drm/amdgpu: Add amdgpu <--> amdkfd gfx8 interface
authorBen Goz <ben.goz@amd.com>
Tue, 7 Oct 2014 11:43:07 +0000 (14:43 +0300)
committerOded Gabbay <oded.gabbay@gmail.com>
Mon, 20 Jul 2015 06:16:48 +0000 (09:16 +0300)
This patch adds the gfx8 interface file between amdgpu and amdkfd. This
interface file is currently in use when running on a Carrizo-based
system.

The interface itself is represented by a pointer to struct
kfd_dev. The pointer is located inside amdgpu_device structure.

All the register accesses that amdkfd need are done using this
interface. This allows us to avoid direct register accesses in
amdkfd proper, while also allows us to avoid locking between
amdkfd and amdgpu.

The single exception is the doorbells that are used in both of
the drivers. However, because they are located in separate pci
bar pages, the danger of sharing registers between the drivers
is minimal.

Having said that, we are planning to move the doorbells as well
to amdgpu.

Signed-off-by: Ben Goz <ben.goz@amd.com>
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
MAINTAINERS
drivers/gpu/drm/amd/amdgpu/Makefile
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c [new file with mode: 0644]
drivers/gpu/drm/amd/amdgpu/vid.h
drivers/gpu/drm/amd/include/vi_structs.h [new file with mode: 0644]

index 5cc0734..9c9dd5f 100644 (file)
@@ -639,9 +639,11 @@ S: Supported
 F:     drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.c
 F:     drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd.h
 F:     drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v7.c
+F:     drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
 F:     drivers/gpu/drm/amd/amdkfd/
 F:     drivers/gpu/drm/amd/include/cik_structs.h
 F:     drivers/gpu/drm/amd/include/kgd_kfd_interface.h
+F:     drivers/gpu/drm/amd/include/vi_structs.h
 F:     drivers/gpu/drm/radeon/radeon_kfd.c
 F:     drivers/gpu/drm/radeon/radeon_kfd.h
 F:     include/uapi/linux/kfd_ioctl.h
index af5397c..9083605 100644 (file)
@@ -74,7 +74,8 @@ amdgpu-y += \
 # add amdkfd interfaces
 amdgpu-y += \
         amdgpu_amdkfd.o \
-        amdgpu_amdkfd_gfx_v7.o
+        amdgpu_amdkfd_gfx_v7.o \
+        amdgpu_amdkfd_gfx_v8.o
 
 amdgpu-$(CONFIG_COMPAT) += amdgpu_ioc32.o
 amdgpu-$(CONFIG_VGA_SWITCHEROO) += amdgpu_atpx_handler.o
index 7aa5ab0..bc763e0 100644 (file)
@@ -53,6 +53,9 @@ bool amdgpu_amdkfd_load_interface(struct amdgpu_device *rdev)
        case CHIP_KAVERI:
                kfd2kgd = amdgpu_amdkfd_gfx_7_get_functions();
                break;
+       case CHIP_CARRIZO:
+               kfd2kgd = amdgpu_amdkfd_gfx_8_0_get_functions();
+               break;
        default:
                return false;
        }
index c81242e..a8be765 100644 (file)
@@ -50,6 +50,7 @@ void amdgpu_amdkfd_device_init(struct amdgpu_device *rdev);
 void amdgpu_amdkfd_device_fini(struct amdgpu_device *rdev);
 
 struct kfd2kgd_calls *amdgpu_amdkfd_gfx_7_get_functions(void);
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions(void);
 
 /* Shared API */
 int alloc_gtt_mem(struct kgd_dev *kgd, size_t size,
diff --git a/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c b/drivers/gpu/drm/amd/amdgpu/amdgpu_amdkfd_gfx_v8.c
new file mode 100644 (file)
index 0000000..dfd1d50
--- /dev/null
@@ -0,0 +1,543 @@
+/*
+ * Copyright 2014 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/fdtable.h>
+#include <linux/uaccess.h>
+#include <linux/firmware.h>
+#include <drm/drmP.h>
+#include "amdgpu.h"
+#include "amdgpu_amdkfd.h"
+#include "amdgpu_ucode.h"
+#include "gca/gfx_8_0_sh_mask.h"
+#include "gca/gfx_8_0_d.h"
+#include "gca/gfx_8_0_enum.h"
+#include "oss/oss_3_0_sh_mask.h"
+#include "oss/oss_3_0_d.h"
+#include "gmc/gmc_8_1_sh_mask.h"
+#include "gmc/gmc_8_1_d.h"
+#include "vi_structs.h"
+#include "vid.h"
+
+#define VI_PIPE_PER_MEC        (4)
+
+struct cik_sdma_rlc_registers;
+
+/*
+ * Register access functions
+ */
+
+static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+               uint32_t sh_mem_config,
+               uint32_t sh_mem_ape1_base, uint32_t sh_mem_ape1_limit,
+               uint32_t sh_mem_bases);
+static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
+               unsigned int vmid);
+static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
+               uint32_t hpd_size, uint64_t hpd_gpu_addr);
+static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id);
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+               uint32_t queue_id, uint32_t __user *wptr);
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd);
+static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
+               uint32_t pipe_id, uint32_t queue_id);
+static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd);
+static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+                               unsigned int timeout, uint32_t pipe_id,
+                               uint32_t queue_id);
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+                               unsigned int timeout);
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
+static int kgd_address_watch_disable(struct kgd_dev *kgd);
+static int kgd_address_watch_execute(struct kgd_dev *kgd,
+                                       unsigned int watch_point_id,
+                                       uint32_t cntl_val,
+                                       uint32_t addr_hi,
+                                       uint32_t addr_lo);
+static int kgd_wave_control_execute(struct kgd_dev *kgd,
+                                       uint32_t gfx_index_val,
+                                       uint32_t sq_cmd);
+static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
+                                       unsigned int watch_point_id,
+                                       unsigned int reg_offset);
+
+static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
+               uint8_t vmid);
+static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
+               uint8_t vmid);
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid);
+static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type);
+
+static const struct kfd2kgd_calls kfd2kgd = {
+       .init_gtt_mem_allocation = alloc_gtt_mem,
+       .free_gtt_mem = free_gtt_mem,
+       .get_vmem_size = get_vmem_size,
+       .get_gpu_clock_counter = get_gpu_clock_counter,
+       .get_max_engine_clock_in_mhz = get_max_engine_clock_in_mhz,
+       .program_sh_mem_settings = kgd_program_sh_mem_settings,
+       .set_pasid_vmid_mapping = kgd_set_pasid_vmid_mapping,
+       .init_pipeline = kgd_init_pipeline,
+       .init_interrupts = kgd_init_interrupts,
+       .hqd_load = kgd_hqd_load,
+       .hqd_sdma_load = kgd_hqd_sdma_load,
+       .hqd_is_occupied = kgd_hqd_is_occupied,
+       .hqd_sdma_is_occupied = kgd_hqd_sdma_is_occupied,
+       .hqd_destroy = kgd_hqd_destroy,
+       .hqd_sdma_destroy = kgd_hqd_sdma_destroy,
+       .address_watch_disable = kgd_address_watch_disable,
+       .address_watch_execute = kgd_address_watch_execute,
+       .wave_control_execute = kgd_wave_control_execute,
+       .address_watch_get_offset = kgd_address_watch_get_offset,
+       .get_atc_vmid_pasid_mapping_pasid =
+                       get_atc_vmid_pasid_mapping_pasid,
+       .get_atc_vmid_pasid_mapping_valid =
+                       get_atc_vmid_pasid_mapping_valid,
+       .write_vmid_invalidate_request = write_vmid_invalidate_request,
+       .get_fw_version = get_fw_version
+};
+
+struct kfd2kgd_calls *amdgpu_amdkfd_gfx_8_0_get_functions()
+{
+       return (struct kfd2kgd_calls *)&kfd2kgd;
+}
+
+static inline struct amdgpu_device *get_amdgpu_device(struct kgd_dev *kgd)
+{
+       return (struct amdgpu_device *)kgd;
+}
+
+static void lock_srbm(struct kgd_dev *kgd, uint32_t mec, uint32_t pipe,
+                       uint32_t queue, uint32_t vmid)
+{
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+       uint32_t value = PIPEID(pipe) | MEID(mec) | VMID(vmid) | QUEUEID(queue);
+
+       mutex_lock(&adev->srbm_mutex);
+       WREG32(mmSRBM_GFX_CNTL, value);
+}
+
+static void unlock_srbm(struct kgd_dev *kgd)
+{
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+       WREG32(mmSRBM_GFX_CNTL, 0);
+       mutex_unlock(&adev->srbm_mutex);
+}
+
+static void acquire_queue(struct kgd_dev *kgd, uint32_t pipe_id,
+                               uint32_t queue_id)
+{
+       uint32_t mec = (++pipe_id / VI_PIPE_PER_MEC) + 1;
+       uint32_t pipe = (pipe_id % VI_PIPE_PER_MEC);
+
+       lock_srbm(kgd, mec, pipe, queue_id, 0);
+}
+
+static void release_queue(struct kgd_dev *kgd)
+{
+       unlock_srbm(kgd);
+}
+
+static void kgd_program_sh_mem_settings(struct kgd_dev *kgd, uint32_t vmid,
+                                       uint32_t sh_mem_config,
+                                       uint32_t sh_mem_ape1_base,
+                                       uint32_t sh_mem_ape1_limit,
+                                       uint32_t sh_mem_bases)
+{
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+       lock_srbm(kgd, 0, 0, 0, vmid);
+
+       WREG32(mmSH_MEM_CONFIG, sh_mem_config);
+       WREG32(mmSH_MEM_APE1_BASE, sh_mem_ape1_base);
+       WREG32(mmSH_MEM_APE1_LIMIT, sh_mem_ape1_limit);
+       WREG32(mmSH_MEM_BASES, sh_mem_bases);
+
+       unlock_srbm(kgd);
+}
+
+static int kgd_set_pasid_vmid_mapping(struct kgd_dev *kgd, unsigned int pasid,
+                                       unsigned int vmid)
+{
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+       /*
+        * We have to assume that there is no outstanding mapping.
+        * The ATC_VMID_PASID_MAPPING_UPDATE_STATUS bit could be 0 because
+        * a mapping is in progress or because a mapping finished
+        * and the SW cleared it.
+        * So the protocol is to always wait & clear.
+        */
+       uint32_t pasid_mapping = (pasid == 0) ? 0 : (uint32_t)pasid |
+                       ATC_VMID0_PASID_MAPPING__VALID_MASK;
+
+       WREG32(mmATC_VMID0_PASID_MAPPING + vmid, pasid_mapping);
+
+       while (!(RREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS) & (1U << vmid)))
+               cpu_relax();
+       WREG32(mmATC_VMID_PASID_MAPPING_UPDATE_STATUS, 1U << vmid);
+
+       /* Mapping vmid to pasid also for IH block */
+       WREG32(mmIH_VMID_0_LUT + vmid, pasid_mapping);
+
+       return 0;
+}
+
+static int kgd_init_pipeline(struct kgd_dev *kgd, uint32_t pipe_id,
+                               uint32_t hpd_size, uint64_t hpd_gpu_addr)
+{
+       return 0;
+}
+
+static int kgd_init_interrupts(struct kgd_dev *kgd, uint32_t pipe_id)
+{
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+       uint32_t mec;
+       uint32_t pipe;
+
+       mec = (++pipe_id / VI_PIPE_PER_MEC) + 1;
+       pipe = (pipe_id % VI_PIPE_PER_MEC);
+
+       lock_srbm(kgd, mec, pipe, 0, 0);
+
+       WREG32(mmCPC_INT_CNTL, CP_INT_CNTL_RING0__TIME_STAMP_INT_ENABLE_MASK);
+
+       unlock_srbm(kgd);
+
+       return 0;
+}
+
+static inline uint32_t get_sdma_base_addr(struct cik_sdma_rlc_registers *m)
+{
+       return 0;
+}
+
+static inline struct vi_mqd *get_mqd(void *mqd)
+{
+       return (struct vi_mqd *)mqd;
+}
+
+static inline struct cik_sdma_rlc_registers *get_sdma_mqd(void *mqd)
+{
+       return (struct cik_sdma_rlc_registers *)mqd;
+}
+
+static int kgd_hqd_load(struct kgd_dev *kgd, void *mqd, uint32_t pipe_id,
+                       uint32_t queue_id, uint32_t __user *wptr)
+{
+       struct vi_mqd *m;
+       uint32_t shadow_wptr, valid_wptr;
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+
+       m = get_mqd(mqd);
+
+       valid_wptr = copy_from_user(&shadow_wptr, wptr, sizeof(shadow_wptr));
+       acquire_queue(kgd, pipe_id, queue_id);
+
+       WREG32(mmCP_MQD_CONTROL, m->cp_mqd_control);
+       WREG32(mmCP_MQD_BASE_ADDR, m->cp_mqd_base_addr_lo);
+       WREG32(mmCP_MQD_BASE_ADDR_HI, m->cp_mqd_base_addr_hi);
+
+       WREG32(mmCP_HQD_VMID, m->cp_hqd_vmid);
+       WREG32(mmCP_HQD_PERSISTENT_STATE, m->cp_hqd_persistent_state);
+       WREG32(mmCP_HQD_PIPE_PRIORITY, m->cp_hqd_pipe_priority);
+       WREG32(mmCP_HQD_QUEUE_PRIORITY, m->cp_hqd_queue_priority);
+       WREG32(mmCP_HQD_QUANTUM, m->cp_hqd_quantum);
+       WREG32(mmCP_HQD_PQ_BASE, m->cp_hqd_pq_base_lo);
+       WREG32(mmCP_HQD_PQ_BASE_HI, m->cp_hqd_pq_base_hi);
+       WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR, m->cp_hqd_pq_rptr_report_addr_lo);
+       WREG32(mmCP_HQD_PQ_RPTR_REPORT_ADDR_HI,
+                       m->cp_hqd_pq_rptr_report_addr_hi);
+
+       if (valid_wptr > 0)
+               WREG32(mmCP_HQD_PQ_WPTR, shadow_wptr);
+
+       WREG32(mmCP_HQD_PQ_CONTROL, m->cp_hqd_pq_control);
+       WREG32(mmCP_HQD_PQ_DOORBELL_CONTROL, m->cp_hqd_pq_doorbell_control);
+
+       WREG32(mmCP_HQD_EOP_BASE_ADDR, m->cp_hqd_eop_base_addr_lo);
+       WREG32(mmCP_HQD_EOP_BASE_ADDR_HI, m->cp_hqd_eop_base_addr_hi);
+       WREG32(mmCP_HQD_EOP_CONTROL, m->cp_hqd_eop_control);
+       WREG32(mmCP_HQD_EOP_RPTR, m->cp_hqd_eop_rptr);
+       WREG32(mmCP_HQD_EOP_WPTR, m->cp_hqd_eop_wptr);
+       WREG32(mmCP_HQD_EOP_EVENTS, m->cp_hqd_eop_done_events);
+
+       WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_LO, m->cp_hqd_ctx_save_base_addr_lo);
+       WREG32(mmCP_HQD_CTX_SAVE_BASE_ADDR_HI, m->cp_hqd_ctx_save_base_addr_hi);
+       WREG32(mmCP_HQD_CTX_SAVE_CONTROL, m->cp_hqd_ctx_save_control);
+       WREG32(mmCP_HQD_CNTL_STACK_OFFSET, m->cp_hqd_cntl_stack_offset);
+       WREG32(mmCP_HQD_CNTL_STACK_SIZE, m->cp_hqd_cntl_stack_size);
+       WREG32(mmCP_HQD_WG_STATE_OFFSET, m->cp_hqd_wg_state_offset);
+       WREG32(mmCP_HQD_CTX_SAVE_SIZE, m->cp_hqd_ctx_save_size);
+
+       WREG32(mmCP_HQD_IB_CONTROL, m->cp_hqd_ib_control);
+
+       WREG32(mmCP_HQD_DEQUEUE_REQUEST, m->cp_hqd_dequeue_request);
+       WREG32(mmCP_HQD_ERROR, m->cp_hqd_error);
+       WREG32(mmCP_HQD_EOP_WPTR_MEM, m->cp_hqd_eop_wptr_mem);
+       WREG32(mmCP_HQD_EOP_DONES, m->cp_hqd_eop_dones);
+
+       WREG32(mmCP_HQD_ACTIVE, m->cp_hqd_active);
+
+       release_queue(kgd);
+
+       return 0;
+}
+
+static int kgd_hqd_sdma_load(struct kgd_dev *kgd, void *mqd)
+{
+       return 0;
+}
+
+static bool kgd_hqd_is_occupied(struct kgd_dev *kgd, uint64_t queue_address,
+                               uint32_t pipe_id, uint32_t queue_id)
+{
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+       uint32_t act;
+       bool retval = false;
+       uint32_t low, high;
+
+       acquire_queue(kgd, pipe_id, queue_id);
+       act = RREG32(mmCP_HQD_ACTIVE);
+       if (act) {
+               low = lower_32_bits(queue_address >> 8);
+               high = upper_32_bits(queue_address >> 8);
+
+               if (low == RREG32(mmCP_HQD_PQ_BASE) &&
+                               high == RREG32(mmCP_HQD_PQ_BASE_HI))
+                       retval = true;
+       }
+       release_queue(kgd);
+       return retval;
+}
+
+static bool kgd_hqd_sdma_is_occupied(struct kgd_dev *kgd, void *mqd)
+{
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+       struct cik_sdma_rlc_registers *m;
+       uint32_t sdma_base_addr;
+       uint32_t sdma_rlc_rb_cntl;
+
+       m = get_sdma_mqd(mqd);
+       sdma_base_addr = get_sdma_base_addr(m);
+
+       sdma_rlc_rb_cntl = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+
+       if (sdma_rlc_rb_cntl & SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK)
+               return true;
+
+       return false;
+}
+
+static int kgd_hqd_destroy(struct kgd_dev *kgd, uint32_t reset_type,
+                               unsigned int timeout, uint32_t pipe_id,
+                               uint32_t queue_id)
+{
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+       uint32_t temp;
+
+       acquire_queue(kgd, pipe_id, queue_id);
+
+       WREG32(mmCP_HQD_DEQUEUE_REQUEST, reset_type);
+
+       while (true) {
+               temp = RREG32(mmCP_HQD_ACTIVE);
+               if (temp & CP_HQD_ACTIVE__ACTIVE_MASK)
+                       break;
+               if (timeout == 0) {
+                       pr_err("kfd: cp queue preemption time out (%dms)\n",
+                               temp);
+                       release_queue(kgd);
+                       return -ETIME;
+               }
+               msleep(20);
+               timeout -= 20;
+       }
+
+       release_queue(kgd);
+       return 0;
+}
+
+static int kgd_hqd_sdma_destroy(struct kgd_dev *kgd, void *mqd,
+                               unsigned int timeout)
+{
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+       struct cik_sdma_rlc_registers *m;
+       uint32_t sdma_base_addr;
+       uint32_t temp;
+
+       m = get_sdma_mqd(mqd);
+       sdma_base_addr = get_sdma_base_addr(m);
+
+       temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL);
+       temp = temp & ~SDMA0_RLC0_RB_CNTL__RB_ENABLE_MASK;
+       WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_CNTL, temp);
+
+       while (true) {
+               temp = RREG32(sdma_base_addr + mmSDMA0_RLC0_CONTEXT_STATUS);
+               if (temp & SDMA0_STATUS_REG__RB_CMD_IDLE__SHIFT)
+                       break;
+               if (timeout == 0)
+                       return -ETIME;
+               msleep(20);
+               timeout -= 20;
+       }
+
+       WREG32(sdma_base_addr + mmSDMA0_RLC0_DOORBELL, 0);
+       WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_RPTR, 0);
+       WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_WPTR, 0);
+       WREG32(sdma_base_addr + mmSDMA0_RLC0_RB_BASE, 0);
+
+       return 0;
+}
+
+static bool get_atc_vmid_pasid_mapping_valid(struct kgd_dev *kgd,
+                                                       uint8_t vmid)
+{
+       uint32_t reg;
+       struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+       reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
+       return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+}
+
+static uint16_t get_atc_vmid_pasid_mapping_pasid(struct kgd_dev *kgd,
+                                                               uint8_t vmid)
+{
+       uint32_t reg;
+       struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+       reg = RREG32(mmATC_VMID0_PASID_MAPPING + vmid);
+       return reg & ATC_VMID0_PASID_MAPPING__VALID_MASK;
+}
+
+static void write_vmid_invalidate_request(struct kgd_dev *kgd, uint8_t vmid)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+
+       WREG32(mmVM_INVALIDATE_REQUEST, 1 << vmid);
+}
+
+static int kgd_address_watch_disable(struct kgd_dev *kgd)
+{
+       return 0;
+}
+
+static int kgd_address_watch_execute(struct kgd_dev *kgd,
+                                       unsigned int watch_point_id,
+                                       uint32_t cntl_val,
+                                       uint32_t addr_hi,
+                                       uint32_t addr_lo)
+{
+       return 0;
+}
+
+static int kgd_wave_control_execute(struct kgd_dev *kgd,
+                                       uint32_t gfx_index_val,
+                                       uint32_t sq_cmd)
+{
+       struct amdgpu_device *adev = get_amdgpu_device(kgd);
+       uint32_t data = 0;
+
+       mutex_lock(&adev->grbm_idx_mutex);
+
+       WREG32(mmGRBM_GFX_INDEX, gfx_index_val);
+       WREG32(mmSQ_CMD, sq_cmd);
+
+       data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+               INSTANCE_BROADCAST_WRITES, 1);
+       data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+               SH_BROADCAST_WRITES, 1);
+       data = REG_SET_FIELD(data, GRBM_GFX_INDEX,
+               SE_BROADCAST_WRITES, 1);
+
+       WREG32(mmGRBM_GFX_INDEX, data);
+       mutex_unlock(&adev->grbm_idx_mutex);
+
+       return 0;
+}
+
+static uint32_t kgd_address_watch_get_offset(struct kgd_dev *kgd,
+                                       unsigned int watch_point_id,
+                                       unsigned int reg_offset)
+{
+       return 0;
+}
+
+static uint16_t get_fw_version(struct kgd_dev *kgd, enum kgd_engine_type type)
+{
+       struct amdgpu_device *adev = (struct amdgpu_device *) kgd;
+       const union amdgpu_firmware_header *hdr;
+
+       BUG_ON(kgd == NULL);
+
+       switch (type) {
+       case KGD_ENGINE_PFP:
+               hdr = (const union amdgpu_firmware_header *)
+                                                       adev->gfx.pfp_fw->data;
+               break;
+
+       case KGD_ENGINE_ME:
+               hdr = (const union amdgpu_firmware_header *)
+                                                       adev->gfx.me_fw->data;
+               break;
+
+       case KGD_ENGINE_CE:
+               hdr = (const union amdgpu_firmware_header *)
+                                                       adev->gfx.ce_fw->data;
+               break;
+
+       case KGD_ENGINE_MEC1:
+               hdr = (const union amdgpu_firmware_header *)
+                                                       adev->gfx.mec_fw->data;
+               break;
+
+       case KGD_ENGINE_MEC2:
+               hdr = (const union amdgpu_firmware_header *)
+                                                       adev->gfx.mec2_fw->data;
+               break;
+
+       case KGD_ENGINE_RLC:
+               hdr = (const union amdgpu_firmware_header *)
+                                                       adev->gfx.rlc_fw->data;
+               break;
+
+       case KGD_ENGINE_SDMA1:
+               hdr = (const union amdgpu_firmware_header *)
+                                                       adev->sdma[0].fw->data;
+               break;
+
+       case KGD_ENGINE_SDMA2:
+               hdr = (const union amdgpu_firmware_header *)
+                                                       adev->sdma[1].fw->data;
+               break;
+
+       default:
+               return 0;
+       }
+
+       if (hdr == NULL)
+               return 0;
+
+       /* Only 12 bit in use*/
+       return hdr->common.ucode_version;
+}
index 31bb894..d98aa9d 100644 (file)
 
 #define AMDGPU_NUM_OF_VMIDS                    8
 
+#define                PIPEID(x)                                       ((x) << 0)
+#define                MEID(x)                                         ((x) << 2)
+#define                VMID(x)                                         ((x) << 4)
+#define                QUEUEID(x)                                      ((x) << 8)
+
 #define RB_BITMAP_WIDTH_PER_SH     2
 
 #define MC_SEQ_MISC0__MT__MASK 0xf0000000
diff --git a/drivers/gpu/drm/amd/include/vi_structs.h b/drivers/gpu/drm/amd/include/vi_structs.h
new file mode 100644 (file)
index 0000000..65cfacd
--- /dev/null
@@ -0,0 +1,417 @@
+/*
+ * Copyright 2012 Advanced Micro Devices, Inc.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included in
+ * all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
+ * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
+ * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
+ * OTHER DEALINGS IN THE SOFTWARE.
+ *
+ */
+
+#ifndef VI_STRUCTS_H_
+#define VI_STRUCTS_H_
+
+struct vi_sdma_mqd {
+       uint32_t sdmax_rlcx_rb_cntl;
+       uint32_t sdmax_rlcx_rb_base;
+       uint32_t sdmax_rlcx_rb_base_hi;
+       uint32_t sdmax_rlcx_rb_rptr;
+       uint32_t sdmax_rlcx_rb_wptr;
+       uint32_t sdmax_rlcx_rb_wptr_poll_cntl;
+       uint32_t sdmax_rlcx_rb_wptr_poll_addr_hi;
+       uint32_t sdmax_rlcx_rb_wptr_poll_addr_lo;
+       uint32_t sdmax_rlcx_rb_rptr_addr_hi;
+       uint32_t sdmax_rlcx_rb_rptr_addr_lo;
+       uint32_t sdmax_rlcx_ib_cntl;
+       uint32_t sdmax_rlcx_ib_rptr;
+       uint32_t sdmax_rlcx_ib_offset;
+       uint32_t sdmax_rlcx_ib_base_lo;
+       uint32_t sdmax_rlcx_ib_base_hi;
+       uint32_t sdmax_rlcx_ib_size;
+       uint32_t sdmax_rlcx_skip_cntl;
+       uint32_t sdmax_rlcx_context_status;
+       uint32_t sdmax_rlcx_doorbell;
+       uint32_t sdmax_rlcx_virtual_addr;
+       uint32_t sdmax_rlcx_ape1_cntl;
+       uint32_t sdmax_rlcx_doorbell_log;
+       uint32_t reserved_22;
+       uint32_t reserved_23;
+       uint32_t reserved_24;
+       uint32_t reserved_25;
+       uint32_t reserved_26;
+       uint32_t reserved_27;
+       uint32_t reserved_28;
+       uint32_t reserved_29;
+       uint32_t reserved_30;
+       uint32_t reserved_31;
+       uint32_t reserved_32;
+       uint32_t reserved_33;
+       uint32_t reserved_34;
+       uint32_t reserved_35;
+       uint32_t reserved_36;
+       uint32_t reserved_37;
+       uint32_t reserved_38;
+       uint32_t reserved_39;
+       uint32_t reserved_40;
+       uint32_t reserved_41;
+       uint32_t reserved_42;
+       uint32_t reserved_43;
+       uint32_t reserved_44;
+       uint32_t reserved_45;
+       uint32_t reserved_46;
+       uint32_t reserved_47;
+       uint32_t reserved_48;
+       uint32_t reserved_49;
+       uint32_t reserved_50;
+       uint32_t reserved_51;
+       uint32_t reserved_52;
+       uint32_t reserved_53;
+       uint32_t reserved_54;
+       uint32_t reserved_55;
+       uint32_t reserved_56;
+       uint32_t reserved_57;
+       uint32_t reserved_58;
+       uint32_t reserved_59;
+       uint32_t reserved_60;
+       uint32_t reserved_61;
+       uint32_t reserved_62;
+       uint32_t reserved_63;
+       uint32_t reserved_64;
+       uint32_t reserved_65;
+       uint32_t reserved_66;
+       uint32_t reserved_67;
+       uint32_t reserved_68;
+       uint32_t reserved_69;
+       uint32_t reserved_70;
+       uint32_t reserved_71;
+       uint32_t reserved_72;
+       uint32_t reserved_73;
+       uint32_t reserved_74;
+       uint32_t reserved_75;
+       uint32_t reserved_76;
+       uint32_t reserved_77;
+       uint32_t reserved_78;
+       uint32_t reserved_79;
+       uint32_t reserved_80;
+       uint32_t reserved_81;
+       uint32_t reserved_82;
+       uint32_t reserved_83;
+       uint32_t reserved_84;
+       uint32_t reserved_85;
+       uint32_t reserved_86;
+       uint32_t reserved_87;
+       uint32_t reserved_88;
+       uint32_t reserved_89;
+       uint32_t reserved_90;
+       uint32_t reserved_91;
+       uint32_t reserved_92;
+       uint32_t reserved_93;
+       uint32_t reserved_94;
+       uint32_t reserved_95;
+       uint32_t reserved_96;
+       uint32_t reserved_97;
+       uint32_t reserved_98;
+       uint32_t reserved_99;
+       uint32_t reserved_100;
+       uint32_t reserved_101;
+       uint32_t reserved_102;
+       uint32_t reserved_103;
+       uint32_t reserved_104;
+       uint32_t reserved_105;
+       uint32_t reserved_106;
+       uint32_t reserved_107;
+       uint32_t reserved_108;
+       uint32_t reserved_109;
+       uint32_t reserved_110;
+       uint32_t reserved_111;
+       uint32_t reserved_112;
+       uint32_t reserved_113;
+       uint32_t reserved_114;
+       uint32_t reserved_115;
+       uint32_t reserved_116;
+       uint32_t reserved_117;
+       uint32_t reserved_118;
+       uint32_t reserved_119;
+       uint32_t reserved_120;
+       uint32_t reserved_121;
+       uint32_t reserved_122;
+       uint32_t reserved_123;
+       uint32_t reserved_124;
+       uint32_t reserved_125;
+       uint32_t reserved_126;
+       uint32_t reserved_127;
+};
+
+struct vi_mqd {
+       uint32_t header;
+       uint32_t compute_dispatch_initiator;
+       uint32_t compute_dim_x;
+       uint32_t compute_dim_y;
+       uint32_t compute_dim_z;
+       uint32_t compute_start_x;
+       uint32_t compute_start_y;
+       uint32_t compute_start_z;
+       uint32_t compute_num_thread_x;
+       uint32_t compute_num_thread_y;
+       uint32_t compute_num_thread_z;
+       uint32_t compute_pipelinestat_enable;
+       uint32_t compute_perfcount_enable;
+       uint32_t compute_pgm_lo;
+       uint32_t compute_pgm_hi;
+       uint32_t compute_tba_lo;
+       uint32_t compute_tba_hi;
+       uint32_t compute_tma_lo;
+       uint32_t compute_tma_hi;
+       uint32_t compute_pgm_rsrc1;
+       uint32_t compute_pgm_rsrc2;
+       uint32_t compute_vmid;
+       uint32_t compute_resource_limits;
+       uint32_t compute_static_thread_mgmt_se0;
+       uint32_t compute_static_thread_mgmt_se1;
+       uint32_t compute_tmpring_size;
+       uint32_t compute_static_thread_mgmt_se2;
+       uint32_t compute_static_thread_mgmt_se3;
+       uint32_t compute_restart_x;
+       uint32_t compute_restart_y;
+       uint32_t compute_restart_z;
+       uint32_t compute_thread_trace_enable;
+       uint32_t compute_misc_reserved;
+       uint32_t compute_dispatch_id;
+       uint32_t compute_threadgroup_id;
+       uint32_t compute_relaunch;
+       uint32_t compute_wave_restore_addr_lo;
+       uint32_t compute_wave_restore_addr_hi;
+       uint32_t compute_wave_restore_control;
+       uint32_t reserved_39;
+       uint32_t reserved_40;
+       uint32_t reserved_41;
+       uint32_t reserved_42;
+       uint32_t reserved_43;
+       uint32_t reserved_44;
+       uint32_t reserved_45;
+       uint32_t reserved_46;
+       uint32_t reserved_47;
+       uint32_t reserved_48;
+       uint32_t reserved_49;
+       uint32_t reserved_50;
+       uint32_t reserved_51;
+       uint32_t reserved_52;
+       uint32_t reserved_53;
+       uint32_t reserved_54;
+       uint32_t reserved_55;
+       uint32_t reserved_56;
+       uint32_t reserved_57;
+       uint32_t reserved_58;
+       uint32_t reserved_59;
+       uint32_t reserved_60;
+       uint32_t reserved_61;
+       uint32_t reserved_62;
+       uint32_t reserved_63;
+       uint32_t reserved_64;
+       uint32_t compute_user_data_0;
+       uint32_t compute_user_data_1;
+       uint32_t compute_user_data_2;
+       uint32_t compute_user_data_3;
+       uint32_t compute_user_data_4;
+       uint32_t compute_user_data_5;
+       uint32_t compute_user_data_6;
+       uint32_t compute_user_data_7;
+       uint32_t compute_user_data_8;
+       uint32_t compute_user_data_9;
+       uint32_t compute_user_data_10;
+       uint32_t compute_user_data_11;
+       uint32_t compute_user_data_12;
+       uint32_t compute_user_data_13;
+       uint32_t compute_user_data_14;
+       uint32_t compute_user_data_15;
+       uint32_t cp_compute_csinvoc_count_lo;
+       uint32_t cp_compute_csinvoc_count_hi;
+       uint32_t reserved_83;
+       uint32_t reserved_84;
+       uint32_t reserved_85;
+       uint32_t cp_mqd_query_time_lo;
+       uint32_t cp_mqd_query_time_hi;
+       uint32_t cp_mqd_connect_start_time_lo;
+       uint32_t cp_mqd_connect_start_time_hi;
+       uint32_t cp_mqd_connect_end_time_lo;
+       uint32_t cp_mqd_connect_end_time_hi;
+       uint32_t cp_mqd_connect_end_wf_count;
+       uint32_t cp_mqd_connect_end_pq_rptr;
+       uint32_t cp_mqd_connect_end_pq_wptr;
+       uint32_t cp_mqd_connect_end_ib_rptr;
+       uint32_t reserved_96;
+       uint32_t reserved_97;
+       uint32_t cp_mqd_save_start_time_lo;
+       uint32_t cp_mqd_save_start_time_hi;
+       uint32_t cp_mqd_save_end_time_lo;
+       uint32_t cp_mqd_save_end_time_hi;
+       uint32_t cp_mqd_restore_start_time_lo;
+       uint32_t cp_mqd_restore_start_time_hi;
+       uint32_t cp_mqd_restore_end_time_lo;
+       uint32_t cp_mqd_restore_end_time_hi;
+       uint32_t reserved_106;
+       uint32_t reserved_107;
+       uint32_t gds_cs_ctxsw_cnt0;
+       uint32_t gds_cs_ctxsw_cnt1;
+       uint32_t gds_cs_ctxsw_cnt2;
+       uint32_t gds_cs_ctxsw_cnt3;
+       uint32_t reserved_112;
+       uint32_t reserved_113;
+       uint32_t cp_pq_exe_status_lo;
+       uint32_t cp_pq_exe_status_hi;
+       uint32_t cp_packet_id_lo;
+       uint32_t cp_packet_id_hi;
+       uint32_t cp_packet_exe_status_lo;
+       uint32_t cp_packet_exe_status_hi;
+       uint32_t gds_save_base_addr_lo;
+       uint32_t gds_save_base_addr_hi;
+       uint32_t gds_save_mask_lo;
+       uint32_t gds_save_mask_hi;
+       uint32_t ctx_save_base_addr_lo;
+       uint32_t ctx_save_base_addr_hi;
+       uint32_t reserved_126;
+       uint32_t reserved_127;
+       uint32_t cp_mqd_base_addr_lo;
+       uint32_t cp_mqd_base_addr_hi;
+       uint32_t cp_hqd_active;
+       uint32_t cp_hqd_vmid;
+       uint32_t cp_hqd_persistent_state;
+       uint32_t cp_hqd_pipe_priority;
+       uint32_t cp_hqd_queue_priority;
+       uint32_t cp_hqd_quantum;
+       uint32_t cp_hqd_pq_base_lo;
+       uint32_t cp_hqd_pq_base_hi;
+       uint32_t cp_hqd_pq_rptr;
+       uint32_t cp_hqd_pq_rptr_report_addr_lo;
+       uint32_t cp_hqd_pq_rptr_report_addr_hi;
+       uint32_t cp_hqd_pq_wptr_poll_addr_lo;
+       uint32_t cp_hqd_pq_wptr_poll_addr_hi;
+       uint32_t cp_hqd_pq_doorbell_control;
+       uint32_t cp_hqd_pq_wptr;
+       uint32_t cp_hqd_pq_control;
+       uint32_t cp_hqd_ib_base_addr_lo;
+       uint32_t cp_hqd_ib_base_addr_hi;
+       uint32_t cp_hqd_ib_rptr;
+       uint32_t cp_hqd_ib_control;
+       uint32_t cp_hqd_iq_timer;
+       uint32_t cp_hqd_iq_rptr;
+       uint32_t cp_hqd_dequeue_request;
+       uint32_t cp_hqd_dma_offload;
+       uint32_t cp_hqd_sema_cmd;
+       uint32_t cp_hqd_msg_type;
+       uint32_t cp_hqd_atomic0_preop_lo;
+       uint32_t cp_hqd_atomic0_preop_hi;
+       uint32_t cp_hqd_atomic1_preop_lo;
+       uint32_t cp_hqd_atomic1_preop_hi;
+       uint32_t cp_hqd_hq_status0;
+       uint32_t cp_hqd_hq_control0;
+       uint32_t cp_mqd_control;
+       uint32_t cp_hqd_hq_status1;
+       uint32_t cp_hqd_hq_control1;
+       uint32_t cp_hqd_eop_base_addr_lo;
+       uint32_t cp_hqd_eop_base_addr_hi;
+       uint32_t cp_hqd_eop_control;
+       uint32_t cp_hqd_eop_rptr;
+       uint32_t cp_hqd_eop_wptr;
+       uint32_t cp_hqd_eop_done_events;
+       uint32_t cp_hqd_ctx_save_base_addr_lo;
+       uint32_t cp_hqd_ctx_save_base_addr_hi;
+       uint32_t cp_hqd_ctx_save_control;
+       uint32_t cp_hqd_cntl_stack_offset;
+       uint32_t cp_hqd_cntl_stack_size;
+       uint32_t cp_hqd_wg_state_offset;
+       uint32_t cp_hqd_ctx_save_size;
+       uint32_t cp_hqd_gds_resource_state;
+       uint32_t cp_hqd_error;
+       uint32_t cp_hqd_eop_wptr_mem;
+       uint32_t cp_hqd_eop_dones;
+       uint32_t reserved_182;
+       uint32_t reserved_183;
+       uint32_t reserved_184;
+       uint32_t reserved_185;
+       uint32_t reserved_186;
+       uint32_t reserved_187;
+       uint32_t reserved_188;
+       uint32_t reserved_189;
+       uint32_t reserved_190;
+       uint32_t reserved_191;
+       uint32_t iqtimer_pkt_header;
+       uint32_t iqtimer_pkt_dw0;
+       uint32_t iqtimer_pkt_dw1;
+       uint32_t iqtimer_pkt_dw2;
+       uint32_t iqtimer_pkt_dw3;
+       uint32_t iqtimer_pkt_dw4;
+       uint32_t iqtimer_pkt_dw5;
+       uint32_t iqtimer_pkt_dw6;
+       uint32_t iqtimer_pkt_dw7;
+       uint32_t iqtimer_pkt_dw8;
+       uint32_t iqtimer_pkt_dw9;
+       uint32_t iqtimer_pkt_dw10;
+       uint32_t iqtimer_pkt_dw11;
+       uint32_t iqtimer_pkt_dw12;
+       uint32_t iqtimer_pkt_dw13;
+       uint32_t iqtimer_pkt_dw14;
+       uint32_t iqtimer_pkt_dw15;
+       uint32_t iqtimer_pkt_dw16;
+       uint32_t iqtimer_pkt_dw17;
+       uint32_t iqtimer_pkt_dw18;
+       uint32_t iqtimer_pkt_dw19;
+       uint32_t iqtimer_pkt_dw20;
+       uint32_t iqtimer_pkt_dw21;
+       uint32_t iqtimer_pkt_dw22;
+       uint32_t iqtimer_pkt_dw23;
+       uint32_t iqtimer_pkt_dw24;
+       uint32_t iqtimer_pkt_dw25;
+       uint32_t iqtimer_pkt_dw26;
+       uint32_t iqtimer_pkt_dw27;
+       uint32_t iqtimer_pkt_dw28;
+       uint32_t iqtimer_pkt_dw29;
+       uint32_t iqtimer_pkt_dw30;
+       uint32_t iqtimer_pkt_dw31;
+       uint32_t reserved_225;
+       uint32_t reserved_226;
+       uint32_t reserved_227;
+       uint32_t set_resources_header;
+       uint32_t set_resources_dw1;
+       uint32_t set_resources_dw2;
+       uint32_t set_resources_dw3;
+       uint32_t set_resources_dw4;
+       uint32_t set_resources_dw5;
+       uint32_t set_resources_dw6;
+       uint32_t set_resources_dw7;
+       uint32_t reserved_236;
+       uint32_t reserved_237;
+       uint32_t reserved_238;
+       uint32_t reserved_239;
+       uint32_t queue_doorbell_id0;
+       uint32_t queue_doorbell_id1;
+       uint32_t queue_doorbell_id2;
+       uint32_t queue_doorbell_id3;
+       uint32_t queue_doorbell_id4;
+       uint32_t queue_doorbell_id5;
+       uint32_t queue_doorbell_id6;
+       uint32_t queue_doorbell_id7;
+       uint32_t queue_doorbell_id8;
+       uint32_t queue_doorbell_id9;
+       uint32_t queue_doorbell_id10;
+       uint32_t queue_doorbell_id11;
+       uint32_t queue_doorbell_id12;
+       uint32_t queue_doorbell_id13;
+       uint32_t queue_doorbell_id14;
+       uint32_t queue_doorbell_id15;
+};
+
+#endif /* VI_STRUCTS_H_ */