On 20/07/14 19:54, Jerome Glisse wrote:
On Thu, Jul 17, 2014 at 04:29:15PM +0300, Oded Gabbay wrote:
KFD_IOC_GET_VERSION: Retrieves the interface version of amdkfd
KFD_IOC_CREATE_QUEUE: Creates a usermode queue that runs on a specific GPU device
KFD_IOC_DESTROY_QUEUE: Destroys an existing usermode queue
KFD_IOC_SET_MEMORY_POLICY: Sets the memory policy of the default and alternate aperture of the calling process
KFD_IOC_GET_CLOCK_COUNTERS: Retrieves counters (timestamps) of CPU and GPU
KFD_IOC_GET_PROCESS_APERTURES: Retrieves information about process apertures that were initialized
during the open() call of the amdkfd device
KFD_IOC_UPDATE_QUEUE: Updates configuration of an existing usermode queue
KFD_IOC_PMC_ACQUIRE_ACCESS: Acquires exclusive access (from other HSA processes) to the performance
counters of the GPU
- KFD_IOC_PMC_RELEASE_ACCESS: Releases exclusive access of the GPU's performance counters
Exclusive userspace access is recipie for failure. This must go and you must come up with better model. Which in my mind involve an ioctl for each command buffer submission.
Removed these 2 ioctls in v3
Signed-off-by: Oded Gabbay oded.gabbay@amd.com
include/uapi/linux/kfd_ioctl.h | 133 +++++++++++++++++++++++++++++++++++++++++ 1 file changed, 133 insertions(+) create mode 100644 include/uapi/linux/kfd_ioctl.h
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h new file mode 100644 index 0000000..3cedd1a --- /dev/null +++ b/include/uapi/linux/kfd_ioctl.h @@ -0,0 +1,133 @@ +/*
- 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.
- */
+#ifndef KFD_IOCTL_H_INCLUDED +#define KFD_IOCTL_H_INCLUDED
+#include <linux/types.h> +#include <linux/ioctl.h>
+#define KFD_IOCTL_CURRENT_VERSION 1
+/* The 64-bit ABI is the authoritative version. */ +#pragma pack(push, 1)
pagram pack must be remove do not use that.
Removed in v3
+struct kfd_ioctl_get_version_args {
- uint32_t min_supported_version; /* from KFD */
- uint32_t max_supported_version; /* from KFD */
+};
+/* For kfd_ioctl_create_queue_args.queue_type. */ +#define KFD_IOC_QUEUE_TYPE_COMPUTE 0 +#define KFD_IOC_QUEUE_TYPE_SDMA 1
+struct kfd_ioctl_create_queue_args {
- uint64_t ring_base_address; /* to KFD */
- uint64_t write_pointer_address; /* from KFD */
- uint64_t read_pointer_address; /* from KFD */
- uint64_t doorbell_address; /* from KFD */
- uint32_t ring_size; /* to KFD */
- uint32_t gpu_id; /* to KFD */
- uint32_t queue_type; /* to KFD */
- uint32_t queue_percentage; /* to KFD */
- uint32_t queue_priority; /* to KFD */
- uint32_t queue_id; /* from KFD */
+};
+struct kfd_ioctl_destroy_queue_args {
- uint32_t queue_id; /* to KFD */
+};
+struct kfd_ioctl_update_queue_args {
- uint64_t ring_base_address; /* to KFD */
- uint32_t queue_id; /* to KFD */
- uint32_t ring_size; /* to KFD */
- uint32_t queue_percentage; /* to KFD */
- uint32_t queue_priority; /* to KFD */
+};
The queue_percentage and queue_priority really needs some explanations. I guess userspace would shed some light on those but still this should be properly explain.
Added explanation in v3
+/* For kfd_ioctl_set_memory_policy_args.default_policy and alternate_policy */ +#define KFD_IOC_CACHE_POLICY_COHERENT 0 +#define KFD_IOC_CACHE_POLICY_NONCOHERENT 1
+struct kfd_ioctl_set_memory_policy_args {
- uint64_t alternate_aperture_base; /* to KFD */
- uint64_t alternate_aperture_size; /* to KFD */
- uint32_t gpu_id; /* to KFD */
- uint32_t default_policy; /* to KFD */
- uint32_t alternate_policy; /* to KFD */
+};
Same what is aperture in this context. I know about all this stuff but i have no idea what aperture is in this context.
Added explanation in v3
+struct kfd_ioctl_get_clock_counters_args {
- uint64_t gpu_clock_counter; /* from KFD */
- uint64_t cpu_clock_counter; /* from KFD */
- uint64_t system_clock_counter; /* from KFD */
- uint64_t system_clock_freq; /* from KFD */
- uint32_t gpu_id; /* to KFD */
+};
Again comment about what kind of counter this is, monotonic, ...
Added explanation in v3
+#define NUM_OF_SUPPORTED_GPUS 7
+struct kfd_process_device_apertures {
- uint64_t lds_base; /* from KFD */
- uint64_t lds_limit; /* from KFD */
- uint64_t scratch_base; /* from KFD */
- uint64_t scratch_limit; /* from KFD */
- uint64_t gpuvm_base; /* from KFD */
- uint64_t gpuvm_limit; /* from KFD */
- uint32_t gpu_id; /* from KFD */
+};
+struct kfd_ioctl_get_process_apertures_args {
- struct kfd_process_device_apertures process_apertures[NUM_OF_SUPPORTED_GPUS];/* from KFD */
- uint8_t num_of_nodes; /* from KFD, should be in the range [1 - NUM_OF_SUPPORTED_GPUS]*/
+};
I would rather see userspace provide a pointer to an array and the size of that array and possibly size of individual element. This would allow you to grow the kfd_process_device_apertures if needs be. Thought as i understand it this is a temporary driver.
This is not going to change in the near or far future. I also agree that we would probably see a common framework before NUM_OF_SUPPORTED_GPUS increases. If you insist, I'll try to squeeze it in at v4
+struct kfd_ioctl_pmc_acquire_access_args {
- uint64_t trace_id; /* to KFD */
- uint32_t gpu_id; /* to KFD */
+};
+struct kfd_ioctl_pmc_release_access_args {
- uint64_t trace_id; /* to KFD */
- uint32_t gpu_id; /* to KFD */
+};
As said above no ioctl to have some exclusive access you can not trust userspace that's rules NUMBER ONE.
Removed in v3.
+#define KFD_IOC_MAGIC 'K'
+#define KFD_IOC_GET_VERSION _IOR(KFD_IOC_MAGIC, 1, struct kfd_ioctl_get_version_args) +#define KFD_IOC_CREATE_QUEUE _IOWR(KFD_IOC_MAGIC, 2, struct kfd_ioctl_create_queue_args) +#define KFD_IOC_DESTROY_QUEUE _IOWR(KFD_IOC_MAGIC, 3, struct kfd_ioctl_destroy_queue_args) +#define KFD_IOC_SET_MEMORY_POLICY _IOW(KFD_IOC_MAGIC, 4, struct kfd_ioctl_set_memory_policy_args) +#define KFD_IOC_GET_CLOCK_COUNTERS _IOWR(KFD_IOC_MAGIC, 5, struct kfd_ioctl_get_clock_counters_args) +#define KFD_IOC_GET_PROCESS_APERTURES _IOR(KFD_IOC_MAGIC, 6, struct kfd_ioctl_get_process_apertures_args) +#define KFD_IOC_UPDATE_QUEUE _IOW(KFD_IOC_MAGIC, 7, struct kfd_ioctl_update_queue_args) +#define KFD_IOC_PMC_ACQUIRE_ACCESS _IOW(KFD_IOC_MAGIC, 12, struct kfd_ioctl_pmc_acquire_access_args) +#define KFD_IOC_PMC_RELEASE_ACCESS _IOW(KFD_IOC_MAGIC, 13, struct kfd_ioctl_pmc_release_access_args)
+#pragma pack(pop)
No pragma packing.
Removed in v3
+#endif
1.9.1
Oded