[PATCH 128/459] drm/amdkfd: Add navi10 support to amdkfd. (v2)
Alex Deucher
alexdeucher at gmail.com
Mon Jun 17 19:25:53 UTC 2019
From: Philip Cox <Philip.Cox at amd.com>
KFD (kernel fusion driver) is the kernel driver
for the compute backend for usermode compute
stack.
v2: squash in updates (Alex)
Signed-off-by: Oak Zeng <Oak.Zeng at amd.com>
Signed-off-by: Philip Cox <Philip.Cox at amd.com>
Acked-by: Alex Deucher <alexander.deucher at amd.com>
Signed-off-by: Hawking Zhang <Hawking.Zhang at amd.com>
Signed-off-by: Alex Deucher <alexander.deucher at amd.com>
---
drivers/gpu/drm/amd/amdkfd/Makefile | 3 +
drivers/gpu/drm/amd/amdkfd/kfd_crat.c | 5 +
drivers/gpu/drm/amd/amdkfd/kfd_device.c | 21 +-
.../drm/amd/amdkfd/kfd_device_queue_manager.c | 26 +-
.../drm/amd/amdkfd/kfd_device_queue_manager.h | 2 +
.../amd/amdkfd/kfd_device_queue_manager_v10.c | 87 +++
drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c | 1 +
drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c | 3 +
drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h | 1 +
.../gpu/drm/amd/amdkfd/kfd_kernel_queue_v10.c | 348 ++++++++++++
.../gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c | 519 ++++++++++++++++++
.../gpu/drm/amd/amdkfd/kfd_packet_manager.c | 3 +
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 11 +-
drivers/gpu/drm/amd/amdkfd/kfd_process.c | 1 +
drivers/gpu/drm/amd/amdkfd/kfd_topology.c | 1 +
15 files changed, 1019 insertions(+), 13 deletions(-)
create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c
create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_v10.c
create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c
diff --git a/drivers/gpu/drm/amd/amdkfd/Makefile b/drivers/gpu/drm/amd/amdkfd/Makefile
index 69ec96998bb9..48155060a57c 100644
--- a/drivers/gpu/drm/amd/amdkfd/Makefile
+++ b/drivers/gpu/drm/amd/amdkfd/Makefile
@@ -36,16 +36,19 @@ AMDKFD_FILES := $(AMDKFD_PATH)/kfd_module.o \
$(AMDKFD_PATH)/kfd_mqd_manager_cik.o \
$(AMDKFD_PATH)/kfd_mqd_manager_vi.o \
$(AMDKFD_PATH)/kfd_mqd_manager_v9.o \
+ $(AMDKFD_PATH)/kfd_mqd_manager_v10.o \
$(AMDKFD_PATH)/kfd_kernel_queue.o \
$(AMDKFD_PATH)/kfd_kernel_queue_cik.o \
$(AMDKFD_PATH)/kfd_kernel_queue_vi.o \
$(AMDKFD_PATH)/kfd_kernel_queue_v9.o \
+ $(AMDKFD_PATH)/kfd_kernel_queue_v10.o \
$(AMDKFD_PATH)/kfd_packet_manager.o \
$(AMDKFD_PATH)/kfd_process_queue_manager.o \
$(AMDKFD_PATH)/kfd_device_queue_manager.o \
$(AMDKFD_PATH)/kfd_device_queue_manager_cik.o \
$(AMDKFD_PATH)/kfd_device_queue_manager_vi.o \
$(AMDKFD_PATH)/kfd_device_queue_manager_v9.o \
+ $(AMDKFD_PATH)/kfd_device_queue_manager_v10.o \
$(AMDKFD_PATH)/kfd_interrupt.o \
$(AMDKFD_PATH)/kfd_events.o \
$(AMDKFD_PATH)/cik_event_interrupt.o \
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
index 59f8ca4297db..792371442195 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_crat.c
@@ -138,6 +138,8 @@ static struct kfd_gpu_cache_info carrizo_cache_info[] = {
/* TODO - check & update Vega10 cache details */
#define vega10_cache_info carrizo_cache_info
#define raven_cache_info carrizo_cache_info
+/* TODO - check & update Navi10 cache details */
+#define navi10_cache_info carrizo_cache_info
static void kfd_populated_cu_info_cpu(struct kfd_topology_device *dev,
struct crat_subtype_computeunit *cu)
@@ -666,6 +668,9 @@ static int kfd_fill_gpu_cache_info(struct kfd_dev *kdev,
case CHIP_RAVEN:
pcache_info = raven_cache_info;
num_of_cache_types = ARRAY_SIZE(raven_cache_info);
+ case CHIP_NAVI10:
+ pcache_info = navi10_cache_info;
+ num_of_cache_types = ARRAY_SIZE(navi10_cache_info);
break;
default:
return -EINVAL;
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device.c b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
index ebac7d7f9956..955d72179da1 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device.c
@@ -317,6 +317,22 @@ static const struct kfd_device_info vega20_device_info = {
.num_sdma_queues_per_engine = 8,
};
+static const struct kfd_device_info navi10_device_info = {
+ .asic_family = CHIP_NAVI10,
+ .max_pasid_bits = 16,
+ .max_no_of_hqd = 24,
+ .doorbell_size = 8,
+ .ih_ring_entry_size = 8 * sizeof(uint32_t),
+ .event_interrupt_class = &event_interrupt_class_v9,
+ .num_of_watch_points = 4,
+ .mqd_size_aligned = MQD_SIZE_ALIGNED,
+ .needs_iommu_device = false,
+ .supports_cwsr = false,
+ .needs_pci_atomics = false,
+ .num_sdma_engines = 2,
+ .num_sdma_queues_per_engine = 8,
+};
+
struct kfd_deviceid {
unsigned short did;
const struct kfd_device_info *device_info;
@@ -434,7 +450,9 @@ static const struct kfd_deviceid supported_devices[] = {
{ 0x66a3, &vega20_device_info }, /* Vega20 */
{ 0x66a4, &vega20_device_info }, /* Vega20 */
{ 0x66a7, &vega20_device_info }, /* Vega20 */
- { 0x66af, &vega20_device_info } /* Vega20 */
+ { 0x66af, &vega20_device_info }, /* Vega20 */
+ /* Navi10 */
+ { 0x7310, &navi10_device_info }, /* Navi10 */
};
static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
@@ -517,6 +535,7 @@ static void kfd_cwsr_init(struct kfd_dev *kfd)
kfd->cwsr_isa = cwsr_trap_gfx8_hex;
kfd->cwsr_isa_size = sizeof(cwsr_trap_gfx8_hex);
} else {
+ /* TODO: Do we need another trap handler for navi10? */
BUILD_BUG_ON(sizeof(cwsr_trap_gfx9_hex) > PAGE_SIZE);
kfd->cwsr_isa = cwsr_trap_gfx9_hex;
kfd->cwsr_isa_size = sizeof(cwsr_trap_gfx9_hex);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
index 3528590ae90b..632e510b5396 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
@@ -1264,6 +1264,7 @@ static int map_queues_cpsch(struct device_queue_manager *dqm)
return 0;
retval = pm_send_runlist(&dqm->packets, &dqm->queues);
+ pr_debug("%s sent runlist\n", __func__);
if (retval) {
pr_err("failed to execute runlist\n");
return retval;
@@ -1785,6 +1786,9 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)
case CHIP_RAVEN:
device_queue_manager_init_v9(&dqm->asic_ops);
break;
+ case CHIP_NAVI10:
+ device_queue_manager_init_v10_navi10(&dqm->asic_ops);
+ break;
default:
WARN(1, "Unexpected ASIC family %u",
dev->device_info->asic_family);
@@ -1875,17 +1879,17 @@ int dqm_debugfs_hqds(struct seq_file *m, void *data)
int pipe, queue;
int r = 0;
- r = dqm->dev->kfd2kgd->hqd_dump(dqm->dev->kgd,
- KFD_CIK_HIQ_PIPE, KFD_CIK_HIQ_QUEUE, &dump, &n_regs);
- if (!r) {
- seq_printf(m, " HIQ on MEC %d Pipe %d Queue %d\n",
- KFD_CIK_HIQ_PIPE/get_pipes_per_mec(dqm)+1,
- KFD_CIK_HIQ_PIPE%get_pipes_per_mec(dqm),
- KFD_CIK_HIQ_QUEUE);
- seq_reg_dump(m, dump, n_regs);
-
- kfree(dump);
- }
+ r = dqm->dev->kfd2kgd->hqd_dump(dqm->dev->kgd,
+ KFD_CIK_HIQ_PIPE, KFD_CIK_HIQ_QUEUE, &dump, &n_regs);
+ if (!r) {
+ seq_printf(m, " HIQ on MEC %d Pipe %d Queue %d\n",
+ KFD_CIK_HIQ_PIPE/get_pipes_per_mec(dqm)+1,
+ KFD_CIK_HIQ_PIPE%get_pipes_per_mec(dqm),
+ KFD_CIK_HIQ_QUEUE);
+ seq_reg_dump(m, dump, n_regs);
+
+ kfree(dump);
+ }
for (pipe = 0; pipe < get_pipes_per_mec(dqm); pipe++) {
int pipe_offset = pipe * get_queues_per_pipe(dqm);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
index 88b4c007696e..ff9cdc584120 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h
@@ -212,6 +212,8 @@ void device_queue_manager_init_vi_tonga(
struct device_queue_manager_asic_ops *asic_ops);
void device_queue_manager_init_v9(
struct device_queue_manager_asic_ops *asic_ops);
+void device_queue_manager_init_v10_navi10(
+ struct device_queue_manager_asic_ops *asic_ops);
void program_sh_mem_settings(struct device_queue_manager *dqm,
struct qcm_process_device *qpd);
unsigned int get_queues_num(struct device_queue_manager *dqm);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c
new file mode 100644
index 000000000000..adb38850366c
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_v10.c
@@ -0,0 +1,87 @@
+/*
+ * Copyright 2018 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 "kfd_device_queue_manager.h"
+#include "navi10_enum.h"
+#include "gc/gc_10_1_0_offset.h"
+#include "gc/gc_10_1_0_sh_mask.h"
+
+static int update_qpd_v10(struct device_queue_manager *dqm,
+ struct qcm_process_device *qpd);
+static void init_sdma_vm_v10(struct device_queue_manager *dqm, struct queue *q,
+ struct qcm_process_device *qpd);
+
+void device_queue_manager_init_v10_navi10(
+ struct device_queue_manager_asic_ops *asic_ops)
+{
+ asic_ops->update_qpd = update_qpd_v10;
+ asic_ops->init_sdma_vm = init_sdma_vm_v10;
+}
+
+static uint32_t compute_sh_mem_bases_64bit(struct kfd_process_device *pdd)
+{
+ uint32_t shared_base = pdd->lds_base >> 48;
+ uint32_t private_base = pdd->scratch_base >> 48;
+
+ return (shared_base << SH_MEM_BASES__SHARED_BASE__SHIFT) |
+ private_base;
+}
+
+static int update_qpd_v10(struct device_queue_manager *dqm,
+ struct qcm_process_device *qpd)
+{
+ struct kfd_process_device *pdd;
+
+ pdd = qpd_to_pdd(qpd);
+
+ /* check if sh_mem_config register already configured */
+ if (qpd->sh_mem_config == 0) {
+ qpd->sh_mem_config =
+ SH_MEM_ALIGNMENT_MODE_UNALIGNED <<
+ SH_MEM_CONFIG__ALIGNMENT_MODE__SHIFT;
+#if 0
+ /* TODO:
+ * This shouldn't be an issue with Navi10. Verify.
+ */
+ if (vega10_noretry)
+ qpd->sh_mem_config |=
+ 1 << SH_MEM_CONFIG__RETRY_DISABLE__SHIFT;
+#endif
+
+ qpd->sh_mem_ape1_limit = 0;
+ qpd->sh_mem_ape1_base = 0;
+ }
+
+ qpd->sh_mem_bases = compute_sh_mem_bases_64bit(pdd);
+
+ pr_debug("sh_mem_bases 0x%X\n", qpd->sh_mem_bases);
+
+ return 0;
+}
+
+static void init_sdma_vm_v10(struct device_queue_manager *dqm, struct queue *q,
+ struct qcm_process_device *qpd)
+{
+ /* Not needed on SDMAv4 onwards any more */
+ q->properties.sdma_vm_addr = 0;
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
index 22a8e88b6a67..60521366dd31 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
@@ -405,6 +405,7 @@ int kfd_init_apertures(struct kfd_process *process)
case CHIP_VEGA12:
case CHIP_VEGA20:
case CHIP_RAVEN:
+ case CHIP_NAVI10:
kfd_init_apertures_v9(pdd, id);
break;
default:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
index 229500c8c958..29c0bd2d7a5c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.c
@@ -332,6 +332,9 @@ struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
case CHIP_RAVEN:
kernel_queue_init_v9(&kq->ops_asic_specific);
break;
+ case CHIP_NAVI10:
+ kernel_queue_init_v10(&kq->ops_asic_specific);
+ break;
default:
WARN(1, "Unexpected ASIC family %u",
dev->device_info->asic_family);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h
index a7116a939029..365fc674fea4 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue.h
@@ -102,5 +102,6 @@ struct kernel_queue {
void kernel_queue_init_cik(struct kernel_queue_ops *ops);
void kernel_queue_init_vi(struct kernel_queue_ops *ops);
void kernel_queue_init_v9(struct kernel_queue_ops *ops);
+void kernel_queue_init_v10(struct kernel_queue_ops *ops);
#endif /* KFD_KERNEL_QUEUE_H_ */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_v10.c
new file mode 100644
index 000000000000..209ad518fba1
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_kernel_queue_v10.c
@@ -0,0 +1,348 @@
+/*
+ * Copyright 2018 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 "kfd_kernel_queue.h"
+#include "kfd_device_queue_manager.h"
+#include "kfd_pm4_headers_ai.h"
+#include "kfd_pm4_opcodes.h"
+#include "gc/gc_10_1_0_sh_mask.h"
+
+static bool initialize_v10(struct kernel_queue *kq, struct kfd_dev *dev,
+ enum kfd_queue_type type, unsigned int queue_size);
+static void uninitialize_v10(struct kernel_queue *kq);
+static void submit_packet_v10(struct kernel_queue *kq);
+
+void kernel_queue_init_v10(struct kernel_queue_ops *ops)
+{
+ ops->initialize = initialize_v10;
+ ops->uninitialize = uninitialize_v10;
+ ops->submit_packet = submit_packet_v10;
+}
+
+static bool initialize_v10(struct kernel_queue *kq, struct kfd_dev *dev,
+ enum kfd_queue_type type, unsigned int queue_size)
+{
+ int retval;
+
+ retval = kfd_gtt_sa_allocate(dev, PAGE_SIZE, &kq->eop_mem);
+ if (retval != 0)
+ return false;
+
+ kq->eop_gpu_addr = kq->eop_mem->gpu_addr;
+ kq->eop_kernel_addr = kq->eop_mem->cpu_ptr;
+
+ memset(kq->eop_kernel_addr, 0, PAGE_SIZE);
+
+ return true;
+}
+
+static void uninitialize_v10(struct kernel_queue *kq)
+{
+ kfd_gtt_sa_free(kq->dev, kq->eop_mem);
+}
+
+static void submit_packet_v10(struct kernel_queue *kq)
+{
+ *kq->wptr64_kernel = kq->pending_wptr64;
+ write_kernel_doorbell64(kq->queue->properties.doorbell_ptr,
+ kq->pending_wptr64);
+}
+
+static int pm_map_process_v10(struct packet_manager *pm,
+ uint32_t *buffer, struct qcm_process_device *qpd)
+{
+ struct pm4_mes_map_process *packet;
+ uint64_t vm_page_table_base_addr = qpd->page_table_base;
+
+ packet = (struct pm4_mes_map_process *)buffer;
+ memset(buffer, 0, sizeof(struct pm4_mes_map_process));
+
+ packet->header.u32All = pm_build_pm4_header(IT_MAP_PROCESS,
+ sizeof(struct pm4_mes_map_process));
+ packet->bitfields2.diq_enable = (qpd->is_debug) ? 1 : 0;
+ packet->bitfields2.process_quantum = 1;
+ packet->bitfields2.pasid = qpd->pqm->process->pasid;
+ packet->bitfields14.gds_size = qpd->gds_size;
+ packet->bitfields14.num_gws = qpd->num_gws;
+ packet->bitfields14.num_oac = qpd->num_oac;
+ packet->bitfields14.sdma_enable = 1;
+
+ packet->bitfields14.num_queues = (qpd->is_debug) ? 0 : qpd->queue_count;
+
+ packet->sh_mem_config = qpd->sh_mem_config;
+ packet->sh_mem_bases = qpd->sh_mem_bases;
+ if (qpd->tba_addr) {
+ packet->sq_shader_tba_lo = lower_32_bits(qpd->tba_addr >> 8);
+ packet->sq_shader_tba_hi = (1 << SQ_SHADER_TBA_HI__TRAP_EN__SHIFT) |
+ upper_32_bits(qpd->tba_addr >> 8);
+ packet->sq_shader_tma_lo = lower_32_bits(qpd->tma_addr >> 8);
+ packet->sq_shader_tma_hi = upper_32_bits(qpd->tma_addr >> 8);
+ }
+
+ packet->gds_addr_lo = lower_32_bits(qpd->gds_context_area);
+ packet->gds_addr_hi = upper_32_bits(qpd->gds_context_area);
+
+ packet->vm_context_page_table_base_addr_lo32 =
+ lower_32_bits(vm_page_table_base_addr);
+ packet->vm_context_page_table_base_addr_hi32 =
+ upper_32_bits(vm_page_table_base_addr);
+
+ return 0;
+}
+
+static int pm_runlist_v10(struct packet_manager *pm, uint32_t *buffer,
+ uint64_t ib, size_t ib_size_in_dwords, bool chain)
+{
+ struct pm4_mes_runlist *packet;
+
+ int concurrent_proc_cnt = 0;
+ struct kfd_dev *kfd = pm->dqm->dev;
+
+ /* Determine the number of processes to map together to HW:
+ * it can not exceed the number of VMIDs available to the
+ * scheduler, and it is determined by the smaller of the number
+ * of processes in the runlist and kfd module parameter
+ * hws_max_conc_proc.
+ * Note: the arbitration between the number of VMIDs and
+ * hws_max_conc_proc has been done in
+ * kgd2kfd_device_init().
+ */
+ concurrent_proc_cnt = min(pm->dqm->processes_count,
+ kfd->max_proc_per_quantum);
+
+
+ packet = (struct pm4_mes_runlist *)buffer;
+
+ memset(buffer, 0, sizeof(struct pm4_mes_runlist));
+ packet->header.u32All = pm_build_pm4_header(IT_RUN_LIST,
+ sizeof(struct pm4_mes_runlist));
+
+ packet->bitfields4.ib_size = ib_size_in_dwords;
+ packet->bitfields4.chain = chain ? 1 : 0;
+ packet->bitfields4.offload_polling = 0;
+ packet->bitfields4.valid = 1;
+ packet->bitfields4.process_cnt = concurrent_proc_cnt;
+ packet->ordinal2 = lower_32_bits(ib);
+ packet->ib_base_hi = upper_32_bits(ib);
+
+ return 0;
+}
+
+static int pm_map_queues_v10(struct packet_manager *pm, uint32_t *buffer,
+ struct queue *q, bool is_static)
+{
+ struct pm4_mes_map_queues *packet;
+ bool use_static = is_static;
+
+ packet = (struct pm4_mes_map_queues *)buffer;
+ memset(buffer, 0, sizeof(struct pm4_mes_map_queues));
+
+ packet->header.u32All = pm_build_pm4_header(IT_MAP_QUEUES,
+ sizeof(struct pm4_mes_map_queues));
+ packet->bitfields2.alloc_format =
+ alloc_format__mes_map_queues__one_per_pipe_vi;
+ packet->bitfields2.num_queues = 1;
+ packet->bitfields2.queue_sel =
+ queue_sel__mes_map_queues__map_to_hws_determined_queue_slots_vi;
+
+ packet->bitfields2.engine_sel =
+ engine_sel__mes_map_queues__compute_vi;
+ packet->bitfields2.queue_type =
+ queue_type__mes_map_queues__normal_compute_vi;
+
+ switch (q->properties.type) {
+ case KFD_QUEUE_TYPE_COMPUTE:
+ if (use_static)
+ packet->bitfields2.queue_type =
+ queue_type__mes_map_queues__normal_latency_static_queue_vi;
+ break;
+ case KFD_QUEUE_TYPE_DIQ:
+ packet->bitfields2.queue_type =
+ queue_type__mes_map_queues__debug_interface_queue_vi;
+ break;
+ case KFD_QUEUE_TYPE_SDMA:
+ packet->bitfields2.engine_sel = q->properties.sdma_engine_id +
+ engine_sel__mes_map_queues__sdma0_vi;
+ use_static = false; /* no static queues under SDMA */
+ break;
+ default:
+ WARN(1, "queue type %d\n", q->properties.type);
+ return -EINVAL;
+ }
+ packet->bitfields3.doorbell_offset =
+ q->properties.doorbell_off;
+
+ packet->mqd_addr_lo =
+ lower_32_bits(q->gart_mqd_addr);
+
+ packet->mqd_addr_hi =
+ upper_32_bits(q->gart_mqd_addr);
+
+ packet->wptr_addr_lo =
+ lower_32_bits((uint64_t)q->properties.write_ptr);
+
+ packet->wptr_addr_hi =
+ upper_32_bits((uint64_t)q->properties.write_ptr);
+
+ return 0;
+}
+
+static int pm_unmap_queues_v10(struct packet_manager *pm, uint32_t *buffer,
+ enum kfd_queue_type type,
+ enum kfd_unmap_queues_filter filter,
+ uint32_t filter_param, bool reset,
+ unsigned int sdma_engine)
+{
+ struct pm4_mes_unmap_queues *packet;
+
+ packet = (struct pm4_mes_unmap_queues *)buffer;
+ memset(buffer, 0, sizeof(struct pm4_mes_unmap_queues));
+
+ packet->header.u32All = pm_build_pm4_header(IT_UNMAP_QUEUES,
+ sizeof(struct pm4_mes_unmap_queues));
+ switch (type) {
+ case KFD_QUEUE_TYPE_COMPUTE:
+ case KFD_QUEUE_TYPE_DIQ:
+ packet->bitfields2.engine_sel =
+ engine_sel__mes_unmap_queues__compute;
+ break;
+ case KFD_QUEUE_TYPE_SDMA:
+ packet->bitfields2.engine_sel =
+ engine_sel__mes_unmap_queues__sdma0 + sdma_engine;
+ break;
+ default:
+ WARN(1, "queue type %d\n", type);
+ break;
+ }
+
+ if (reset)
+ packet->bitfields2.action =
+ action__mes_unmap_queues__reset_queues;
+ else
+ packet->bitfields2.action =
+ action__mes_unmap_queues__preempt_queues;
+
+ switch (filter) {
+ case KFD_UNMAP_QUEUES_FILTER_SINGLE_QUEUE:
+ packet->bitfields2.queue_sel =
+ queue_sel__mes_unmap_queues__perform_request_on_specified_queues;
+ packet->bitfields2.num_queues = 1;
+ packet->bitfields3b.doorbell_offset0 = filter_param;
+ break;
+ case KFD_UNMAP_QUEUES_FILTER_BY_PASID:
+ packet->bitfields2.queue_sel =
+ queue_sel__mes_unmap_queues__perform_request_on_pasid_queues;
+ packet->bitfields3a.pasid = filter_param;
+ break;
+ case KFD_UNMAP_QUEUES_FILTER_ALL_QUEUES:
+ packet->bitfields2.queue_sel =
+ queue_sel__mes_unmap_queues__unmap_all_queues;
+ break;
+ case KFD_UNMAP_QUEUES_FILTER_DYNAMIC_QUEUES:
+ /* in this case, we do not preempt static queues */
+ packet->bitfields2.queue_sel =
+ queue_sel__mes_unmap_queues__unmap_all_non_static_queues;
+ break;
+ default:
+ WARN(1, "filter %d\n", filter);
+ break;
+ }
+
+ return 0;
+
+}
+
+static int pm_query_status_v10(struct packet_manager *pm, uint32_t *buffer,
+ uint64_t fence_address, uint32_t fence_value)
+{
+ struct pm4_mes_query_status *packet;
+
+ packet = (struct pm4_mes_query_status *)buffer;
+ memset(buffer, 0, sizeof(struct pm4_mes_query_status));
+
+
+ packet->header.u32All = pm_build_pm4_header(IT_QUERY_STATUS,
+ sizeof(struct pm4_mes_query_status));
+
+ packet->bitfields2.context_id = 0;
+ packet->bitfields2.interrupt_sel =
+ interrupt_sel__mes_query_status__completion_status;
+ packet->bitfields2.command =
+ command__mes_query_status__fence_only_after_write_ack;
+
+ packet->addr_hi = upper_32_bits((uint64_t)fence_address);
+ packet->addr_lo = lower_32_bits((uint64_t)fence_address);
+ packet->data_hi = upper_32_bits((uint64_t)fence_value);
+ packet->data_lo = lower_32_bits((uint64_t)fence_value);
+
+ return 0;
+}
+
+
+static int pm_release_mem_v10(uint64_t gpu_addr, uint32_t *buffer)
+{
+ struct pm4_mec_release_mem *packet;
+
+ WARN_ON(!buffer);
+
+ packet = (struct pm4_mec_release_mem *)buffer;
+ memset(buffer, 0, sizeof(struct pm4_mec_release_mem));
+
+ packet->header.u32All = pm_build_pm4_header(IT_RELEASE_MEM,
+ sizeof(struct pm4_mec_release_mem));
+
+ packet->bitfields2.event_type = CACHE_FLUSH_AND_INV_TS_EVENT;
+ packet->bitfields2.event_index = event_index__mec_release_mem__end_of_pipe;
+ packet->bitfields2.tcl1_action_ena = 1;
+ packet->bitfields2.tc_action_ena = 1;
+ packet->bitfields2.cache_policy = cache_policy__mec_release_mem__lru;
+
+ packet->bitfields3.data_sel = data_sel__mec_release_mem__send_32_bit_low;
+ packet->bitfields3.int_sel =
+ int_sel__mec_release_mem__send_interrupt_after_write_confirm;
+
+ packet->bitfields4.address_lo_32b = (gpu_addr & 0xffffffff) >> 2;
+ packet->address_hi = upper_32_bits(gpu_addr);
+
+ packet->data_lo = 0;
+
+ return sizeof(struct pm4_mec_release_mem) / sizeof(unsigned int);
+}
+
+const struct packet_manager_funcs kfd_v10_pm_funcs = {
+ .map_process = pm_map_process_v10,
+ .runlist = pm_runlist_v10,
+ .set_resources = pm_set_resources_vi,
+ .map_queues = pm_map_queues_v10,
+ .unmap_queues = pm_unmap_queues_v10,
+ .query_status = pm_query_status_v10,
+ .release_mem = pm_release_mem_v10,
+ .map_process_size = sizeof(struct pm4_mes_map_process),
+ .runlist_size = sizeof(struct pm4_mes_runlist),
+ .set_resources_size = sizeof(struct pm4_mes_set_resources),
+ .map_queues_size = sizeof(struct pm4_mes_map_queues),
+ .unmap_queues_size = sizeof(struct pm4_mes_unmap_queues),
+ .query_status_size = sizeof(struct pm4_mes_query_status),
+ .release_mem_size = sizeof(struct pm4_mec_release_mem)
+};
+
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c
new file mode 100644
index 000000000000..6663b72370f6
--- /dev/null
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_mqd_manager_v10.c
@@ -0,0 +1,519 @@
+/*
+ * Copyright 2018 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/printk.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include "kfd_priv.h"
+#include "kfd_mqd_manager.h"
+#include "v10_structs.h"
+#include "gc/gc_10_1_0_offset.h"
+#include "gc/gc_10_1_0_sh_mask.h"
+#include "amdgpu_amdkfd.h"
+
+static inline struct v10_compute_mqd *get_mqd(void *mqd)
+{
+ return (struct v10_compute_mqd *)mqd;
+}
+
+static inline struct v10_sdma_mqd *get_sdma_mqd(void *mqd)
+{
+ return (struct v10_sdma_mqd *)mqd;
+}
+
+static void update_cu_mask(struct mqd_manager *mm, void *mqd,
+ struct queue_properties *q)
+{
+ struct v10_compute_mqd *m;
+ uint32_t se_mask[4] = {0}; /* 4 is the max # of SEs */
+
+ if (q->cu_mask_count == 0)
+ return;
+
+ mqd_symmetrically_map_cu_mask(mm,
+ q->cu_mask, q->cu_mask_count, se_mask);
+
+ m = get_mqd(mqd);
+ m->compute_static_thread_mgmt_se0 = se_mask[0];
+ m->compute_static_thread_mgmt_se1 = se_mask[1];
+ m->compute_static_thread_mgmt_se2 = se_mask[2];
+ m->compute_static_thread_mgmt_se3 = se_mask[3];
+
+ pr_debug("update cu mask to %#x %#x %#x %#x\n",
+ m->compute_static_thread_mgmt_se0,
+ m->compute_static_thread_mgmt_se1,
+ m->compute_static_thread_mgmt_se2,
+ m->compute_static_thread_mgmt_se3);
+}
+
+static struct kfd_mem_obj *allocate_mqd(struct kfd_dev *kfd,
+ struct queue_properties *q)
+{
+ int retval;
+ struct kfd_mem_obj *mqd_mem_obj = NULL;
+
+ /* From V9, for CWSR, the control stack is located on the next page
+ * boundary after the mqd, we will use the gtt allocation function
+ * instead of sub-allocation function.
+ */
+ if (kfd->cwsr_enabled && (q->type == KFD_QUEUE_TYPE_COMPUTE)) {
+ mqd_mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_NOIO);
+ if (!mqd_mem_obj)
+ return NULL;
+ retval = amdgpu_amdkfd_alloc_gtt_mem(kfd->kgd,
+ ALIGN(q->ctl_stack_size, PAGE_SIZE) +
+ ALIGN(sizeof(struct v10_compute_mqd), PAGE_SIZE),
+ &(mqd_mem_obj->gtt_mem),
+ &(mqd_mem_obj->gpu_addr),
+ (void *)&(mqd_mem_obj->cpu_ptr), true);
+ } else {
+ retval = kfd_gtt_sa_allocate(kfd, sizeof(struct v10_compute_mqd),
+ &mqd_mem_obj);
+ }
+
+ if (retval) {
+ kfree(mqd_mem_obj);
+ return NULL;
+ }
+
+ return mqd_mem_obj;
+
+}
+
+static int init_mqd(struct mqd_manager *mm, void **mqd,
+ struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
+ struct queue_properties *q)
+{
+ int retval;
+ uint64_t addr;
+ struct v10_compute_mqd *m;
+ struct kfd_dev *kfd = mm->dev;
+
+ *mqd_mem_obj = allocate_mqd(kfd, q);
+ if (!*mqd_mem_obj)
+ return -ENOMEM;
+
+ m = (struct v10_compute_mqd *) (*mqd_mem_obj)->cpu_ptr;
+ addr = (*mqd_mem_obj)->gpu_addr;
+
+ memset(m, 0, sizeof(struct v10_compute_mqd));
+
+ m->header = 0xC0310800;
+ m->compute_pipelinestat_enable = 1;
+ m->compute_static_thread_mgmt_se0 = 0xFFFFFFFF;
+ m->compute_static_thread_mgmt_se1 = 0xFFFFFFFF;
+ m->compute_static_thread_mgmt_se2 = 0xFFFFFFFF;
+ m->compute_static_thread_mgmt_se3 = 0xFFFFFFFF;
+
+ m->cp_hqd_persistent_state = CP_HQD_PERSISTENT_STATE__PRELOAD_REQ_MASK |
+ 0x53 << CP_HQD_PERSISTENT_STATE__PRELOAD_SIZE__SHIFT;
+
+ m->cp_mqd_control = 1 << CP_MQD_CONTROL__PRIV_STATE__SHIFT;
+
+ m->cp_mqd_base_addr_lo = lower_32_bits(addr);
+ m->cp_mqd_base_addr_hi = upper_32_bits(addr);
+
+ m->cp_hqd_quantum = 1 << CP_HQD_QUANTUM__QUANTUM_EN__SHIFT |
+ 1 << CP_HQD_QUANTUM__QUANTUM_SCALE__SHIFT |
+ 10 << CP_HQD_QUANTUM__QUANTUM_DURATION__SHIFT;
+
+ m->cp_hqd_pipe_priority = 1;
+ m->cp_hqd_queue_priority = 15;
+
+ if (q->format == KFD_QUEUE_FORMAT_AQL) {
+ m->cp_hqd_aql_control =
+ 1 << CP_HQD_AQL_CONTROL__CONTROL0__SHIFT;
+ }
+
+ if (mm->dev->cwsr_enabled) {
+ m->cp_hqd_persistent_state |=
+ (1 << CP_HQD_PERSISTENT_STATE__QSWITCH_MODE__SHIFT);
+ m->cp_hqd_ctx_save_base_addr_lo =
+ lower_32_bits(q->ctx_save_restore_area_address);
+ m->cp_hqd_ctx_save_base_addr_hi =
+ upper_32_bits(q->ctx_save_restore_area_address);
+ m->cp_hqd_ctx_save_size = q->ctx_save_restore_area_size;
+ m->cp_hqd_cntl_stack_size = q->ctl_stack_size;
+ m->cp_hqd_cntl_stack_offset = q->ctl_stack_size;
+ m->cp_hqd_wg_state_offset = q->ctl_stack_size;
+ }
+
+ *mqd = m;
+ if (gart_addr)
+ *gart_addr = addr;
+ retval = mm->update_mqd(mm, m, q);
+
+ return retval;
+}
+
+static int load_mqd(struct mqd_manager *mm, void *mqd,
+ uint32_t pipe_id, uint32_t queue_id,
+ struct queue_properties *p, struct mm_struct *mms)
+{
+ int r = 0;
+ /* AQL write pointer counts in 64B packets, PM4/CP counts in dwords. */
+ uint32_t wptr_shift = (p->format == KFD_QUEUE_FORMAT_AQL ? 4 : 0);
+
+ r = mm->dev->kfd2kgd->hqd_load(mm->dev->kgd, mqd, pipe_id, queue_id,
+ (uint32_t __user *)p->write_ptr,
+ wptr_shift, 0, mms);
+ return r;
+}
+
+static int update_mqd(struct mqd_manager *mm, void *mqd,
+ struct queue_properties *q)
+{
+ struct v10_compute_mqd *m;
+
+ m = get_mqd(mqd);
+
+ m->cp_hqd_pq_control = 5 << CP_HQD_PQ_CONTROL__RPTR_BLOCK_SIZE__SHIFT;
+ m->cp_hqd_pq_control |=
+ ffs(q->queue_size / sizeof(unsigned int)) - 1 - 1;
+ pr_debug("cp_hqd_pq_control 0x%x\n", m->cp_hqd_pq_control);
+
+ m->cp_hqd_pq_base_lo = lower_32_bits((uint64_t)q->queue_address >> 8);
+ m->cp_hqd_pq_base_hi = upper_32_bits((uint64_t)q->queue_address >> 8);
+
+ m->cp_hqd_pq_rptr_report_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
+ m->cp_hqd_pq_rptr_report_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
+ m->cp_hqd_pq_wptr_poll_addr_lo = lower_32_bits((uint64_t)q->write_ptr);
+ m->cp_hqd_pq_wptr_poll_addr_hi = upper_32_bits((uint64_t)q->write_ptr);
+
+ m->cp_hqd_pq_doorbell_control =
+ q->doorbell_off <<
+ CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_OFFSET__SHIFT;
+ pr_debug("cp_hqd_pq_doorbell_control 0x%x\n",
+ m->cp_hqd_pq_doorbell_control);
+
+ m->cp_hqd_ib_control = 3 << CP_HQD_IB_CONTROL__MIN_IB_AVAIL_SIZE__SHIFT;
+
+ /*
+ * HW does not clamp this field correctly. Maximum EOP queue size
+ * is constrained by per-SE EOP done signal count, which is 8-bit.
+ * Limit is 0xFF EOP entries (= 0x7F8 dwords). CP will not submit
+ * more than (EOP entry count - 1) so a queue size of 0x800 dwords
+ * is safe, giving a maximum field value of 0xA.
+ */
+ m->cp_hqd_eop_control = min(0xA,
+ ffs(q->eop_ring_buffer_size / sizeof(unsigned int)) - 1 - 1);
+ m->cp_hqd_eop_base_addr_lo =
+ lower_32_bits(q->eop_ring_buffer_address >> 8);
+ m->cp_hqd_eop_base_addr_hi =
+ upper_32_bits(q->eop_ring_buffer_address >> 8);
+
+ m->cp_hqd_iq_timer = 0;
+
+ m->cp_hqd_vmid = q->vmid;
+
+ if (q->format == KFD_QUEUE_FORMAT_AQL) {
+ /* GC 10 removed WPP_CLAMP from PQ Control */
+ m->cp_hqd_pq_control |= CP_HQD_PQ_CONTROL__NO_UPDATE_RPTR_MASK |
+ 2 << CP_HQD_PQ_CONTROL__SLOT_BASED_WPTR__SHIFT |
+ 1 << CP_HQD_PQ_CONTROL__QUEUE_FULL_EN__SHIFT ;
+ m->cp_hqd_pq_doorbell_control |=
+ 1 << CP_HQD_PQ_DOORBELL_CONTROL__DOORBELL_BIF_DROP__SHIFT;
+ }
+ if (mm->dev->cwsr_enabled)
+ m->cp_hqd_ctx_save_control = 0;
+
+ update_cu_mask(mm, mqd, q);
+
+ q->is_active = (q->queue_size > 0 &&
+ q->queue_address != 0 &&
+ q->queue_percent > 0 &&
+ !q->is_evicted);
+
+ return 0;
+}
+
+static int destroy_mqd(struct mqd_manager *mm, void *mqd,
+ enum kfd_preempt_type type,
+ unsigned int timeout, uint32_t pipe_id,
+ uint32_t queue_id)
+{
+ return mm->dev->kfd2kgd->hqd_destroy
+ (mm->dev->kgd, mqd, type, timeout,
+ pipe_id, queue_id);
+}
+
+static void uninit_mqd(struct mqd_manager *mm, void *mqd,
+ struct kfd_mem_obj *mqd_mem_obj)
+{
+ struct kfd_dev *kfd = mm->dev;
+
+ if (mqd_mem_obj->gtt_mem) {
+ amdgpu_amdkfd_free_gtt_mem(kfd->kgd, mqd_mem_obj->gtt_mem);
+ kfree(mqd_mem_obj);
+ } else {
+ kfd_gtt_sa_free(mm->dev, mqd_mem_obj);
+ }
+}
+
+static bool is_occupied(struct mqd_manager *mm, void *mqd,
+ uint64_t queue_address, uint32_t pipe_id,
+ uint32_t queue_id)
+{
+ return mm->dev->kfd2kgd->hqd_is_occupied(
+ mm->dev->kgd, queue_address,
+ pipe_id, queue_id);
+}
+
+static int get_wave_state(struct mqd_manager *mm, void *mqd,
+ void __user *ctl_stack,
+ u32 *ctl_stack_used_size,
+ u32 *save_area_used_size)
+{
+ struct v10_compute_mqd *m;
+
+ /* Control stack is located one page after MQD. */
+ void *mqd_ctl_stack = (void *)((uintptr_t)mqd + PAGE_SIZE);
+
+ m = get_mqd(mqd);
+
+ *ctl_stack_used_size = m->cp_hqd_cntl_stack_size -
+ m->cp_hqd_cntl_stack_offset;
+ *save_area_used_size = m->cp_hqd_wg_state_offset -
+ m->cp_hqd_cntl_stack_size;
+
+ if (copy_to_user(ctl_stack, mqd_ctl_stack, m->cp_hqd_cntl_stack_size))
+ return -EFAULT;
+
+ return 0;
+}
+
+static int init_mqd_hiq(struct mqd_manager *mm, void **mqd,
+ struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
+ struct queue_properties *q)
+{
+ struct v10_compute_mqd *m;
+ int retval;
+
+
+ retval = init_mqd(mm, mqd, mqd_mem_obj, gart_addr, q);
+
+ if (retval != 0)
+ return retval;
+
+ m = get_mqd(*mqd);
+
+ m->cp_hqd_pq_control |= 1 << CP_HQD_PQ_CONTROL__PRIV_STATE__SHIFT |
+ 1 << CP_HQD_PQ_CONTROL__KMD_QUEUE__SHIFT;
+
+ return retval;
+}
+
+static int update_mqd_hiq(struct mqd_manager *mm, void *mqd,
+ struct queue_properties *q)
+{
+ struct v10_compute_mqd *m;
+ int retval;
+
+ retval = update_mqd(mm, mqd, q);
+
+ if (retval != 0)
+ return retval;
+
+ /* TODO: what's the point? update_mqd already does this. */
+ m = get_mqd(mqd);
+ m->cp_hqd_vmid = q->vmid;
+ return retval;
+}
+
+static int init_mqd_sdma(struct mqd_manager *mm, void **mqd,
+ struct kfd_mem_obj **mqd_mem_obj, uint64_t *gart_addr,
+ struct queue_properties *q)
+{
+ int retval;
+ struct v10_sdma_mqd *m;
+
+
+ retval = kfd_gtt_sa_allocate(mm->dev,
+ sizeof(struct v10_sdma_mqd),
+ mqd_mem_obj);
+
+ if (retval != 0)
+ return -ENOMEM;
+
+ m = (struct v10_sdma_mqd *) (*mqd_mem_obj)->cpu_ptr;
+
+ memset(m, 0, sizeof(struct v10_sdma_mqd));
+
+ *mqd = m;
+ if (gart_addr)
+ *gart_addr = (*mqd_mem_obj)->gpu_addr;
+
+ retval = mm->update_mqd(mm, m, q);
+
+ return retval;
+}
+
+static void uninit_mqd_sdma(struct mqd_manager *mm, void *mqd,
+ struct kfd_mem_obj *mqd_mem_obj)
+{
+ kfd_gtt_sa_free(mm->dev, mqd_mem_obj);
+}
+
+static int load_mqd_sdma(struct mqd_manager *mm, void *mqd,
+ uint32_t pipe_id, uint32_t queue_id,
+ struct queue_properties *p, struct mm_struct *mms)
+{
+ return mm->dev->kfd2kgd->hqd_sdma_load(mm->dev->kgd, mqd,
+ (uint32_t __user *)p->write_ptr,
+ mms);
+}
+
+#define SDMA_RLC_DUMMY_DEFAULT 0xf
+
+static int update_mqd_sdma(struct mqd_manager *mm, void *mqd,
+ struct queue_properties *q)
+{
+ struct v10_sdma_mqd *m;
+
+ m = get_sdma_mqd(mqd);
+ m->sdmax_rlcx_rb_cntl = (ffs(q->queue_size / sizeof(unsigned int)) - 1)
+ << SDMA0_RLC0_RB_CNTL__RB_SIZE__SHIFT |
+ q->vmid << SDMA0_RLC0_RB_CNTL__RB_VMID__SHIFT |
+ 1 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_ENABLE__SHIFT |
+ 6 << SDMA0_RLC0_RB_CNTL__RPTR_WRITEBACK_TIMER__SHIFT;
+
+ m->sdmax_rlcx_rb_base = lower_32_bits(q->queue_address >> 8);
+ m->sdmax_rlcx_rb_base_hi = upper_32_bits(q->queue_address >> 8);
+ m->sdmax_rlcx_rb_rptr_addr_lo = lower_32_bits((uint64_t)q->read_ptr);
+ m->sdmax_rlcx_rb_rptr_addr_hi = upper_32_bits((uint64_t)q->read_ptr);
+ m->sdmax_rlcx_doorbell_offset =
+ q->doorbell_off << SDMA0_RLC0_DOORBELL_OFFSET__OFFSET__SHIFT;
+
+ m->sdma_engine_id = q->sdma_engine_id;
+ m->sdma_queue_id = q->sdma_queue_id;
+ m->sdmax_rlcx_dummy_reg = SDMA_RLC_DUMMY_DEFAULT;
+
+
+ q->is_active = (q->queue_size > 0 &&
+ q->queue_address != 0 &&
+ q->queue_percent > 0 &&
+ !q->is_evicted);
+ return 0;
+}
+
+/*
+ * * preempt type here is ignored because there is only one way
+ * * to preempt sdma queue
+ */
+static int destroy_mqd_sdma(struct mqd_manager *mm, void *mqd,
+ enum kfd_preempt_type type,
+ unsigned int timeout, uint32_t pipe_id,
+ uint32_t queue_id)
+{
+ return mm->dev->kfd2kgd->hqd_sdma_destroy(mm->dev->kgd, mqd, timeout);
+}
+
+static bool is_occupied_sdma(struct mqd_manager *mm, void *mqd,
+ uint64_t queue_address, uint32_t pipe_id,
+ uint32_t queue_id)
+{
+ return mm->dev->kfd2kgd->hqd_sdma_is_occupied(mm->dev->kgd, mqd);
+}
+
+#if defined(CONFIG_DEBUG_FS)
+
+static int debugfs_show_mqd(struct seq_file *m, void *data)
+{
+ seq_hex_dump(m, " ", DUMP_PREFIX_OFFSET, 32, 4,
+ data, sizeof(struct v10_compute_mqd), false);
+ return 0;
+}
+
+static int debugfs_show_mqd_sdma(struct seq_file *m, void *data)
+{
+ seq_hex_dump(m, " ", DUMP_PREFIX_OFFSET, 32, 4,
+ data, sizeof(struct v10_sdma_mqd), false);
+ return 0;
+}
+
+#endif
+
+struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type,
+ struct kfd_dev *dev)
+{
+ struct mqd_manager *mqd;
+
+ if (WARN_ON(type >= KFD_MQD_TYPE_MAX))
+ return NULL;
+
+ mqd = kzalloc(sizeof(*mqd), GFP_NOIO);
+ if (!mqd)
+ return NULL;
+
+ mqd->dev = dev;
+
+ switch (type) {
+ case KFD_MQD_TYPE_CP:
+ pr_debug("%s@%i\n", __func__, __LINE__);
+ case KFD_MQD_TYPE_COMPUTE:
+ pr_debug("%s@%i\n", __func__, __LINE__);
+ mqd->init_mqd = init_mqd;
+ mqd->uninit_mqd = uninit_mqd;
+ mqd->load_mqd = load_mqd;
+ mqd->update_mqd = update_mqd;
+ mqd->destroy_mqd = destroy_mqd;
+ mqd->is_occupied = is_occupied;
+ mqd->get_wave_state = get_wave_state;
+#if defined(CONFIG_DEBUG_FS)
+ mqd->debugfs_show_mqd = debugfs_show_mqd;
+#endif
+ pr_debug("%s@%i\n", __func__, __LINE__);
+ break;
+ case KFD_MQD_TYPE_HIQ:
+ pr_debug("%s@%i\n", __func__, __LINE__);
+ mqd->init_mqd = init_mqd_hiq;
+ mqd->uninit_mqd = uninit_mqd;
+ mqd->load_mqd = load_mqd;
+ mqd->update_mqd = update_mqd_hiq;
+ mqd->destroy_mqd = destroy_mqd;
+ mqd->is_occupied = is_occupied;
+#if defined(CONFIG_DEBUG_FS)
+ mqd->debugfs_show_mqd = debugfs_show_mqd;
+#endif
+ pr_debug("%s@%i\n", __func__, __LINE__);
+ break;
+ case KFD_MQD_TYPE_SDMA:
+ pr_debug("%s@%i\n", __func__, __LINE__);
+ mqd->init_mqd = init_mqd_sdma;
+ mqd->uninit_mqd = uninit_mqd_sdma;
+ mqd->load_mqd = load_mqd_sdma;
+ mqd->update_mqd = update_mqd_sdma;
+ mqd->destroy_mqd = destroy_mqd_sdma;
+ mqd->is_occupied = is_occupied_sdma;
+#if defined(CONFIG_DEBUG_FS)
+ mqd->debugfs_show_mqd = debugfs_show_mqd_sdma;
+#endif
+ pr_debug("%s@%i\n", __func__, __LINE__);
+ break;
+ default:
+ kfree(mqd);
+ return NULL;
+ }
+
+ return mqd;
+}
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
index 808194663a7d..c72c8f5fd54c 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_packet_manager.c
@@ -237,6 +237,9 @@ int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm)
case CHIP_RAVEN:
pm->pmf = &kfd_v9_pm_funcs;
break;
+ case CHIP_NAVI10:
+ pm->pmf = &kfd_v10_pm_funcs;
+ break;
default:
WARN(1, "Unexpected ASIC family %u",
dqm->dev->device_info->asic_family);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index da589ee1366c..40e40d1e4dd2 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -171,6 +171,10 @@ enum cache_policy {
cache_policy_noncoherent
};
+#define KFD_IS_VI(chip) ((chip) >= CHIP_CARRIZO && (chip) <= CHIP_POLARIS11)
+#define KFD_IS_DGPU(chip) (((chip) >= CHIP_TONGA && \
+ (chip) <= CHIP_NAVI10) || \
+ (chip) == CHIP_HAWAII)
#define KFD_IS_SOC15(chip) ((chip) >= CHIP_VEGA10)
struct kfd_event_interrupt_class {
@@ -861,6 +865,8 @@ struct mqd_manager *mqd_manager_init_vi_tonga(enum KFD_MQD_TYPE type,
struct kfd_dev *dev);
struct mqd_manager *mqd_manager_init_v9(enum KFD_MQD_TYPE type,
struct kfd_dev *dev);
+struct mqd_manager *mqd_manager_init_v10(enum KFD_MQD_TYPE type,
+ struct kfd_dev *dev);
struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev);
void device_queue_manager_uninit(struct device_queue_manager *dqm);
struct kernel_queue *kernel_queue_init(struct kfd_dev *dev,
@@ -950,6 +956,7 @@ struct packet_manager_funcs {
extern const struct packet_manager_funcs kfd_vi_pm_funcs;
extern const struct packet_manager_funcs kfd_v9_pm_funcs;
+extern const struct packet_manager_funcs kfd_v10_pm_funcs;
int pm_init(struct packet_manager *pm, struct device_queue_manager *dqm);
void pm_uninit(struct packet_manager *pm);
@@ -969,7 +976,9 @@ void pm_release_ib(struct packet_manager *pm);
/* Following PM funcs can be shared among VI and AI */
unsigned int pm_build_pm4_header(unsigned int opcode, size_t packet_size);
int pm_set_resources_vi(struct packet_manager *pm, uint32_t *buffer,
- struct scheduling_resources *res);
+ struct scheduling_resources *res);
+void kfd_pm_func_init_v10(struct packet_manager *pm, uint16_t fw_ver);
+
uint64_t kfd_get_number_elems(struct kfd_dev *kfd);
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 4bdae78bab8e..8382742e296a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -1107,3 +1107,4 @@ int kfd_debugfs_mqds_by_process(struct seq_file *m, void *data)
}
#endif
+
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
index 2c40ab4fe8de..c2e6e47abaf2 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_topology.c
@@ -1321,6 +1321,7 @@ int kfd_topology_add_device(struct kfd_dev *gpu)
case CHIP_VEGA12:
case CHIP_VEGA20:
case CHIP_RAVEN:
+ case CHIP_NAVI10:
dev->node_props.capability |= ((HSA_CAP_DOORBELL_TYPE_2_0 <<
HSA_CAP_DOORBELL_TYPE_TOTALBITS_SHIFT) &
HSA_CAP_DOORBELL_TYPE_TOTALBITS_MASK);
--
2.20.1
More information about the amd-gfx
mailing list