[PATCH 2/4] drm/amdkfd: Add initial VI support for DQM

From: Ben Goz ben.goz@amd.com

This patch starts to add support for the VI APU in the DQM module.

Because most (more than 90%) of the DQM code is shared among AMD's APUs, we chose a design that performs most/all the code in the shared DQM file (kfd_device_queue_manager.c). If there is H/W specific code to be executed, than it is written in an asic-specific extension function for that H/W.

That asic-specific extension function is called from the shared function at the appropriate time. This requires that for every asic-specific extension function that is implemented in a specific ASIC, there will be an equivalent implementation in ALL ASICs, even if those implementations are just stubs.

That way we achieve:

- Maintainability: by having one copy of most of the code, we only need to fix bugs at one locations

- Readability: very clear what is the shared code and what is done per ASIC

- Extensibility: very easy to add new H/W specific files/functions

Signed-off-by: Ben Goz ben.goz@amd.com Signed-off-by: Oded Gabbay oded.gabbay@amd.com --- drivers/gpu/drm/amd/amdkfd/Makefile | 1 + .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.c | 113 +++++------------ .../gpu/drm/amd/amdkfd/kfd_device_queue_manager.h | 11 +- .../drm/amd/amdkfd/kfd_device_queue_manager_cik.c | 135 +++++++++++++++++++++ .../drm/amd/amdkfd/kfd_device_queue_manager_vi.c | 64 ++++++++++ 5 files changed, 238 insertions(+), 86 deletions(-) create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c create mode 100644 drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c

diff --git a/drivers/gpu/drm/amd/amdkfd/Makefile b/drivers/gpu/drm/amd/amdkfd/Makefile index bc6053f..7558683 100644 --- a/drivers/gpu/drm/amd/amdkfd/Makefile +++ b/drivers/gpu/drm/amd/amdkfd/Makefile @@ -10,6 +10,7 @@ amdkfd-y := kfd_module.o kfd_device.o kfd_chardev.o kfd_topology.o \ kfd_mqd_manager_cik.o kfd_mqd_manager_vi.o \ kfd_kernel_queue.o kfd_packet_manager.o \ kfd_process_queue_manager.o kfd_device_queue_manager.o \ + kfd_device_queue_manager_cik.o kfd_device_queue_manager_vi.o \ kfd_interrupt.o

obj-$(CONFIG_HSA_AMD) += amdkfd.o 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 12c8448..b201624 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c @@ -61,7 +61,7 @@ enum KFD_MQD_TYPE get_mqd_type_from_queue_type(enum kfd_queue_type type) return KFD_MQD_TYPE_CP; }

-static inline unsigned int get_pipes_num(struct device_queue_manager *dqm) +inline unsigned int get_pipes_num(struct device_queue_manager *dqm) { BUG_ON(!dqm || !dqm->dev); return dqm->dev->shared_resources.compute_pipe_count; @@ -78,7 +78,7 @@ static inline unsigned int get_pipes_num_cpsch(void) return PIPE_PER_ME_CP_SCHEDULING; }

-static inline unsigned int +inline unsigned int get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd) { uint32_t nybble; @@ -88,7 +88,7 @@ get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd) return nybble; }

-static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd) +inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd) { unsigned int shared_base;

@@ -97,41 +97,7 @@ static inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *pdd) return shared_base; }

-static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble); -static void init_process_memory(struct device_queue_manager *dqm, - struct qcm_process_device *qpd) -{ - struct kfd_process_device *pdd; - unsigned int temp; - - BUG_ON(!dqm || !qpd); - - pdd = qpd_to_pdd(qpd); - - /* check if sh_mem_config register already configured */ - if (qpd->sh_mem_config == 0) { - qpd->sh_mem_config = - ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED) | - DEFAULT_MTYPE(MTYPE_NONCACHED) | - APE1_MTYPE(MTYPE_NONCACHED); - qpd->sh_mem_ape1_limit = 0; - qpd->sh_mem_ape1_base = 0; - } - - if (qpd->pqm->process->is_32bit_user_mode) { - temp = get_sh_mem_bases_32(pdd); - qpd->sh_mem_bases = SHARED_BASE(temp); - qpd->sh_mem_config |= PTR32; - } else { - temp = get_sh_mem_bases_nybble_64(pdd); - qpd->sh_mem_bases = compute_sh_mem_bases_64bit(temp); - } - - pr_debug("kfd: is32bit process: %d sh_mem_bases nybble: 0x%X and register 0x%X\n", - qpd->pqm->process->is_32bit_user_mode, temp, qpd->sh_mem_bases); -} - -static void program_sh_mem_settings(struct device_queue_manager *dqm, +void program_sh_mem_settings(struct device_queue_manager *dqm, struct qcm_process_device *qpd) { return kfd2kgd->program_sh_mem_settings(dqm->dev->kgd, qpd->vmid, @@ -391,6 +357,7 @@ static int register_process_nocpsch(struct device_queue_manager *dqm, struct qcm_process_device *qpd) { struct device_process_node *n; + int retval;

BUG_ON(!dqm || !qpd);

@@ -405,12 +372,13 @@ static int register_process_nocpsch(struct device_queue_manager *dqm, mutex_lock(&dqm->lock); list_add(&n->list, &dqm->queues);

- init_process_memory(dqm, qpd); + retval = dqm->ops_asic_specific.register_process(dqm, qpd); + dqm->processes_count++;

mutex_unlock(&dqm->lock);

- return 0; + return retval; }

static int unregister_process_nocpsch(struct device_queue_manager *dqm, @@ -455,34 +423,7 @@ set_pasid_vmid_mapping(struct device_queue_manager *dqm, unsigned int pasid, vmid); }

-static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble) -{ - /* In 64-bit mode, we can only control the top 3 bits of the LDS, - * scratch and GPUVM apertures. - * The hardware fills in the remaining 59 bits according to the - * following pattern: - * LDS: X0000000'00000000 - X0000001'00000000 (4GB) - * Scratch: X0000001'00000000 - X0000002'00000000 (4GB) - * GPUVM: Y0010000'00000000 - Y0020000'00000000 (1TB) - * - * (where X/Y is the configurable nybble with the low-bit 0) - * - * LDS and scratch will have the same top nybble programmed in the - * top 3 bits of SH_MEM_BASES.PRIVATE_BASE. - * GPUVM can have a different top nybble programmed in the - * top 3 bits of SH_MEM_BASES.SHARED_BASE. - * We don't bother to support different top nybbles - * for LDS/Scratch and GPUVM. - */ - - BUG_ON((top_address_nybble & 1) || top_address_nybble > 0xE || - top_address_nybble == 0); - - return PRIVATE_BASE(top_address_nybble << 12) | - SHARED_BASE(top_address_nybble << 12); -} - -static int init_pipelines(struct device_queue_manager *dqm, +int init_pipelines(struct device_queue_manager *dqm, unsigned int pipes_num, unsigned int first_pipe) { void *hpdptr; @@ -715,7 +656,7 @@ static int initialize_cpsch(struct device_queue_manager *dqm) dqm->queue_count = dqm->processes_count = 0; dqm->sdma_queue_count = 0; dqm->active_runlist = false; - retval = init_pipelines(dqm, get_pipes_num(dqm), 0); + retval = dqm->ops_asic_specific.initialize(dqm); if (retval != 0) goto fail_init_pipelines;

@@ -1035,8 +976,7 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm, void __user *alternate_aperture_base, uint64_t alternate_aperture_size) { - uint32_t default_mtype; - uint32_t ape1_mtype; + bool retval;

pr_debug("kfd: In func %s\n", __func__);

@@ -1073,18 +1013,13 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm, qpd->sh_mem_ape1_limit = limit >> 16; }

- default_mtype = (default_policy == cache_policy_coherent) ? - MTYPE_NONCACHED : - MTYPE_CACHED; - - ape1_mtype = (alternate_policy == cache_policy_coherent) ? - MTYPE_NONCACHED : - MTYPE_CACHED; - - qpd->sh_mem_config = (qpd->sh_mem_config & PTR32) - | ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED) - | DEFAULT_MTYPE(default_mtype) - | APE1_MTYPE(ape1_mtype); + retval = dqm->ops_asic_specific.set_cache_memory_policy( + dqm, + qpd, + default_policy, + alternate_policy, + alternate_aperture_base, + alternate_aperture_size);

if ((sched_policy == KFD_SCHED_POLICY_NO_HWS) && (qpd->vmid != 0)) program_sh_mem_settings(dqm, qpd); @@ -1094,7 +1029,7 @@ static bool set_cache_memory_policy(struct device_queue_manager *dqm, qpd->sh_mem_ape1_limit);

mutex_unlock(&dqm->lock); - return true; + return retval;

out: mutex_unlock(&dqm->lock); @@ -1107,6 +1042,8 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev)

BUG_ON(!dev);

+ pr_debug("kfd: loading device queue manager\n"); + dqm = kzalloc(sizeof(struct device_queue_manager), GFP_KERNEL); if (!dqm) return NULL; @@ -1149,6 +1086,13 @@ struct device_queue_manager *device_queue_manager_init(struct kfd_dev *dev) break; }

+ switch (dev->device_info->asic_family) { + case CHIP_CARRIZO: + device_queue_manager_init_vi(&dqm->ops_asic_specific); + case CHIP_KAVERI: + device_queue_manager_init_cik(&dqm->ops_asic_specific); + } + if (dqm->ops.initialize(dqm) != 0) { kfree(dqm); return NULL; @@ -1164,4 +1108,3 @@ void device_queue_manager_uninit(struct device_queue_manager *dqm) dqm->ops.uninitialize(dqm); kfree(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 72d2ca0..1934795 100644 --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.h @@ -134,6 +134,7 @@ struct device_queue_manager_ops {

struct device_queue_manager { struct device_queue_manager_ops ops; + struct device_queue_manager_ops ops_asic_specific;

struct mqd_manager *mqds[KFD_MQD_TYPE_MAX]; struct packet_manager packets; @@ -155,6 +156,14 @@ struct device_queue_manager { bool active_runlist; };

- +void device_queue_manager_init_cik(struct device_queue_manager_ops *ops); +void device_queue_manager_init_vi(struct device_queue_manager_ops *ops); +void program_sh_mem_settings(struct device_queue_manager *dqm, + struct qcm_process_device *qpd); +inline unsigned int get_sh_mem_bases_32(struct kfd_process_device *qpd); +inline unsigned int get_sh_mem_bases_nybble_64(struct kfd_process_device *pdd); +int init_pipelines(struct device_queue_manager *dqm, + unsigned int pipes_num, unsigned int first_pipe); +inline unsigned int get_pipes_num(struct device_queue_manager *dqm);

#endif /* KFD_DEVICE_QUEUE_MANAGER_H_ */ diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c new file mode 100644 index 0000000..6b07246 --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_cik.c @@ -0,0 +1,135 @@ +/* + * 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 "kfd_device_queue_manager.h" +#include "cik_regs.h" + +static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm, + struct qcm_process_device *qpd, + enum cache_policy default_policy, + enum cache_policy alternate_policy, + void __user *alternate_aperture_base, + uint64_t alternate_aperture_size); +static int register_process_cik(struct device_queue_manager *dqm, + struct qcm_process_device *qpd); +static int initialize_cpsch_cik(struct device_queue_manager *dqm); + +void device_queue_manager_init_cik(struct device_queue_manager_ops *ops) +{ + ops->set_cache_memory_policy = set_cache_memory_policy_cik; + ops->register_process = register_process_cik; + ops->initialize = initialize_cpsch_cik; +} + +static uint32_t compute_sh_mem_bases_64bit(unsigned int top_address_nybble) +{ + /* In 64-bit mode, we can only control the top 3 bits of the LDS, + * scratch and GPUVM apertures. + * The hardware fills in the remaining 59 bits according to the + * following pattern: + * LDS: X0000000'00000000 - X0000001'00000000 (4GB) + * Scratch: X0000001'00000000 - X0000002'00000000 (4GB) + * GPUVM: Y0010000'00000000 - Y0020000'00000000 (1TB) + * + * (where X/Y is the configurable nybble with the low-bit 0) + * + * LDS and scratch will have the same top nybble programmed in the + * top 3 bits of SH_MEM_BASES.PRIVATE_BASE. + * GPUVM can have a different top nybble programmed in the + * top 3 bits of SH_MEM_BASES.SHARED_BASE. + * We don't bother to support different top nybbles + * for LDS/Scratch and GPUVM. + */ + + BUG_ON((top_address_nybble & 1) || top_address_nybble > 0xE || + top_address_nybble == 0); + + return PRIVATE_BASE(top_address_nybble << 12) | + SHARED_BASE(top_address_nybble << 12); +} + +static bool set_cache_memory_policy_cik(struct device_queue_manager *dqm, + struct qcm_process_device *qpd, + enum cache_policy default_policy, + enum cache_policy alternate_policy, + void __user *alternate_aperture_base, + uint64_t alternate_aperture_size) +{ + uint32_t default_mtype; + uint32_t ape1_mtype; + + default_mtype = (default_policy == cache_policy_coherent) ? + MTYPE_NONCACHED : + MTYPE_CACHED; + + ape1_mtype = (alternate_policy == cache_policy_coherent) ? + MTYPE_NONCACHED : + MTYPE_CACHED; + + qpd->sh_mem_config = (qpd->sh_mem_config & PTR32) + | ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED) + | DEFAULT_MTYPE(default_mtype) + | APE1_MTYPE(ape1_mtype); + + return true; +} + +static int register_process_cik(struct device_queue_manager *dqm, + struct qcm_process_device *qpd) +{ + struct kfd_process_device *pdd; + unsigned int temp; + + BUG_ON(!dqm || !qpd); + + pdd = qpd_to_pdd(qpd); + + /* check if sh_mem_config register already configured */ + if (qpd->sh_mem_config == 0) { + qpd->sh_mem_config = + ALIGNMENT_MODE(SH_MEM_ALIGNMENT_MODE_UNALIGNED) | + DEFAULT_MTYPE(MTYPE_NONCACHED) | + APE1_MTYPE(MTYPE_NONCACHED); + qpd->sh_mem_ape1_limit = 0; + qpd->sh_mem_ape1_base = 0; + } + + if (qpd->pqm->process->is_32bit_user_mode) { + temp = get_sh_mem_bases_32(pdd); + qpd->sh_mem_bases = SHARED_BASE(temp); + qpd->sh_mem_config |= PTR32; + } else { + temp = get_sh_mem_bases_nybble_64(pdd); + qpd->sh_mem_bases = compute_sh_mem_bases_64bit(temp); + } + + pr_debug("kfd: is32bit process: %d sh_mem_bases nybble: 0x%X and register 0x%X\n", + qpd->pqm->process->is_32bit_user_mode, temp, qpd->sh_mem_bases); + + return 0; +} + +static int initialize_cpsch_cik(struct device_queue_manager *dqm) +{ + return init_pipelines(dqm, get_pipes_num(dqm), 0); +} diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c new file mode 100644 index 0000000..20553dc --- /dev/null +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager_vi.c @@ -0,0 +1,64 @@ +/* + * 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 "kfd_device_queue_manager.h" + +static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm, + struct qcm_process_device *qpd, + enum cache_policy default_policy, + enum cache_policy alternate_policy, + void __user *alternate_aperture_base, + uint64_t alternate_aperture_size); +static int register_process_vi(struct device_queue_manager *dqm, + struct qcm_process_device *qpd); +static int initialize_cpsch_vi(struct device_queue_manager *dqm); + +void device_queue_manager_init_vi(struct device_queue_manager_ops *ops) +{ + pr_warn("amdkfd: VI DQM is not currently supported\n"); + + ops->set_cache_memory_policy = set_cache_memory_policy_vi; + ops->register_process = register_process_vi; + ops->initialize = initialize_cpsch_vi; +} + +static bool set_cache_memory_policy_vi(struct device_queue_manager *dqm, + struct qcm_process_device *qpd, + enum cache_policy default_policy, + enum cache_policy alternate_policy, + void __user *alternate_aperture_base, + uint64_t alternate_aperture_size) +{ + return false; +} + +static int register_process_vi(struct device_queue_manager *dqm, + struct qcm_process_device *qpd) +{ + return -1; +} + +static int initialize_cpsch_vi(struct device_queue_manager *dqm) +{ + return 0; +}