amdgpu_gfx_rlc_init_cpt - Init cp table - * - * @adev: amdgpu_device pointer - * - * Allocate and setup value to cp table of rlc. - * Returns 0 on succeess or negative error code if allocate failed. - */ -int amdgpu_gfx_rlc_init_cpt(struct amdgpu_device *adev) -{ - int r; - - r = amdgpu_bo_create_reserved(adev, adev->gfx.rlc.cp_table_size, - PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, - &adev->gfx.rlc.cp_table_obj, - &adev->gfx.rlc.cp_table_gpu_addr, - (void **)&adev->gfx.rlc.cp_table_ptr); - if (r) { - dev_err(adev->dev, "(%d) failed to create cp table bo\n", r); - amdgpu_gfx_rlc_fini(adev); - return r; - } - - /* set up the cp table */ - amdgpu_gfx_rlc_setup_cp_table(adev); - amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj); - amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj); - - return 0; -} - -/** - * amdgpu_gfx_rlc_setup_cp_table - setup cp the buffer of cp table - * - * @adev: amdgpu_device pointer - * - * Write cp firmware data into cp table. - */ -void amdgpu_gfx_rlc_setup_cp_table(struct amdgpu_device *adev) -{ - const __le32 *fw_data; - volatile u32 *dst_ptr; - int me, i, max_me; - u32 bo_offset = 0; - u32 table_offset, table_size; - - max_me = adev->gfx.rlc.funcs->get_cp_table_num(adev); - - /* write the cp table buffer */ - dst_ptr = adev->gfx.rlc.cp_table_ptr; - for (me = 0; me < max_me; me++) { - if (me == 0) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; - fw_data = (const __le32 *) - (adev->gfx.ce_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 1) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; - fw_data = (const __le32 *) - (adev->gfx.pfp_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 2) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; - fw_data = (const __le32 *) - (adev->gfx.me_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 3) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; - fw_data = (const __le32 *) - (adev->gfx.mec_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } else if (me == 4) { - const struct gfx_firmware_header_v1_0 *hdr = - (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; - fw_data = (const __le32 *) - (adev->gfx.mec2_fw->data + - le32_to_cpu(hdr->header.ucode_array_offset_bytes)); - table_offset = le32_to_cpu(hdr->jt_offset); - table_size = le32_to_cpu(hdr->jt_size); - } - - for (i = 0; i < table_size; i ++) { - dst_ptr[bo_offset + i] = - cpu_to_le32(le32_to_cpu(fw_data[table_offset + i])); - } - - bo_offset += table_size; - } -} - -/** * amdgpu_gfx_rlc_fini - Free BO which used for RLC * * @adev: amdgpu_device pointer diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.h b/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.h index 49a8ab5..b3b0920 100644 --- a/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.h +++ b/drivers/gpu/drm/amd/amdgpu/amdgpu_rlc.h @@ -1,3 +1,4 @@ + /* * Copyright 2014 Advanced Micro Devices, Inc. * @@ -27,13 +28,9 @@ #include "clearstate_defs.h" struct amdgpu_rlc_funcs { - bool (*is_rlc_enabled)(struct amdgpu_device *adev); - void (*set_safe_mode)(struct amdgpu_device *adev); - void (*unset_safe_mode)(struct amdgpu_device *adev); + void (*enter_safe_mode)(struct amdgpu_device *adev); + void (*exit_safe_mode)(struct amdgpu_device *adev); int (*init)(struct amdgpu_device *adev); - u32 (*get_csb_size)(struct amdgpu_device *adev); - void (*get_csb_buffer)(struct amdgpu_device *adev, volatile u32 *buffer); - int (*get_cp_table_num)(struct amdgpu_device *adev); int (*resume)(struct amdgpu_device *adev); void (*stop)(struct amdgpu_device *adev); void (*reset)(struct amdgpu_device *adev); @@ -42,21 +39,21 @@ struct amdgpu_rlc_funcs { struct amdgpu_rlc { /* for power gating */ - struct amdgpu_bo *save_restore_obj; - uint64_t save_restore_gpu_addr; - volatile uint32_t *sr_ptr; + struct amdgpu_bo *save_restore_obj; + uint64_t save_restore_gpu_addr; + volatile uint32_t *sr_ptr; const u32 *reg_list; u32 reg_list_size; /* for clear state */ - struct amdgpu_bo *clear_state_obj; - uint64_t clear_state_gpu_addr; - volatile uint32_t *cs_ptr; + struct amdgpu_bo *clear_state_obj; + uint64_t clear_state_gpu_addr; + volatile uint32_t *cs_ptr; const struct cs_section_def *cs_data; u32 clear_state_size; /* for cp tables */ - struct amdgpu_bo *cp_table_obj; - uint64_t cp_table_gpu_addr; - volatile uint32_t *cp_table_ptr; + struct amdgpu_bo *cp_table_obj; + uint64_t cp_table_gpu_addr; + volatile uint32_t *cp_table_ptr; u32 cp_table_size; /* safe mode for updating CG/PG state */ @@ -87,12 +84,6 @@ struct amdgpu_rlc { bool is_rlc_v2_1; }; -void amdgpu_gfx_rlc_enter_safe_mode(struct amdgpu_device *adev); -void amdgpu_gfx_rlc_exit_safe_mode(struct amdgpu_device *adev); -int amdgpu_gfx_rlc_init_sr(struct amdgpu_device *adev, u32 dws); -int amdgpu_gfx_rlc_init_csb(struct amdgpu_device *adev); -int amdgpu_gfx_rlc_init_cpt(struct amdgpu_device *adev); -void amdgpu_gfx_rlc_setup_cp_table(struct amdgpu_device *adev); void amdgpu_gfx_rlc_fini(struct amdgpu_device *adev); #endif diff --git a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c index e02631d..ecd69ab 100644 --- a/drivers/gpu/drm/amd/amdgpu/ci_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/ci_dpm.c @@ -743,19 +743,19 @@ static int ci_enable_didt(struct amdgpu_device *adev, bool enable) if (pi->caps_sq_ramping || pi->caps_db_ramping || pi->caps_td_ramping || pi->caps_tcp_ramping) { - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); if (enable) { ret = ci_program_pt_config_registers(adev, didt_config_ci); if (ret) { - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return ret; } } ci_do_enable_didt(adev, enable); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); } return 0; diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c index 075407e..abc8ec6 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v6_0.c @@ -2401,7 +2401,7 @@ static int gfx_v6_0_rlc_init(struct amdgpu_device *adev) { const u32 *src_ptr; volatile u32 *dst_ptr; - u32 dws; + u32 dws, i; u64 reg_list_mc_addr; const struct cs_section_def *cs_data; int r; @@ -2416,10 +2416,26 @@ static int gfx_v6_0_rlc_init(struct amdgpu_device *adev) cs_data = adev->gfx.rlc.cs_data; if (src_ptr) { - /* init save restore block */ - r = amdgpu_gfx_rlc_init_sr(adev, dws); - if (r) + /* save restore block */ + r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.save_restore_obj, + &adev->gfx.rlc.save_restore_gpu_addr, + (void **)&adev->gfx.rlc.sr_ptr); + if (r) { + dev_warn(adev->dev, "(%d) create RLC sr bo failed\n", + r); + amdgpu_gfx_rlc_fini(adev); return r; + } + + /* write the sr buffer */ + dst_ptr = adev->gfx.rlc.sr_ptr; + for (i = 0; i < adev->gfx.rlc.reg_list_size; i++) + dst_ptr[i] = cpu_to_le32(src_ptr[i]); + + amdgpu_bo_kunmap(adev->gfx.rlc.save_restore_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj); } if (cs_data) { diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c index 6815153..19a0e4f 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v7_0.c @@ -882,6 +882,7 @@ static const u32 kalindi_rlc_save_restore_register_list[] = static u32 gfx_v7_0_get_csb_size(struct amdgpu_device *adev); static void gfx_v7_0_get_csb_buffer(struct amdgpu_device *adev, volatile u32 *buffer); +static void gfx_v7_0_init_cp_pg_table(struct amdgpu_device *adev); static void gfx_v7_0_init_pg(struct amdgpu_device *adev); static void gfx_v7_0_get_cu_info(struct amdgpu_device *adev); @@ -3290,7 +3291,8 @@ static void gfx_v7_0_ring_emit_wreg(struct amdgpu_ring *ring, static int gfx_v7_0_rlc_init(struct amdgpu_device *adev) { const u32 *src_ptr; - u32 dws; + volatile u32 *dst_ptr; + u32 dws, i; const struct cs_section_def *cs_data; int r; @@ -3317,23 +3319,66 @@ static int gfx_v7_0_rlc_init(struct amdgpu_device *adev) cs_data = adev->gfx.rlc.cs_data; if (src_ptr) { - /* init save restore block */ - r = amdgpu_gfx_rlc_init_sr(adev, dws); - if (r) + /* save restore block */ + r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.save_restore_obj, + &adev->gfx.rlc.save_restore_gpu_addr, + (void **)&adev->gfx.rlc.sr_ptr); + if (r) { + dev_warn(adev->dev, "(%d) create, pin or map of RLC sr bo failed\n", r); + amdgpu_gfx_rlc_fini(adev); return r; + } + + /* write the sr buffer */ + dst_ptr = adev->gfx.rlc.sr_ptr; + for (i = 0; i < adev->gfx.rlc.reg_list_size; i++) + dst_ptr[i] = cpu_to_le32(src_ptr[i]); + amdgpu_bo_kunmap(adev->gfx.rlc.save_restore_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.save_restore_obj); } if (cs_data) { - /* init clear state block */ - r = amdgpu_gfx_rlc_init_csb(adev); - if (r) + /* clear state block */ + adev->gfx.rlc.clear_state_size = dws = gfx_v7_0_get_csb_size(adev); + + r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.clear_state_obj, + &adev->gfx.rlc.clear_state_gpu_addr, + (void **)&adev->gfx.rlc.cs_ptr); + if (r) { + dev_warn(adev->dev, "(%d) create RLC c bo failed\n", r); + amdgpu_gfx_rlc_fini(adev); return r; + } + + /* set up the cs buffer */ + dst_ptr = adev->gfx.rlc.cs_ptr; + gfx_v7_0_get_csb_buffer(adev, dst_ptr); + amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj); } if (adev->gfx.rlc.cp_table_size) { - r = amdgpu_gfx_rlc_init_cpt(adev); - if (r) + + r = amdgpu_bo_create_reserved(adev, adev->gfx.rlc.cp_table_size, + PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.cp_table_obj, + &adev->gfx.rlc.cp_table_gpu_addr, + (void **)&adev->gfx.rlc.cp_table_ptr); + if (r) { + dev_warn(adev->dev, "(%d) create RLC cp table bo failed\n", r); + amdgpu_gfx_rlc_fini(adev); return r; + } + + gfx_v7_0_init_cp_pg_table(adev); + + amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj); + } return 0; @@ -3414,12 +3459,7 @@ static u32 gfx_v7_0_halt_rlc(struct amdgpu_device *adev) return orig; } -static bool gfx_v7_0_is_rlc_enabled(struct amdgpu_device *adev) -{ - return true; -} - -static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev) +static void gfx_v7_0_enter_rlc_safe_mode(struct amdgpu_device *adev) { u32 tmp, i, mask; @@ -3441,7 +3481,7 @@ static void gfx_v7_0_set_safe_mode(struct amdgpu_device *adev) } } -static void gfx_v7_0_unset_safe_mode(struct amdgpu_device *adev) +static void gfx_v7_0_exit_rlc_safe_mode(struct amdgpu_device *adev) { u32 tmp; @@ -3757,12 +3797,72 @@ static void gfx_v7_0_enable_gds_pg(struct amdgpu_device *adev, bool enable) WREG32(mmRLC_PG_CNTL, data); } -static int gfx_v7_0_cp_pg_table_num(struct amdgpu_device *adev) +static void gfx_v7_0_init_cp_pg_table(struct amdgpu_device *adev) { + const __le32 *fw_data; + volatile u32 *dst_ptr; + int me, i, max_me = 4; + u32 bo_offset = 0; + u32 table_offset, table_size; + if (adev->asic_type == CHIP_KAVERI) - return 5; - else - return 4; + max_me = 5; + + if (adev->gfx.rlc.cp_table_ptr == NULL) + return; + + /* write the cp table buffer */ + dst_ptr = adev->gfx.rlc.cp_table_ptr; + for (me = 0; me < max_me; me++) { + if (me == 0) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; + fw_data = (const __le32 *) + (adev->gfx.ce_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 1) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; + fw_data = (const __le32 *) + (adev->gfx.pfp_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 2) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; + fw_data = (const __le32 *) + (adev->gfx.me_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 3) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; + fw_data = (const __le32 *) + (adev->gfx.mec_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; + fw_data = (const __le32 *) + (adev->gfx.mec2_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } + + for (i = 0; i < table_size; i ++) { + dst_ptr[bo_offset + i] = + cpu_to_le32(le32_to_cpu(fw_data[table_offset + i])); + } + + bo_offset += table_size; + } } static void gfx_v7_0_enable_gfx_cgpg(struct amdgpu_device *adev, @@ -4201,12 +4301,8 @@ static const struct amdgpu_gfx_funcs gfx_v7_0_gfx_funcs = { }; static const struct amdgpu_rlc_funcs gfx_v7_0_rlc_funcs = { - .is_rlc_enabled = gfx_v7_0_is_rlc_enabled, - .set_safe_mode = gfx_v7_0_set_safe_mode, - .unset_safe_mode = gfx_v7_0_unset_safe_mode, - .get_csb_size = gfx_v7_0_get_csb_size, - .get_csb_buffer = gfx_v7_0_get_csb_buffer, - .get_cp_table_num = gfx_v7_0_cp_pg_table_num, + .enter_safe_mode = gfx_v7_0_enter_rlc_safe_mode, + .exit_safe_mode = gfx_v7_0_exit_rlc_safe_mode }; static int gfx_v7_0_early_init(void *handle) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c index 43272fb..3ec5832 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c @@ -1298,16 +1298,75 @@ static void gfx_v8_0_get_csb_buffer(struct amdgpu_device *adev, buffer[count++] = cpu_to_le32(0); } -static int gfx_v8_0_cp_jump_table_num(struct amdgpu_device *adev) +static void cz_init_cp_jump_table(struct amdgpu_device *adev) { + const __le32 *fw_data; + volatile u32 *dst_ptr; + int me, i, max_me = 4; + u32 bo_offset = 0; + u32 table_offset, table_size; + if (adev->asic_type == CHIP_CARRIZO) - return 5; - else - return 4; + max_me = 5; + + /* write the cp table buffer */ + dst_ptr = adev->gfx.rlc.cp_table_ptr; + for (me = 0; me < max_me; me++) { + if (me == 0) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; + fw_data = (const __le32 *) + (adev->gfx.ce_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 1) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; + fw_data = (const __le32 *) + (adev->gfx.pfp_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 2) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; + fw_data = (const __le32 *) + (adev->gfx.me_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 3) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; + fw_data = (const __le32 *) + (adev->gfx.mec_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 4) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; + fw_data = (const __le32 *) + (adev->gfx.mec2_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } + + for (i = 0; i < table_size; i ++) { + dst_ptr[bo_offset + i] = + cpu_to_le32(le32_to_cpu(fw_data[table_offset + i])); + } + + bo_offset += table_size; + } } static int gfx_v8_0_rlc_init(struct amdgpu_device *adev) { + volatile u32 *dst_ptr; + u32 dws; const struct cs_section_def *cs_data; int r; @@ -1316,18 +1375,44 @@ static int gfx_v8_0_rlc_init(struct amdgpu_device *adev) cs_data = adev->gfx.rlc.cs_data; if (cs_data) { - /* init clear state block */ - r = amdgpu_gfx_rlc_init_csb(adev); - if (r) + /* clear state block */ + adev->gfx.rlc.clear_state_size = dws = gfx_v8_0_get_csb_size(adev); + + r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.clear_state_obj, + &adev->gfx.rlc.clear_state_gpu_addr, + (void **)&adev->gfx.rlc.cs_ptr); + if (r) { + dev_warn(adev->dev, "(%d) create RLC c bo failed\n", r); + amdgpu_gfx_rlc_fini(adev); return r; + } + + /* set up the cs buffer */ + dst_ptr = adev->gfx.rlc.cs_ptr; + gfx_v8_0_get_csb_buffer(adev, dst_ptr); + amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj); } if ((adev->asic_type == CHIP_CARRIZO) || (adev->asic_type == CHIP_STONEY)) { adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */ - r = amdgpu_gfx_rlc_init_cpt(adev); - if (r) + r = amdgpu_bo_create_reserved(adev, adev->gfx.rlc.cp_table_size, + PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.cp_table_obj, + &adev->gfx.rlc.cp_table_gpu_addr, + (void **)&adev->gfx.rlc.cp_table_ptr); + if (r) { + dev_warn(adev->dev, "(%d) create RLC cp table bo failed\n", r); return r; + } + + cz_init_cp_jump_table(adev); + + amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj); } return 0; @@ -4880,7 +4965,7 @@ static int gfx_v8_0_hw_fini(void *handle) pr_debug("For SRIOV client, shouldn't do anything.\n"); return 0; } - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); if (!gfx_v8_0_wait_for_idle(adev)) gfx_v8_0_cp_enable(adev, false); else @@ -4889,7 +4974,7 @@ static int gfx_v8_0_hw_fini(void *handle) gfx_v8_0_rlc_stop(adev); else pr_err("rlc is busy, skip halt rlc\n"); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return 0; } @@ -5352,7 +5437,7 @@ static int gfx_v8_0_set_powergating_state(void *handle, AMD_PG_SUPPORT_RLC_SMU_HS | AMD_PG_SUPPORT_CP | AMD_PG_SUPPORT_GFX_DMG)) - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); switch (adev->asic_type) { case CHIP_CARRIZO: case CHIP_STONEY: @@ -5406,7 +5491,7 @@ static int gfx_v8_0_set_powergating_state(void *handle, AMD_PG_SUPPORT_RLC_SMU_HS | AMD_PG_SUPPORT_CP | AMD_PG_SUPPORT_GFX_DMG)) - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return 0; } @@ -5500,53 +5585,57 @@ static void gfx_v8_0_send_serdes_cmd(struct amdgpu_device *adev, #define RLC_GPR_REG2__MESSAGE__SHIFT 0x00000001 #define RLC_GPR_REG2__MESSAGE_MASK 0x0000001e -static bool gfx_v8_0_is_rlc_enabled(struct amdgpu_device *adev) +static void iceland_enter_rlc_safe_mode(struct amdgpu_device *adev) { - uint32_t rlc_setting; + u32 data; + unsigned i; - rlc_setting = RREG32(mmRLC_CNTL); - if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK)) - return false; + data = RREG32(mmRLC_CNTL); + if (!(data & RLC_CNTL__RLC_ENABLE_F32_MASK)) + return; - return true; -} + if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG)) { + data |= RLC_SAFE_MODE__CMD_MASK; + data &= ~RLC_SAFE_MODE__MESSAGE_MASK; + data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); + WREG32(mmRLC_SAFE_MODE, data); -static void gfx_v8_0_set_safe_mode(struct amdgpu_device *adev) -{ - uint32_t data; - unsigned i; - data = RREG32(mmRLC_CNTL); - data |= RLC_SAFE_MODE__CMD_MASK; - data &= ~RLC_SAFE_MODE__MESSAGE_MASK; - data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); - WREG32(mmRLC_SAFE_MODE, data); + for (i = 0; i < adev->usec_timeout; i++) { + if ((RREG32(mmRLC_GPM_STAT) & + (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK | + RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) == + (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK | + RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) + break; + udelay(1); + } - /* wait for RLC_SAFE_MODE */ - for (i = 0; i < adev->usec_timeout; i++) { - if ((RREG32(mmRLC_GPM_STAT) & - (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK | - RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) == - (RLC_GPM_STAT__GFX_CLOCK_STATUS_MASK | - RLC_GPM_STAT__GFX_POWER_STATUS_MASK)) - break; - udelay(1); - } - for (i = 0; i < adev->usec_timeout; i++) { - if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) - break; - udelay(1); + for (i = 0; i < adev->usec_timeout; i++) { + if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) + break; + udelay(1); + } + adev->gfx.rlc.in_safe_mode = true; } } -static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev) +static void iceland_exit_rlc_safe_mode(struct amdgpu_device *adev) { - uint32_t data; + u32 data = 0; unsigned i; data = RREG32(mmRLC_CNTL); - data |= RLC_SAFE_MODE__CMD_MASK; - data &= ~RLC_SAFE_MODE__MESSAGE_MASK; - WREG32(mmRLC_SAFE_MODE, data); + if (!(data & RLC_CNTL__RLC_ENABLE_F32_MASK)) + return; + + if (adev->cg_flags & (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG)) { + if (adev->gfx.rlc.in_safe_mode) { + data |= RLC_SAFE_MODE__CMD_MASK; + data &= ~RLC_SAFE_MODE__MESSAGE_MASK; + WREG32(mmRLC_SAFE_MODE, data); + adev->gfx.rlc.in_safe_mode = false; + } + } for (i = 0; i < adev->usec_timeout; i++) { if (!REG_GET_FIELD(RREG32(mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) @@ -5556,12 +5645,8 @@ static void gfx_v8_0_unset_safe_mode(struct amdgpu_device *adev) } static const struct amdgpu_rlc_funcs iceland_rlc_funcs = { - .is_rlc_enabled = gfx_v8_0_is_rlc_enabled, - .set_safe_mode = gfx_v8_0_set_safe_mode, - .unset_safe_mode = gfx_v8_0_unset_safe_mode, - .get_csb_size = gfx_v8_0_get_csb_size, - .get_csb_buffer = gfx_v8_0_get_csb_buffer, - .get_cp_table_num = gfx_v8_0_cp_jump_table_num, + .enter_safe_mode = iceland_enter_rlc_safe_mode, + .exit_safe_mode = iceland_exit_rlc_safe_mode }; static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev, @@ -5569,7 +5654,7 @@ static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev { uint32_t temp, data; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); /* It is disabled by HW by default */ if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_MGCG)) { @@ -5665,7 +5750,7 @@ static void gfx_v8_0_update_medium_grain_clock_gating(struct amdgpu_device *adev gfx_v8_0_wait_for_rlc_serdes(adev); } - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); } static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev, @@ -5675,7 +5760,7 @@ static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev temp = data = RREG32(mmRLC_CGCG_CGLS_CTRL); - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { temp1 = data1 = RREG32(mmRLC_CGTT_MGCG_OVERRIDE); @@ -5758,7 +5843,7 @@ static void gfx_v8_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev gfx_v8_0_wait_for_rlc_serdes(adev); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); } static int gfx_v8_0_update_gfx_clock_gating(struct amdgpu_device *adev, bool enable) diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c index ac2a843..69fcc77 100644 --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c @@ -1079,13 +1079,72 @@ static void gfx_v9_0_enable_lbpw(struct amdgpu_device *adev, bool enable) WREG32_FIELD15(GC, 0, RLC_LB_CNTL, LOAD_BALANCE_ENABLE, enable ? 1 : 0); } -static int gfx_v9_0_cp_jump_table_num(struct amdgpu_device *adev) +static void rv_init_cp_jump_table(struct amdgpu_device *adev) { - return 5; + const __le32 *fw_data; + volatile u32 *dst_ptr; + int me, i, max_me = 5; + u32 bo_offset = 0; + u32 table_offset, table_size; + + /* write the cp table buffer */ + dst_ptr = adev->gfx.rlc.cp_table_ptr; + for (me = 0; me < max_me; me++) { + if (me == 0) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.ce_fw->data; + fw_data = (const __le32 *) + (adev->gfx.ce_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 1) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.pfp_fw->data; + fw_data = (const __le32 *) + (adev->gfx.pfp_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 2) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.me_fw->data; + fw_data = (const __le32 *) + (adev->gfx.me_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 3) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec_fw->data; + fw_data = (const __le32 *) + (adev->gfx.mec_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } else if (me == 4) { + const struct gfx_firmware_header_v1_0 *hdr = + (const struct gfx_firmware_header_v1_0 *)adev->gfx.mec2_fw->data; + fw_data = (const __le32 *) + (adev->gfx.mec2_fw->data + + le32_to_cpu(hdr->header.ucode_array_offset_bytes)); + table_offset = le32_to_cpu(hdr->jt_offset); + table_size = le32_to_cpu(hdr->jt_size); + } + + for (i = 0; i < table_size; i ++) { + dst_ptr[bo_offset + i] = + cpu_to_le32(le32_to_cpu(fw_data[table_offset + i])); + } + + bo_offset += table_size; + } } static int gfx_v9_0_rlc_init(struct amdgpu_device *adev) { + volatile u32 *dst_ptr; + u32 dws; const struct cs_section_def *cs_data; int r; @@ -1094,18 +1153,45 @@ static int gfx_v9_0_rlc_init(struct amdgpu_device *adev) cs_data = adev->gfx.rlc.cs_data; if (cs_data) { - /* init clear state block */ - r = amdgpu_gfx_rlc_init_csb(adev); - if (r) + /* clear state block */ + adev->gfx.rlc.clear_state_size = dws = gfx_v9_0_get_csb_size(adev); + r = amdgpu_bo_create_reserved(adev, dws * 4, PAGE_SIZE, + AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.clear_state_obj, + &adev->gfx.rlc.clear_state_gpu_addr, + (void **)&adev->gfx.rlc.cs_ptr); + if (r) { + dev_err(adev->dev, "(%d) failed to create rlc csb bo\n", + r); + amdgpu_gfx_rlc_fini(adev); return r; + } + /* set up the cs buffer */ + dst_ptr = adev->gfx.rlc.cs_ptr; + gfx_v9_0_get_csb_buffer(adev, dst_ptr); + amdgpu_bo_kunmap(adev->gfx.rlc.clear_state_obj); + amdgpu_bo_unpin(adev->gfx.rlc.clear_state_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.clear_state_obj); } if (adev->asic_type == CHIP_RAVEN) { /* TODO: double check the cp_table_size for RV */ adev->gfx.rlc.cp_table_size = ALIGN(96 * 5 * 4, 2048) + (64 * 1024); /* JT + GDS */ - r = amdgpu_gfx_rlc_init_cpt(adev); - if (r) + r = amdgpu_bo_create_reserved(adev, adev->gfx.rlc.cp_table_size, + PAGE_SIZE, AMDGPU_GEM_DOMAIN_VRAM, + &adev->gfx.rlc.cp_table_obj, + &adev->gfx.rlc.cp_table_gpu_addr, + (void **)&adev->gfx.rlc.cp_table_ptr); + if (r) { + dev_err(adev->dev, + "(%d) failed to create cp table bo\n", r); + amdgpu_gfx_rlc_fini(adev); return r; + } + + rv_init_cp_jump_table(adev); + amdgpu_bo_kunmap(adev->gfx.rlc.cp_table_obj); + amdgpu_bo_unreserve(adev->gfx.rlc.cp_table_obj); } switch (adev->asic_type) { @@ -3527,47 +3613,64 @@ static int gfx_v9_0_late_init(void *handle) return 0; } -static bool gfx_v9_0_is_rlc_enabled(struct amdgpu_device *adev) +static void gfx_v9_0_enter_rlc_safe_mode(struct amdgpu_device *adev) { - uint32_t rlc_setting; + uint32_t rlc_setting, data; + unsigned i; + + if (adev->gfx.rlc.in_safe_mode) + return; /* if RLC is not enabled, do nothing */ rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL); if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK)) - return false; - - return true; -} - -static void gfx_v9_0_set_safe_mode(struct amdgpu_device *adev) -{ - uint32_t data; - unsigned i; + return; - data = RLC_SAFE_MODE__CMD_MASK; - data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); - WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); + if (adev->cg_flags & + (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG | + AMD_CG_SUPPORT_GFX_3D_CGCG)) { + data = RLC_SAFE_MODE__CMD_MASK; + data |= (1 << RLC_SAFE_MODE__MESSAGE__SHIFT); + WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); - /* wait for RLC_SAFE_MODE */ - for (i = 0; i < adev->usec_timeout; i++) { - if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) - break; - udelay(1); + /* wait for RLC_SAFE_MODE */ + for (i = 0; i < adev->usec_timeout; i++) { + if (!REG_GET_FIELD(RREG32_SOC15(GC, 0, mmRLC_SAFE_MODE), RLC_SAFE_MODE, CMD)) + break; + udelay(1); + } + adev->gfx.rlc.in_safe_mode = true; } } -static void gfx_v9_0_unset_safe_mode(struct amdgpu_device *adev) +static void gfx_v9_0_exit_rlc_safe_mode(struct amdgpu_device *adev) { - uint32_t data; + uint32_t rlc_setting, data; + + if (!adev->gfx.rlc.in_safe_mode) + return; + + /* if RLC is not enabled, do nothing */ + rlc_setting = RREG32_SOC15(GC, 0, mmRLC_CNTL); + if (!(rlc_setting & RLC_CNTL__RLC_ENABLE_F32_MASK)) + return; - data = RLC_SAFE_MODE__CMD_MASK; - WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); + if (adev->cg_flags & + (AMD_CG_SUPPORT_GFX_CGCG | AMD_CG_SUPPORT_GFX_MGCG)) { + /* + * Try to exit safe mode only if it is already in safe + * mode. + */ + data = RLC_SAFE_MODE__CMD_MASK; + WREG32_SOC15(GC, 0, mmRLC_SAFE_MODE, data); + adev->gfx.rlc.in_safe_mode = false; + } } static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev, bool enable) { - amdgpu_gfx_rlc_enter_safe_mode(adev); + gfx_v9_0_enter_rlc_safe_mode(adev); if ((adev->pg_flags & AMD_PG_SUPPORT_GFX_PG) && enable) { gfx_v9_0_enable_gfx_cg_power_gating(adev, true); @@ -3578,7 +3681,7 @@ static void gfx_v9_0_update_gfx_cg_power_gating(struct amdgpu_device *adev, gfx_v9_0_enable_gfx_pipeline_powergating(adev, false); } - amdgpu_gfx_rlc_exit_safe_mode(adev); + gfx_v9_0_exit_rlc_safe_mode(adev); } static void gfx_v9_0_update_gfx_mg_power_gating(struct amdgpu_device *adev, @@ -3676,7 +3779,7 @@ static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev, { uint32_t data, def; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); /* Enable 3D CGCG/CGLS */ if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_3D_CGCG)) { @@ -3716,7 +3819,7 @@ static void gfx_v9_0_update_3d_clock_gating(struct amdgpu_device *adev, WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL_3D, data); } - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); } static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev, @@ -3724,7 +3827,7 @@ static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev { uint32_t def, data; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); if (enable && (adev->cg_flags & AMD_CG_SUPPORT_GFX_CGCG)) { def = data = RREG32_SOC15(GC, 0, mmRLC_CGTT_MGCG_OVERRIDE); @@ -3764,7 +3867,7 @@ static void gfx_v9_0_update_coarse_grain_clock_gating(struct amdgpu_device *adev WREG32_SOC15(GC, 0, mmRLC_CGCG_CGLS_CTRL, data); } - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); } static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev, @@ -3793,12 +3896,8 @@ static int gfx_v9_0_update_gfx_clock_gating(struct amdgpu_device *adev, } static const struct amdgpu_rlc_funcs gfx_v9_0_rlc_funcs = { - .is_rlc_enabled = gfx_v9_0_is_rlc_enabled, - .set_safe_mode = gfx_v9_0_set_safe_mode, - .unset_safe_mode = gfx_v9_0_unset_safe_mode, - .get_csb_size = gfx_v9_0_get_csb_size, - .get_csb_buffer = gfx_v9_0_get_csb_buffer, - .get_cp_table_num = gfx_v9_0_cp_jump_table_num, + .enter_safe_mode = gfx_v9_0_enter_rlc_safe_mode, + .exit_safe_mode = gfx_v9_0_exit_rlc_safe_mode }; static int gfx_v9_0_set_powergating_state(void *handle, diff --git a/drivers/gpu/drm/amd/amdgpu/kv_dpm.c b/drivers/gpu/drm/amd/amdgpu/kv_dpm.c index 36bcba96..faf06fd 100644 --- a/drivers/gpu/drm/amd/amdgpu/kv_dpm.c +++ b/drivers/gpu/drm/amd/amdgpu/kv_dpm.c @@ -508,19 +508,19 @@ static int kv_enable_didt(struct amdgpu_device *adev, bool enable) pi->caps_db_ramping || pi->caps_td_ramping || pi->caps_tcp_ramping) { - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); if (enable) { ret = kv_program_pt_config_registers(adev, didt_config_kv); if (ret) { - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return ret; } } kv_do_enable_didt(adev, enable); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); } return 0; diff --git a/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_powertune.c b/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_powertune.c index d138ddae..5e19f59 100644 --- a/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_powertune.c +++ b/drivers/gpu/drm/amd/powerplay/hwmgr/smu7_powertune.c @@ -967,7 +967,7 @@ int smu7_enable_didt_config(struct pp_hwmgr *hwmgr) PP_CAP(PHM_PlatformCaps_TDRamping) || PP_CAP(PHM_PlatformCaps_TCPRamping)) { - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); mutex_lock(&adev->grbm_idx_mutex); value = 0; value2 = cgs_read_register(hwmgr->device, mmGRBM_GFX_INDEX); @@ -1014,13 +1014,13 @@ int smu7_enable_didt_config(struct pp_hwmgr *hwmgr) "Failed to enable DPM DIDT.", goto error); } mutex_unlock(&adev->grbm_idx_mutex); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); } return 0; error: mutex_unlock(&adev->grbm_idx_mutex); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return result; } @@ -1034,7 +1034,7 @@ int smu7_disable_didt_config(struct pp_hwmgr *hwmgr) PP_CAP(PHM_PlatformCaps_TDRamping) || PP_CAP(PHM_PlatformCaps_TCPRamping)) { - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); result = smu7_enable_didt(hwmgr, false); PP_ASSERT_WITH_CODE((result == 0), @@ -1046,12 +1046,12 @@ int smu7_disable_didt_config(struct pp_hwmgr *hwmgr) PP_ASSERT_WITH_CODE((0 == result), "Failed to disable DPM DIDT.", goto error); } - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); } return 0; error: - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return result; } diff --git a/drivers/gpu/drm/amd/powerplay/hwmgr/vega10_powertune.c b/drivers/gpu/drm/amd/powerplay/hwmgr/vega10_powertune.c index 6f26cb2..2d88abf 100644 --- a/drivers/gpu/drm/amd/powerplay/hwmgr/vega10_powertune.c +++ b/drivers/gpu/drm/amd/powerplay/hwmgr/vega10_powertune.c @@ -937,7 +937,7 @@ static int vega10_enable_cac_driving_se_didt_config(struct pp_hwmgr *hwmgr) num_se = adev->gfx.config.max_shader_engines; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); mutex_lock(&adev->grbm_idx_mutex); for (count = 0; count < num_se; count++) { @@ -962,7 +962,7 @@ static int vega10_enable_cac_driving_se_didt_config(struct pp_hwmgr *hwmgr) vega10_didt_set_mask(hwmgr, true); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return 0; } @@ -971,11 +971,11 @@ static int vega10_disable_cac_driving_se_didt_config(struct pp_hwmgr *hwmgr) { struct amdgpu_device *adev = hwmgr->adev; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); vega10_didt_set_mask(hwmgr, false); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return 0; } @@ -988,7 +988,7 @@ static int vega10_enable_psm_gc_didt_config(struct pp_hwmgr *hwmgr) num_se = adev->gfx.config.max_shader_engines; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); mutex_lock(&adev->grbm_idx_mutex); for (count = 0; count < num_se; count++) { @@ -1007,7 +1007,7 @@ static int vega10_enable_psm_gc_didt_config(struct pp_hwmgr *hwmgr) vega10_didt_set_mask(hwmgr, true); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); vega10_program_gc_didt_config_registers(hwmgr, GCDiDtDroopCtrlConfig_vega10); if (PP_CAP(PHM_PlatformCaps_GCEDC)) @@ -1024,11 +1024,11 @@ static int vega10_disable_psm_gc_didt_config(struct pp_hwmgr *hwmgr) struct amdgpu_device *adev = hwmgr->adev; uint32_t data; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); vega10_didt_set_mask(hwmgr, false); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); if (PP_CAP(PHM_PlatformCaps_GCEDC)) { data = 0x00000000; @@ -1049,7 +1049,7 @@ static int vega10_enable_se_edc_config(struct pp_hwmgr *hwmgr) num_se = adev->gfx.config.max_shader_engines; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); mutex_lock(&adev->grbm_idx_mutex); for (count = 0; count < num_se; count++) { @@ -1070,7 +1070,7 @@ static int vega10_enable_se_edc_config(struct pp_hwmgr *hwmgr) vega10_didt_set_mask(hwmgr, true); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return 0; } @@ -1079,11 +1079,11 @@ static int vega10_disable_se_edc_config(struct pp_hwmgr *hwmgr) { struct amdgpu_device *adev = hwmgr->adev; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); vega10_didt_set_mask(hwmgr, false); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return 0; } @@ -1097,7 +1097,7 @@ static int vega10_enable_psm_gc_edc_config(struct pp_hwmgr *hwmgr) num_se = adev->gfx.config.max_shader_engines; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); vega10_program_gc_didt_config_registers(hwmgr, AvfsPSMResetConfig_vega10); @@ -1118,7 +1118,7 @@ static int vega10_enable_psm_gc_edc_config(struct pp_hwmgr *hwmgr) vega10_didt_set_mask(hwmgr, true); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); vega10_program_gc_didt_config_registers(hwmgr, PSMGCEDCDroopCtrlConfig_vega10); @@ -1138,11 +1138,11 @@ static int vega10_disable_psm_gc_edc_config(struct pp_hwmgr *hwmgr) struct amdgpu_device *adev = hwmgr->adev; uint32_t data; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); vega10_didt_set_mask(hwmgr, false); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); if (PP_CAP(PHM_PlatformCaps_GCEDC)) { data = 0x00000000; @@ -1160,7 +1160,7 @@ static int vega10_enable_se_edc_force_stall_config(struct pp_hwmgr *hwmgr) struct amdgpu_device *adev = hwmgr->adev; int result; - amdgpu_gfx_rlc_enter_safe_mode(adev); + adev->gfx.rlc.funcs->enter_safe_mode(adev); mutex_lock(&adev->grbm_idx_mutex); WREG32_SOC15(GC, 0, mmGRBM_GFX_INDEX, 0xE0000000); @@ -1173,7 +1173,7 @@ static int vega10_enable_se_edc_force_stall_config(struct pp_hwmgr *hwmgr) vega10_didt_set_mask(hwmgr, false); - amdgpu_gfx_rlc_exit_safe_mode(adev); + adev->gfx.rlc.funcs->exit_safe_mode(adev); return 0; } -- 2.7.4