linux_dsm_epyc7002/drivers/gpu/drm/amd/amdkfd/kfd_device.c

1061 lines
30 KiB
C
Raw Normal View History

/*
* 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/bsearch.h>
#include <linux/pci.h>
#include <linux/slab.h>
#include "kfd_priv.h"
2014-07-17 05:27:00 +07:00
#include "kfd_device_queue_manager.h"
#include "kfd_pm4_headers_vi.h"
#include "cwsr_trap_handler.h"
#include "kfd_iommu.h"
#include "amdgpu_amdkfd.h"
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
#define MQD_SIZE_ALIGNED 768
/*
* kfd_locked is used to lock the kfd driver during suspend or reset
* once locked, kfd driver will stop any further GPU execution.
* create process (open) will return -EAGAIN.
*/
static atomic_t kfd_locked = ATOMIC_INIT(0);
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
#ifdef KFD_SUPPORT_IOMMU_V2
static const struct kfd_device_info kaveri_device_info = {
.asic_family = CHIP_KAVERI,
.max_pasid_bits = 16,
/* max num of queues for KV.TODO should be a dynamic value */
.max_no_of_hqd = 24,
.doorbell_size = 4,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
.num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED,
.supports_cwsr = false,
.needs_iommu_device = true,
.needs_pci_atomics = false,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info carrizo_device_info = {
.asic_family = CHIP_CARRIZO,
.max_pasid_bits = 16,
/* max num of queues for CZ.TODO should be a dynamic value */
.max_no_of_hqd = 24,
.doorbell_size = 4,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
.num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED,
.supports_cwsr = true,
.needs_iommu_device = true,
.needs_pci_atomics = false,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info raven_device_info = {
.asic_family = CHIP_RAVEN,
.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,
.supports_cwsr = true,
.needs_iommu_device = true,
.needs_pci_atomics = true,
.num_sdma_engines = 1,
.num_sdma_queues_per_engine = 2,
};
#endif
static const struct kfd_device_info hawaii_device_info = {
.asic_family = CHIP_HAWAII,
.max_pasid_bits = 16,
/* max num of queues for KV.TODO should be a dynamic value */
.max_no_of_hqd = 24,
.doorbell_size = 4,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
.num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED,
.supports_cwsr = false,
.needs_iommu_device = false,
.needs_pci_atomics = false,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info tonga_device_info = {
.asic_family = CHIP_TONGA,
.max_pasid_bits = 16,
.max_no_of_hqd = 24,
.doorbell_size = 4,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
.num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED,
.supports_cwsr = false,
.needs_iommu_device = false,
.needs_pci_atomics = true,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info fiji_device_info = {
.asic_family = CHIP_FIJI,
.max_pasid_bits = 16,
.max_no_of_hqd = 24,
.doorbell_size = 4,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
.num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED,
.supports_cwsr = true,
.needs_iommu_device = false,
.needs_pci_atomics = true,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info fiji_vf_device_info = {
.asic_family = CHIP_FIJI,
.max_pasid_bits = 16,
.max_no_of_hqd = 24,
.doorbell_size = 4,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
.num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED,
.supports_cwsr = true,
.needs_iommu_device = false,
.needs_pci_atomics = false,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info polaris10_device_info = {
.asic_family = CHIP_POLARIS10,
.max_pasid_bits = 16,
.max_no_of_hqd = 24,
.doorbell_size = 4,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
.num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED,
.supports_cwsr = true,
.needs_iommu_device = false,
.needs_pci_atomics = true,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info polaris10_vf_device_info = {
.asic_family = CHIP_POLARIS10,
.max_pasid_bits = 16,
.max_no_of_hqd = 24,
.doorbell_size = 4,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
.num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED,
.supports_cwsr = true,
.needs_iommu_device = false,
.needs_pci_atomics = false,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info polaris11_device_info = {
.asic_family = CHIP_POLARIS11,
.max_pasid_bits = 16,
.max_no_of_hqd = 24,
.doorbell_size = 4,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
.num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED,
.supports_cwsr = true,
.needs_iommu_device = false,
.needs_pci_atomics = true,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info polaris12_device_info = {
.asic_family = CHIP_POLARIS12,
.max_pasid_bits = 16,
.max_no_of_hqd = 24,
.doorbell_size = 4,
.ih_ring_entry_size = 4 * sizeof(uint32_t),
.event_interrupt_class = &event_interrupt_class_cik,
.num_of_watch_points = 4,
.mqd_size_aligned = MQD_SIZE_ALIGNED,
.supports_cwsr = true,
.needs_iommu_device = false,
.needs_pci_atomics = true,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info vega10_device_info = {
.asic_family = CHIP_VEGA10,
.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,
.supports_cwsr = true,
.needs_iommu_device = false,
.needs_pci_atomics = false,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info vega10_vf_device_info = {
.asic_family = CHIP_VEGA10,
.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,
.supports_cwsr = true,
.needs_iommu_device = false,
.needs_pci_atomics = false,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info vega12_device_info = {
.asic_family = CHIP_VEGA12,
.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,
.supports_cwsr = true,
.needs_iommu_device = false,
.needs_pci_atomics = false,
.num_sdma_engines = 2,
.num_sdma_queues_per_engine = 2,
};
static const struct kfd_device_info vega20_device_info = {
.asic_family = CHIP_VEGA20,
.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,
.supports_cwsr = true,
.needs_iommu_device = 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;
};
static const struct kfd_deviceid supported_devices[] = {
#ifdef KFD_SUPPORT_IOMMU_V2
{ 0x1304, &kaveri_device_info }, /* Kaveri */
{ 0x1305, &kaveri_device_info }, /* Kaveri */
{ 0x1306, &kaveri_device_info }, /* Kaveri */
{ 0x1307, &kaveri_device_info }, /* Kaveri */
{ 0x1309, &kaveri_device_info }, /* Kaveri */
{ 0x130A, &kaveri_device_info }, /* Kaveri */
{ 0x130B, &kaveri_device_info }, /* Kaveri */
{ 0x130C, &kaveri_device_info }, /* Kaveri */
{ 0x130D, &kaveri_device_info }, /* Kaveri */
{ 0x130E, &kaveri_device_info }, /* Kaveri */
{ 0x130F, &kaveri_device_info }, /* Kaveri */
{ 0x1310, &kaveri_device_info }, /* Kaveri */
{ 0x1311, &kaveri_device_info }, /* Kaveri */
{ 0x1312, &kaveri_device_info }, /* Kaveri */
{ 0x1313, &kaveri_device_info }, /* Kaveri */
{ 0x1315, &kaveri_device_info }, /* Kaveri */
{ 0x1316, &kaveri_device_info }, /* Kaveri */
{ 0x1317, &kaveri_device_info }, /* Kaveri */
{ 0x1318, &kaveri_device_info }, /* Kaveri */
{ 0x131B, &kaveri_device_info }, /* Kaveri */
{ 0x131C, &kaveri_device_info }, /* Kaveri */
{ 0x131D, &kaveri_device_info }, /* Kaveri */
{ 0x9870, &carrizo_device_info }, /* Carrizo */
{ 0x9874, &carrizo_device_info }, /* Carrizo */
{ 0x9875, &carrizo_device_info }, /* Carrizo */
{ 0x9876, &carrizo_device_info }, /* Carrizo */
{ 0x9877, &carrizo_device_info }, /* Carrizo */
{ 0x15DD, &raven_device_info }, /* Raven */
{ 0x15D8, &raven_device_info }, /* Raven */
#endif
{ 0x67A0, &hawaii_device_info }, /* Hawaii */
{ 0x67A1, &hawaii_device_info }, /* Hawaii */
{ 0x67A2, &hawaii_device_info }, /* Hawaii */
{ 0x67A8, &hawaii_device_info }, /* Hawaii */
{ 0x67A9, &hawaii_device_info }, /* Hawaii */
{ 0x67AA, &hawaii_device_info }, /* Hawaii */
{ 0x67B0, &hawaii_device_info }, /* Hawaii */
{ 0x67B1, &hawaii_device_info }, /* Hawaii */
{ 0x67B8, &hawaii_device_info }, /* Hawaii */
{ 0x67B9, &hawaii_device_info }, /* Hawaii */
{ 0x67BA, &hawaii_device_info }, /* Hawaii */
{ 0x67BE, &hawaii_device_info }, /* Hawaii */
{ 0x6920, &tonga_device_info }, /* Tonga */
{ 0x6921, &tonga_device_info }, /* Tonga */
{ 0x6928, &tonga_device_info }, /* Tonga */
{ 0x6929, &tonga_device_info }, /* Tonga */
{ 0x692B, &tonga_device_info }, /* Tonga */
{ 0x6938, &tonga_device_info }, /* Tonga */
{ 0x6939, &tonga_device_info }, /* Tonga */
{ 0x7300, &fiji_device_info }, /* Fiji */
{ 0x730F, &fiji_vf_device_info }, /* Fiji vf*/
{ 0x67C0, &polaris10_device_info }, /* Polaris10 */
{ 0x67C1, &polaris10_device_info }, /* Polaris10 */
{ 0x67C2, &polaris10_device_info }, /* Polaris10 */
{ 0x67C4, &polaris10_device_info }, /* Polaris10 */
{ 0x67C7, &polaris10_device_info }, /* Polaris10 */
{ 0x67C8, &polaris10_device_info }, /* Polaris10 */
{ 0x67C9, &polaris10_device_info }, /* Polaris10 */
{ 0x67CA, &polaris10_device_info }, /* Polaris10 */
{ 0x67CC, &polaris10_device_info }, /* Polaris10 */
{ 0x67CF, &polaris10_device_info }, /* Polaris10 */
{ 0x67D0, &polaris10_vf_device_info }, /* Polaris10 vf*/
{ 0x67DF, &polaris10_device_info }, /* Polaris10 */
{ 0x67E0, &polaris11_device_info }, /* Polaris11 */
{ 0x67E1, &polaris11_device_info }, /* Polaris11 */
{ 0x67E3, &polaris11_device_info }, /* Polaris11 */
{ 0x67E7, &polaris11_device_info }, /* Polaris11 */
{ 0x67E8, &polaris11_device_info }, /* Polaris11 */
{ 0x67E9, &polaris11_device_info }, /* Polaris11 */
{ 0x67EB, &polaris11_device_info }, /* Polaris11 */
{ 0x67EF, &polaris11_device_info }, /* Polaris11 */
{ 0x67FF, &polaris11_device_info }, /* Polaris11 */
{ 0x6980, &polaris12_device_info }, /* Polaris12 */
{ 0x6981, &polaris12_device_info }, /* Polaris12 */
{ 0x6985, &polaris12_device_info }, /* Polaris12 */
{ 0x6986, &polaris12_device_info }, /* Polaris12 */
{ 0x6987, &polaris12_device_info }, /* Polaris12 */
{ 0x6995, &polaris12_device_info }, /* Polaris12 */
{ 0x6997, &polaris12_device_info }, /* Polaris12 */
{ 0x699F, &polaris12_device_info }, /* Polaris12 */
{ 0x6860, &vega10_device_info }, /* Vega10 */
{ 0x6861, &vega10_device_info }, /* Vega10 */
{ 0x6862, &vega10_device_info }, /* Vega10 */
{ 0x6863, &vega10_device_info }, /* Vega10 */
{ 0x6864, &vega10_device_info }, /* Vega10 */
{ 0x6867, &vega10_device_info }, /* Vega10 */
{ 0x6868, &vega10_device_info }, /* Vega10 */
{ 0x6869, &vega10_device_info }, /* Vega10 */
{ 0x686A, &vega10_device_info }, /* Vega10 */
{ 0x686B, &vega10_device_info }, /* Vega10 */
{ 0x686C, &vega10_vf_device_info }, /* Vega10 vf*/
{ 0x686D, &vega10_device_info }, /* Vega10 */
{ 0x686E, &vega10_device_info }, /* Vega10 */
{ 0x686F, &vega10_device_info }, /* Vega10 */
{ 0x687F, &vega10_device_info }, /* Vega10 */
{ 0x69A0, &vega12_device_info }, /* Vega12 */
{ 0x69A1, &vega12_device_info }, /* Vega12 */
{ 0x69A2, &vega12_device_info }, /* Vega12 */
{ 0x69A3, &vega12_device_info }, /* Vega12 */
{ 0x69AF, &vega12_device_info }, /* Vega12 */
{ 0x66a0, &vega20_device_info }, /* Vega20 */
{ 0x66a1, &vega20_device_info }, /* Vega20 */
{ 0x66a2, &vega20_device_info }, /* Vega20 */
{ 0x66a3, &vega20_device_info }, /* Vega20 */
{ 0x66a4, &vega20_device_info }, /* Vega20 */
{ 0x66a7, &vega20_device_info }, /* Vega20 */
{ 0x66af, &vega20_device_info } /* Vega20 */
};
static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
unsigned int chunk_size);
static void kfd_gtt_sa_fini(struct kfd_dev *kfd);
static int kfd_resume(struct kfd_dev *kfd);
static const struct kfd_device_info *lookup_device_info(unsigned short did)
{
size_t i;
for (i = 0; i < ARRAY_SIZE(supported_devices); i++) {
if (supported_devices[i].did == did) {
WARN_ON(!supported_devices[i].device_info);
return supported_devices[i].device_info;
}
}
dev_warn(kfd_device, "DID %04x is missing in supported_devices\n",
did);
return NULL;
}
struct kfd_dev *kgd2kfd_probe(struct kgd_dev *kgd,
struct pci_dev *pdev, const struct kfd2kgd_calls *f2g)
{
struct kfd_dev *kfd;
int ret;
const struct kfd_device_info *device_info =
lookup_device_info(pdev->device);
if (!device_info) {
dev_err(kfd_device, "kgd2kfd_probe failed\n");
return NULL;
}
kfd = kzalloc(sizeof(*kfd), GFP_KERNEL);
if (!kfd)
return NULL;
/* Allow BIF to recode atomics to PCIe 3.0 AtomicOps.
* 32 and 64-bit requests are possible and must be
* supported.
*/
ret = pci_enable_atomic_ops_to_root(pdev,
PCI_EXP_DEVCAP2_ATOMIC_COMP32 |
PCI_EXP_DEVCAP2_ATOMIC_COMP64);
if (device_info->needs_pci_atomics && ret < 0) {
dev_info(kfd_device,
"skipped device %x:%x, PCI rejects atomics\n",
pdev->vendor, pdev->device);
kfree(kfd);
return NULL;
} else if (!ret)
kfd->pci_atomic_requested = true;
kfd->kgd = kgd;
kfd->device_info = device_info;
kfd->pdev = pdev;
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
kfd->init_complete = false;
kfd->kfd2kgd = f2g;
mutex_init(&kfd->doorbell_mutex);
memset(&kfd->doorbell_available_index, 0,
sizeof(kfd->doorbell_available_index));
atomic_set(&kfd->sram_ecc_flag, 0);
return kfd;
}
static void kfd_cwsr_init(struct kfd_dev *kfd)
{
if (cwsr_enable && kfd->device_info->supports_cwsr) {
if (kfd->device_info->asic_family < CHIP_VEGA10) {
BUILD_BUG_ON(sizeof(cwsr_trap_gfx8_hex) > PAGE_SIZE);
kfd->cwsr_isa = cwsr_trap_gfx8_hex;
kfd->cwsr_isa_size = sizeof(cwsr_trap_gfx8_hex);
} else {
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);
}
kfd->cwsr_enabled = true;
}
}
bool kgd2kfd_device_init(struct kfd_dev *kfd,
const struct kgd2kfd_shared_resources *gpu_resources)
{
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
unsigned int size;
kfd->mec_fw_version = amdgpu_amdkfd_get_fw_version(kfd->kgd,
KGD_ENGINE_MEC1);
kfd->sdma_fw_version = amdgpu_amdkfd_get_fw_version(kfd->kgd,
KGD_ENGINE_SDMA1);
kfd->shared_resources = *gpu_resources;
kfd->vm_info.first_vmid_kfd = ffs(gpu_resources->compute_vmid_bitmap)-1;
kfd->vm_info.last_vmid_kfd = fls(gpu_resources->compute_vmid_bitmap)-1;
kfd->vm_info.vmid_num_kfd = kfd->vm_info.last_vmid_kfd
- kfd->vm_info.first_vmid_kfd + 1;
/* Verify module parameters regarding mapped process number*/
if ((hws_max_conc_proc < 0)
|| (hws_max_conc_proc > kfd->vm_info.vmid_num_kfd)) {
dev_err(kfd_device,
"hws_max_conc_proc %d must be between 0 and %d, use %d instead\n",
hws_max_conc_proc, kfd->vm_info.vmid_num_kfd,
kfd->vm_info.vmid_num_kfd);
kfd->max_proc_per_quantum = kfd->vm_info.vmid_num_kfd;
} else
kfd->max_proc_per_quantum = hws_max_conc_proc;
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
/* calculate max size of mqds needed for queues */
drm/amdkfd: Allow user to limit only queues per device This patch replaces the two current amdkfd module parameters with a new one. The current parameters that are being replaced are: - Maximum number of HSA processes - Maximum number of queues per process The new parameter that replaces them is called "Maximum queues per device" This replacement achieves two goals: - Allows the user to have as many HSA processes as it wants (until a maximum of 512 HSA processes in Kaveri). - Removes the limitation the user had on maximum number of queues per HSA process. E.g. the user can now have processes which only have one queue and other processes which have hundreds of queues, while before the user couldn't have more than 128 queues per process (as default). The default value of the new parameter is 4096 (32 * 128, which were the defaults of the old parameters). There is almost no additional GART memory required for the default case. As a reminder, this amount of queues requires a little bit below 4MB of GART memory. v2: In addition, This patch defines a new counter for queues accounting in the DQM structure. This is done because the current counter only counts active queues which allows the user to create more queues than the max_num_of_queues_per_device module parameter allows. However, we need the current counter for the runlist packet build process, so the solution is to have a dedicated counter for this accounting. Signed-off-by: Oded Gabbay <oded.gabbay@amd.com> Reviewed-by: Ben Goz <ben.goz@amd.com>
2015-01-18 18:18:01 +07:00
size = max_num_of_queues_per_device *
kfd->device_info->mqd_size_aligned;
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
/*
* calculate max size of runlist packet.
* There can be only 2 packets at once
*/
size += (KFD_MAX_NUM_OF_PROCESSES * sizeof(struct pm4_mes_map_process) +
max_num_of_queues_per_device * sizeof(struct pm4_mes_map_queues)
+ sizeof(struct pm4_mes_runlist)) * 2;
/* Add size of HIQ & DIQ */
size += KFD_KERNEL_QUEUE_SIZE * 2;
/* add another 512KB for all other allocations on gart (HPD, fences) */
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
size += 512 * 1024;
if (amdgpu_amdkfd_alloc_gtt_mem(
kfd->kgd, size, &kfd->gtt_mem,
&kfd->gtt_start_gpu_addr, &kfd->gtt_start_cpu_ptr,
false)) {
dev_err(kfd_device, "Could not allocate %d bytes\n", size);
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
goto out;
}
dev_info(kfd_device, "Allocated %d bytes on gart\n", size);
/* Initialize GTT sa with 512 byte chunk size */
if (kfd_gtt_sa_init(kfd, size, 512) != 0) {
dev_err(kfd_device, "Error initializing gtt sub-allocator\n");
goto kfd_gtt_sa_init_error;
}
if (kfd_doorbell_init(kfd)) {
dev_err(kfd_device,
"Error initializing doorbell aperture\n");
goto kfd_doorbell_error;
}
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
if (kfd->kfd2kgd->get_hive_id)
kfd->hive_id = kfd->kfd2kgd->get_hive_id(kfd->kgd);
if (kfd_topology_add_device(kfd)) {
dev_err(kfd_device, "Error adding device to topology\n");
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
goto kfd_topology_add_device_error;
}
if (kfd_interrupt_init(kfd)) {
dev_err(kfd_device, "Error initializing interrupts\n");
goto kfd_interrupt_error;
}
2014-07-17 05:27:00 +07:00
kfd->dqm = device_queue_manager_init(kfd);
if (!kfd->dqm) {
dev_err(kfd_device, "Error initializing queue manager\n");
2014-07-17 05:27:00 +07:00
goto device_queue_manager_error;
}
if (kfd_iommu_device_init(kfd)) {
dev_err(kfd_device, "Error initializing iommuv2\n");
goto device_iommu_error;
2014-07-17 05:27:00 +07:00
}
kfd_cwsr_init(kfd);
if (kfd_resume(kfd))
goto kfd_resume_error;
kfd->dbgmgr = NULL;
kfd->init_complete = true;
dev_info(kfd_device, "added device %x:%x\n", kfd->pdev->vendor,
kfd->pdev->device);
pr_debug("Starting kfd with the following scheduling policy %d\n",
kfd->dqm->sched_policy);
2014-07-17 05:27:00 +07:00
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
goto out;
kfd_resume_error:
device_iommu_error:
2014-07-17 05:27:00 +07:00
device_queue_manager_uninit(kfd->dqm);
device_queue_manager_error:
kfd_interrupt_exit(kfd);
kfd_interrupt_error:
kfd_topology_remove_device(kfd);
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
kfd_topology_add_device_error:
kfd_doorbell_fini(kfd);
kfd_doorbell_error:
kfd_gtt_sa_fini(kfd);
kfd_gtt_sa_init_error:
amdgpu_amdkfd_free_gtt_mem(kfd->kgd, kfd->gtt_mem);
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
dev_err(kfd_device,
"device %x:%x NOT added due to errors\n",
amdkfd: Add basic modules to amdkfd This patch adds the process module and three helper modules: - kfd_process, which handles process which open /dev/kfd - kfd_doorbell, which provides helper functions for doorbell allocation, release and mapping to userspace - kfd_pasid, which provides helper functions for pasid allocation and release - kfd_aperture, which provides helper functions for managing the LDS, Local GPU memory and Scratch memory apertures of the process This patch only contains the basic kfd_process module, which doesn't contain the reference to the queue scheduler. This was done to allow easier code review. Also, this patch doesn't contain the calls to the IOMMU driver for binding the pasid to the device. Again, this was done to allow easier code review The kfd_process object is created when a process opens /dev/kfd and is closed when the mm_struct of that process is teared-down. v3: Removed kfd_vidmem.c file Replaced direct mmput call to mmu_notifier release Removed typedefs Moved bool field to end of the structure Added new kernel params for gart usage limitation Added initialization of sa manager Fixed debug messages Remove support for LDS in 32 bit Changed code to support mmap of doorbell pages from userspace Added documentation for apertures v4: Replaced RCU by SRCU for kfd_process list management v5: Move amdkfd from drm/radeon/ to drm/amd/ Rename kfd_aperture.c to kfd_flat_memory.c Protect against multiple init calls MQD size is H/W dependent so moved it to device info structure Rename kfd_mem_obj structure's members Use delayed function for process tear-down Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 03:25:31 +07:00
kfd->pdev->vendor, kfd->pdev->device);
out:
return kfd->init_complete;
}
void kgd2kfd_device_exit(struct kfd_dev *kfd)
{
if (kfd->init_complete) {
kgd2kfd_suspend(kfd);
2014-07-17 05:27:00 +07:00
device_queue_manager_uninit(kfd->dqm);
kfd_interrupt_exit(kfd);
kfd_topology_remove_device(kfd);
kfd_doorbell_fini(kfd);
kfd_gtt_sa_fini(kfd);
amdgpu_amdkfd_free_gtt_mem(kfd->kgd, kfd->gtt_mem);
}
amdkfd: Add topology module to amdkfd This patch adds the topology module to the driver. The topology is exposed to userspace through the sysfs. The calls to add and remove a device to/from topology are done by the radeon driver. v3: The CPU information, that is provided in the topology section of the amdkfd driver, is extracted from the CRAT table. Unlike the CPU information located in /sys/devices/system/cpu/cpu*, which is extracted from the SRAT table. While the CPU information provided by the CRAT and the SRAT tables might be identical, the node topology might be different. The SRAT table contains the topology of CPU nodes only. The CRAT table contains the topology of CPU and GPU nodes together (and can be interleaved). For example CPU node 1 in SRAT can be CPU node 3 in CRAT. Furthermore it's worth to mention that the CRAT table contains only HSA compatible nodes (nodes which are compliant with the HSA spec). To recap, amdkfd exposes a different kind of topology than the one exposed by /sys/devices/system/cpu/cpu even though it may contain similar information. v4: The topology module doesn't support uevent handling and doesn't notify the userspace about runtime modifications. It is up to the userspace to acquire snapshots of the topology information created by the amdkfd and exposed in sysfs. The following is an example of how the topology looks on a Kaveri A10-7850K system with amdkfd installed: /sys/devices/virtual/kfd/kfd/ | --- topology/ | |--- generation_id |--- system_properties |--- nodes/ | |--- 0/ | |--- gpu_id |--- name |--- properties |--- caches/ | |--- 0/ | |--- properties |--- 1/ | |--- properties |--- 2/ | |--- properties |--- io_links/ | |--- mem_banks/ | |--- 0/ | |--- properties |--- 1/ | |--- properties |--- 2/ | |--- properties |--- 3/ | |--- properties v5: Move amdkfd from drm/radeon/ to drm/amd/ Add a check if dev->gpu pointer is null before accessing it in the node_show function in kfd_topology.c This situation may occur when amdkfd is loaded and there is a GPU with a CRAT table, but that GPU isn't supported by amdkfd Signed-off-by: Evgeny Pinchuk <evgeny.pinchuk@amd.com> Signed-off-by: Oded Gabbay <oded.gabbay@amd.com>
2014-07-17 01:22:32 +07:00
kfree(kfd);
}
int kgd2kfd_pre_reset(struct kfd_dev *kfd)
{
if (!kfd->init_complete)
return 0;
kgd2kfd_suspend(kfd);
/* hold dqm->lock to prevent further execution*/
dqm_lock(kfd->dqm);
kfd_signal_reset_event(kfd);
return 0;
}
/*
* Fix me. KFD won't be able to resume existing process for now.
* We will keep all existing process in a evicted state and
* wait the process to be terminated.
*/
int kgd2kfd_post_reset(struct kfd_dev *kfd)
{
int ret, count;
if (!kfd->init_complete)
return 0;
dqm_unlock(kfd->dqm);
ret = kfd_resume(kfd);
if (ret)
return ret;
count = atomic_dec_return(&kfd_locked);
WARN_ONCE(count != 0, "KFD reset ref. error");
atomic_set(&kfd->sram_ecc_flag, 0);
return 0;
}
bool kfd_is_locked(void)
{
return (atomic_read(&kfd_locked) > 0);
}
void kgd2kfd_suspend(struct kfd_dev *kfd)
{
if (!kfd->init_complete)
return;
/* For first KFD device suspend all the KFD processes */
if (atomic_inc_return(&kfd_locked) == 1)
kfd_suspend_all_processes();
kfd->dqm->ops.stop(kfd->dqm);
kfd_iommu_suspend(kfd);
}
int kgd2kfd_resume(struct kfd_dev *kfd)
{
int ret, count;
if (!kfd->init_complete)
return 0;
ret = kfd_resume(kfd);
if (ret)
return ret;
count = atomic_dec_return(&kfd_locked);
WARN_ONCE(count < 0, "KFD suspend / resume ref. error");
if (count == 0)
ret = kfd_resume_all_processes();
return ret;
}
static int kfd_resume(struct kfd_dev *kfd)
{
int err = 0;
err = kfd_iommu_resume(kfd);
if (err) {
dev_err(kfd_device,
"Failed to resume IOMMU for device %x:%x\n",
kfd->pdev->vendor, kfd->pdev->device);
return err;
}
err = kfd->dqm->ops.start(kfd->dqm);
if (err) {
dev_err(kfd_device,
"Error starting queue manager for device %x:%x\n",
kfd->pdev->vendor, kfd->pdev->device);
goto dqm_start_error;
}
return err;
dqm_start_error:
kfd_iommu_suspend(kfd);
return err;
}
/* This is called directly from KGD at ISR. */
void kgd2kfd_interrupt(struct kfd_dev *kfd, const void *ih_ring_entry)
{
uint32_t patched_ihre[KFD_MAX_RING_ENTRY_SIZE];
bool is_patched = false;
drm/amdkfd: fix interrupt spin lock Vega10 has multiple interrupt rings, so this can be called from multiple calles at the same time resulting in: [ 71.779334] ================================ [ 71.779406] WARNING: inconsistent lock state [ 71.779478] 4.19.0-rc1+ #44 Tainted: G W [ 71.779565] -------------------------------- [ 71.779637] inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage. [ 71.779740] kworker/6:1/120 [HC0[0]:SC0[0]:HE1:SE1] takes: [ 71.779832] 00000000ad761971 (&(&kfd->interrupt_lock)->rlock){?...}, at: kgd2kfd_interrupt+0x75/0x100 [amdgpu] [ 71.780058] {IN-HARDIRQ-W} state was registered at: [ 71.780115] _raw_spin_lock+0x2c/0x40 [ 71.780180] kgd2kfd_interrupt+0x75/0x100 [amdgpu] [ 71.780248] amdgpu_irq_callback+0x6c/0x150 [amdgpu] [ 71.780315] amdgpu_ih_process+0x88/0x100 [amdgpu] [ 71.780380] amdgpu_irq_handler+0x20/0x40 [amdgpu] [ 71.780409] __handle_irq_event_percpu+0x49/0x2a0 [ 71.780436] handle_irq_event_percpu+0x30/0x70 [ 71.780461] handle_irq_event+0x37/0x60 [ 71.780484] handle_edge_irq+0x83/0x1b0 [ 71.780506] handle_irq+0x1f/0x30 [ 71.780526] do_IRQ+0x53/0x110 [ 71.780544] ret_from_intr+0x0/0x22 [ 71.780566] cpuidle_enter_state+0xaa/0x330 [ 71.780591] do_idle+0x203/0x280 [ 71.780610] cpu_startup_entry+0x6f/0x80 [ 71.780634] start_secondary+0x1b0/0x200 [ 71.780657] secondary_startup_64+0xa4/0xb0 Fix this by always using irq save spin locks. Signed-off-by: Christian König <christian.koenig@amd.com> Acked-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2018-11-02 20:46:24 +07:00
unsigned long flags;
if (!kfd->init_complete)
return;
if (kfd->device_info->ih_ring_entry_size > sizeof(patched_ihre)) {
dev_err_once(kfd_device, "Ring entry too small\n");
return;
}
drm/amdkfd: fix interrupt spin lock Vega10 has multiple interrupt rings, so this can be called from multiple calles at the same time resulting in: [ 71.779334] ================================ [ 71.779406] WARNING: inconsistent lock state [ 71.779478] 4.19.0-rc1+ #44 Tainted: G W [ 71.779565] -------------------------------- [ 71.779637] inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage. [ 71.779740] kworker/6:1/120 [HC0[0]:SC0[0]:HE1:SE1] takes: [ 71.779832] 00000000ad761971 (&(&kfd->interrupt_lock)->rlock){?...}, at: kgd2kfd_interrupt+0x75/0x100 [amdgpu] [ 71.780058] {IN-HARDIRQ-W} state was registered at: [ 71.780115] _raw_spin_lock+0x2c/0x40 [ 71.780180] kgd2kfd_interrupt+0x75/0x100 [amdgpu] [ 71.780248] amdgpu_irq_callback+0x6c/0x150 [amdgpu] [ 71.780315] amdgpu_ih_process+0x88/0x100 [amdgpu] [ 71.780380] amdgpu_irq_handler+0x20/0x40 [amdgpu] [ 71.780409] __handle_irq_event_percpu+0x49/0x2a0 [ 71.780436] handle_irq_event_percpu+0x30/0x70 [ 71.780461] handle_irq_event+0x37/0x60 [ 71.780484] handle_edge_irq+0x83/0x1b0 [ 71.780506] handle_irq+0x1f/0x30 [ 71.780526] do_IRQ+0x53/0x110 [ 71.780544] ret_from_intr+0x0/0x22 [ 71.780566] cpuidle_enter_state+0xaa/0x330 [ 71.780591] do_idle+0x203/0x280 [ 71.780610] cpu_startup_entry+0x6f/0x80 [ 71.780634] start_secondary+0x1b0/0x200 [ 71.780657] secondary_startup_64+0xa4/0xb0 Fix this by always using irq save spin locks. Signed-off-by: Christian König <christian.koenig@amd.com> Acked-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2018-11-02 20:46:24 +07:00
spin_lock_irqsave(&kfd->interrupt_lock, flags);
if (kfd->interrupts_active
&& interrupt_is_wanted(kfd, ih_ring_entry,
patched_ihre, &is_patched)
&& enqueue_ih_ring_entry(kfd,
is_patched ? patched_ihre : ih_ring_entry))
queue_work(kfd->ih_wq, &kfd->interrupt_work);
drm/amdkfd: fix interrupt spin lock Vega10 has multiple interrupt rings, so this can be called from multiple calles at the same time resulting in: [ 71.779334] ================================ [ 71.779406] WARNING: inconsistent lock state [ 71.779478] 4.19.0-rc1+ #44 Tainted: G W [ 71.779565] -------------------------------- [ 71.779637] inconsistent {IN-HARDIRQ-W} -> {HARDIRQ-ON-W} usage. [ 71.779740] kworker/6:1/120 [HC0[0]:SC0[0]:HE1:SE1] takes: [ 71.779832] 00000000ad761971 (&(&kfd->interrupt_lock)->rlock){?...}, at: kgd2kfd_interrupt+0x75/0x100 [amdgpu] [ 71.780058] {IN-HARDIRQ-W} state was registered at: [ 71.780115] _raw_spin_lock+0x2c/0x40 [ 71.780180] kgd2kfd_interrupt+0x75/0x100 [amdgpu] [ 71.780248] amdgpu_irq_callback+0x6c/0x150 [amdgpu] [ 71.780315] amdgpu_ih_process+0x88/0x100 [amdgpu] [ 71.780380] amdgpu_irq_handler+0x20/0x40 [amdgpu] [ 71.780409] __handle_irq_event_percpu+0x49/0x2a0 [ 71.780436] handle_irq_event_percpu+0x30/0x70 [ 71.780461] handle_irq_event+0x37/0x60 [ 71.780484] handle_edge_irq+0x83/0x1b0 [ 71.780506] handle_irq+0x1f/0x30 [ 71.780526] do_IRQ+0x53/0x110 [ 71.780544] ret_from_intr+0x0/0x22 [ 71.780566] cpuidle_enter_state+0xaa/0x330 [ 71.780591] do_idle+0x203/0x280 [ 71.780610] cpu_startup_entry+0x6f/0x80 [ 71.780634] start_secondary+0x1b0/0x200 [ 71.780657] secondary_startup_64+0xa4/0xb0 Fix this by always using irq save spin locks. Signed-off-by: Christian König <christian.koenig@amd.com> Acked-by: Alex Deucher <alexander.deucher@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
2018-11-02 20:46:24 +07:00
spin_unlock_irqrestore(&kfd->interrupt_lock, flags);
}
int kgd2kfd_quiesce_mm(struct mm_struct *mm)
{
struct kfd_process *p;
int r;
/* Because we are called from arbitrary context (workqueue) as opposed
* to process context, kfd_process could attempt to exit while we are
* running so the lookup function increments the process ref count.
*/
p = kfd_lookup_process_by_mm(mm);
if (!p)
return -ESRCH;
r = kfd_process_evict_queues(p);
kfd_unref_process(p);
return r;
}
int kgd2kfd_resume_mm(struct mm_struct *mm)
{
struct kfd_process *p;
int r;
/* Because we are called from arbitrary context (workqueue) as opposed
* to process context, kfd_process could attempt to exit while we are
* running so the lookup function increments the process ref count.
*/
p = kfd_lookup_process_by_mm(mm);
if (!p)
return -ESRCH;
r = kfd_process_restore_queues(p);
kfd_unref_process(p);
return r;
}
/** kgd2kfd_schedule_evict_and_restore_process - Schedules work queue that will
* prepare for safe eviction of KFD BOs that belong to the specified
* process.
*
* @mm: mm_struct that identifies the specified KFD process
* @fence: eviction fence attached to KFD process BOs
*
*/
int kgd2kfd_schedule_evict_and_restore_process(struct mm_struct *mm,
struct dma_fence *fence)
{
struct kfd_process *p;
unsigned long active_time;
unsigned long delay_jiffies = msecs_to_jiffies(PROCESS_ACTIVE_TIME_MS);
if (!fence)
return -EINVAL;
if (dma_fence_is_signaled(fence))
return 0;
p = kfd_lookup_process_by_mm(mm);
if (!p)
return -ENODEV;
if (fence->seqno == p->last_eviction_seqno)
goto out;
p->last_eviction_seqno = fence->seqno;
/* Avoid KFD process starvation. Wait for at least
* PROCESS_ACTIVE_TIME_MS before evicting the process again
*/
active_time = get_jiffies_64() - p->last_restore_timestamp;
if (delay_jiffies > active_time)
delay_jiffies -= active_time;
else
delay_jiffies = 0;
/* During process initialization eviction_work.dwork is initialized
* to kfd_evict_bo_worker
*/
schedule_delayed_work(&p->eviction_work, delay_jiffies);
out:
kfd_unref_process(p);
return 0;
}
static int kfd_gtt_sa_init(struct kfd_dev *kfd, unsigned int buf_size,
unsigned int chunk_size)
{
unsigned int num_of_longs;
if (WARN_ON(buf_size < chunk_size))
return -EINVAL;
if (WARN_ON(buf_size == 0))
return -EINVAL;
if (WARN_ON(chunk_size == 0))
return -EINVAL;
kfd->gtt_sa_chunk_size = chunk_size;
kfd->gtt_sa_num_of_chunks = buf_size / chunk_size;
num_of_longs = (kfd->gtt_sa_num_of_chunks + BITS_PER_LONG - 1) /
BITS_PER_LONG;
kfd->gtt_sa_bitmap = kcalloc(num_of_longs, sizeof(long), GFP_KERNEL);
if (!kfd->gtt_sa_bitmap)
return -ENOMEM;
pr_debug("gtt_sa_num_of_chunks = %d, gtt_sa_bitmap = %p\n",
kfd->gtt_sa_num_of_chunks, kfd->gtt_sa_bitmap);
mutex_init(&kfd->gtt_sa_lock);
return 0;
}
static void kfd_gtt_sa_fini(struct kfd_dev *kfd)
{
mutex_destroy(&kfd->gtt_sa_lock);
kfree(kfd->gtt_sa_bitmap);
}
static inline uint64_t kfd_gtt_sa_calc_gpu_addr(uint64_t start_addr,
unsigned int bit_num,
unsigned int chunk_size)
{
return start_addr + bit_num * chunk_size;
}
static inline uint32_t *kfd_gtt_sa_calc_cpu_addr(void *start_addr,
unsigned int bit_num,
unsigned int chunk_size)
{
return (uint32_t *) ((uint64_t) start_addr + bit_num * chunk_size);
}
int kfd_gtt_sa_allocate(struct kfd_dev *kfd, unsigned int size,
struct kfd_mem_obj **mem_obj)
{
unsigned int found, start_search, cur_size;
if (size == 0)
return -EINVAL;
if (size > kfd->gtt_sa_num_of_chunks * kfd->gtt_sa_chunk_size)
return -ENOMEM;
*mem_obj = kzalloc(sizeof(struct kfd_mem_obj), GFP_KERNEL);
if (!(*mem_obj))
return -ENOMEM;
pr_debug("Allocated mem_obj = %p for size = %d\n", *mem_obj, size);
start_search = 0;
mutex_lock(&kfd->gtt_sa_lock);
kfd_gtt_restart_search:
/* Find the first chunk that is free */
found = find_next_zero_bit(kfd->gtt_sa_bitmap,
kfd->gtt_sa_num_of_chunks,
start_search);
pr_debug("Found = %d\n", found);
/* If there wasn't any free chunk, bail out */
if (found == kfd->gtt_sa_num_of_chunks)
goto kfd_gtt_no_free_chunk;
/* Update fields of mem_obj */
(*mem_obj)->range_start = found;
(*mem_obj)->range_end = found;
(*mem_obj)->gpu_addr = kfd_gtt_sa_calc_gpu_addr(
kfd->gtt_start_gpu_addr,
found,
kfd->gtt_sa_chunk_size);
(*mem_obj)->cpu_ptr = kfd_gtt_sa_calc_cpu_addr(
kfd->gtt_start_cpu_ptr,
found,
kfd->gtt_sa_chunk_size);
pr_debug("gpu_addr = %p, cpu_addr = %p\n",
(uint64_t *) (*mem_obj)->gpu_addr, (*mem_obj)->cpu_ptr);
/* If we need only one chunk, mark it as allocated and get out */
if (size <= kfd->gtt_sa_chunk_size) {
pr_debug("Single bit\n");
set_bit(found, kfd->gtt_sa_bitmap);
goto kfd_gtt_out;
}
/* Otherwise, try to see if we have enough contiguous chunks */
cur_size = size - kfd->gtt_sa_chunk_size;
do {
(*mem_obj)->range_end =
find_next_zero_bit(kfd->gtt_sa_bitmap,
kfd->gtt_sa_num_of_chunks, ++found);
/*
* If next free chunk is not contiguous than we need to
* restart our search from the last free chunk we found (which
* wasn't contiguous to the previous ones
*/
if ((*mem_obj)->range_end != found) {
start_search = found;
goto kfd_gtt_restart_search;
}
/*
* If we reached end of buffer, bail out with error
*/
if (found == kfd->gtt_sa_num_of_chunks)
goto kfd_gtt_no_free_chunk;
/* Check if we don't need another chunk */
if (cur_size <= kfd->gtt_sa_chunk_size)
cur_size = 0;
else
cur_size -= kfd->gtt_sa_chunk_size;
} while (cur_size > 0);
pr_debug("range_start = %d, range_end = %d\n",
(*mem_obj)->range_start, (*mem_obj)->range_end);
/* Mark the chunks as allocated */
for (found = (*mem_obj)->range_start;
found <= (*mem_obj)->range_end;
found++)
set_bit(found, kfd->gtt_sa_bitmap);
kfd_gtt_out:
mutex_unlock(&kfd->gtt_sa_lock);
return 0;
kfd_gtt_no_free_chunk:
pr_debug("Allocation failed with mem_obj = %p\n", mem_obj);
mutex_unlock(&kfd->gtt_sa_lock);
kfree(mem_obj);
return -ENOMEM;
}
int kfd_gtt_sa_free(struct kfd_dev *kfd, struct kfd_mem_obj *mem_obj)
{
unsigned int bit;
/* Act like kfree when trying to free a NULL object */
if (!mem_obj)
return 0;
pr_debug("Free mem_obj = %p, range_start = %d, range_end = %d\n",
mem_obj, mem_obj->range_start, mem_obj->range_end);
mutex_lock(&kfd->gtt_sa_lock);
/* Mark the chunks as free */
for (bit = mem_obj->range_start;
bit <= mem_obj->range_end;
bit++)
clear_bit(bit, kfd->gtt_sa_bitmap);
mutex_unlock(&kfd->gtt_sa_lock);
kfree(mem_obj);
return 0;
}
void kgd2kfd_set_sram_ecc_flag(struct kfd_dev *kfd)
{
if (kfd)
atomic_inc(&kfd->sram_ecc_flag);
}
#if defined(CONFIG_DEBUG_FS)
/* This function will send a package to HIQ to hang the HWS
* which will trigger a GPU reset and bring the HWS back to normal state
*/
int kfd_debugfs_hang_hws(struct kfd_dev *dev)
{
int r = 0;
if (dev->dqm->sched_policy != KFD_SCHED_POLICY_HWS) {
pr_err("HWS is not enabled");
return -EINVAL;
}
r = pm_debugfs_hang_hws(&dev->dqm->packets);
if (!r)
r = dqm_debugfs_execute_queues(dev->dqm);
return r;
}
#endif