[PATCH 03/44] drm/amdkfd: add svm ioctl API
Felix Kuehling
Felix.Kuehling at amd.com
Mon Mar 22 10:58:19 UTC 2021
From: Philip Yang <Philip.Yang at amd.com>
Add svm (shared virtual memory) ioctl data structure and API definition.
The svm ioctl API is designed to be extensible in the future. All
operations are provided by a single IOCTL to preserve ioctl number
space. The arguments structure ends with a variable size array of
attributes that can be used to set or get one or multiple attributes.
Signed-off-by: Philip Yang <Philip.Yang at amd.com>
Signed-off-by: Alex Sierra <alex.sierra at amd.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling at amd.com>
---
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 12 ++
drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c | 4 +
drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 2 +
drivers/gpu/drm/amd/amdkfd/kfd_process.c | 1 +
include/uapi/linux/kfd_ioctl.h | 130 ++++++++++++++++++-
5 files changed, 147 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
index 43de260b2230..dbc824cc6b32 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_chardev.c
@@ -1742,6 +1742,16 @@ static int kfd_ioctl_smi_events(struct file *filep,
return kfd_smi_event_open(dev, &args->anon_fd);
}
+static int kfd_ioctl_svm(struct file *filep, struct kfd_process *p, void *data)
+{
+ int r = 0;
+
+ if (p->svm_disabled)
+ return -EPERM;
+
+ return r;
+}
+
#define AMDKFD_IOCTL_DEF(ioctl, _func, _flags) \
[_IOC_NR(ioctl)] = {.cmd = ioctl, .func = _func, .flags = _flags, \
.cmd_drv = 0, .name = #ioctl}
@@ -1840,6 +1850,8 @@ static const struct amdkfd_ioctl_desc amdkfd_ioctls[] = {
AMDKFD_IOCTL_DEF(AMDKFD_IOC_SMI_EVENTS,
kfd_ioctl_smi_events, 0),
+
+ AMDKFD_IOCTL_DEF(AMDKFD_IOC_SVM, kfd_ioctl_svm, 0),
};
#define AMDKFD_CORE_IOCTL_COUNT ARRAY_SIZE(amdkfd_ioctls)
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
index a2c9063076cc..52da1a3b2c7a 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_flat_memory.c
@@ -405,6 +405,10 @@ int kfd_init_apertures(struct kfd_process *process)
case CHIP_POLARIS12:
case CHIP_VEGAM:
kfd_init_apertures_vi(pdd, id);
+ /* VI GPUs cannot support SVM with only
+ * 40 bits of virtual address space.
+ */
+ process->svm_disabled |= true;
break;
case CHIP_VEGA10:
case CHIP_VEGA12:
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
index 99b4624ef4c7..18fc2ccd1a77 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
@@ -809,6 +809,8 @@ struct kfd_process {
struct kobject *kobj;
struct kobject *kobj_queues;
struct attribute attr_pasid;
+
+ bool svm_disabled;
};
#define KFD_PROCESS_TABLE_SIZE 5 /* bits: 32 entries */
diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
index 028ebb0deddd..89e7c125d334 100644
--- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
+++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
@@ -1208,6 +1208,7 @@ static struct kfd_process *create_process(const struct task_struct *thread)
process->mm = thread->mm;
process->lead_thread = thread->group_leader;
process->n_pdds = 0;
+ process->svm_disabled = false;
INIT_DELAYED_WORK(&process->eviction_work, evict_process_worker);
INIT_DELAYED_WORK(&process->restore_work, restore_process_worker);
process->last_restore_timestamp = get_jiffies_64();
diff --git a/include/uapi/linux/kfd_ioctl.h b/include/uapi/linux/kfd_ioctl.h
index bf5e7d7846dd..247b57baa94f 100644
--- a/include/uapi/linux/kfd_ioctl.h
+++ b/include/uapi/linux/kfd_ioctl.h
@@ -30,9 +30,10 @@
* - 1.1 - initial version
* - 1.3 - Add SMI events support
* - 1.4 - Indicate new SRAM EDC bit in device properties
+ * - 1.5 - Add SVM API
*/
#define KFD_IOCTL_MAJOR_VERSION 1
-#define KFD_IOCTL_MINOR_VERSION 4
+#define KFD_IOCTL_MINOR_VERSION 5
struct kfd_ioctl_get_version_args {
__u32 major_version; /* from KFD */
@@ -473,6 +474,129 @@ enum kfd_mmio_remap {
KFD_MMIO_REMAP_HDP_REG_FLUSH_CNTL = 4,
};
+/* Guarantee host access to memory */
+#define KFD_IOCTL_SVM_FLAG_HOST_ACCESS 0x00000001
+/* Fine grained coherency between all devices with access */
+#define KFD_IOCTL_SVM_FLAG_COHERENT 0x00000002
+/* Use any GPU in same hive as preferred device */
+#define KFD_IOCTL_SVM_FLAG_HIVE_LOCAL 0x00000004
+/* GPUs only read, allows replication */
+#define KFD_IOCTL_SVM_FLAG_GPU_RO 0x00000008
+/* Allow execution on GPU */
+#define KFD_IOCTL_SVM_FLAG_GPU_EXEC 0x00000010
+/* GPUs mostly read, may allow similar optimizations as RO, but writes fault */
+#define KFD_IOCTL_SVM_FLAG_GPU_READ_MOSTLY 0x00000020
+
+/**
+ * kfd_ioctl_svm_op - SVM ioctl operations
+ *
+ * @KFD_IOCTL_SVM_OP_SET_ATTR: Modify one or more attributes
+ * @KFD_IOCTL_SVM_OP_GET_ATTR: Query one or more attributes
+ */
+enum kfd_ioctl_svm_op {
+ KFD_IOCTL_SVM_OP_SET_ATTR,
+ KFD_IOCTL_SVM_OP_GET_ATTR
+};
+
+/** kfd_ioctl_svm_location - Enum for preferred and prefetch locations
+ *
+ * GPU IDs are used to specify GPUs as preferred and prefetch locations.
+ * Below definitions are used for system memory or for leaving the preferred
+ * location unspecified.
+ */
+enum kfd_ioctl_svm_location {
+ KFD_IOCTL_SVM_LOCATION_SYSMEM = 0,
+ KFD_IOCTL_SVM_LOCATION_UNDEFINED = 0xffffffff
+};
+
+/**
+ * kfd_ioctl_svm_attr_type - SVM attribute types
+ *
+ * @KFD_IOCTL_SVM_ATTR_PREFERRED_LOC: gpuid of the preferred location, 0 for
+ * system memory
+ * @KFD_IOCTL_SVM_ATTR_PREFETCH_LOC: gpuid of the prefetch location, 0 for
+ * system memory. Setting this triggers an
+ * immediate prefetch (migration).
+ * @KFD_IOCTL_SVM_ATTR_ACCESS:
+ * @KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE:
+ * @KFD_IOCTL_SVM_ATTR_NO_ACCESS: specify memory access for the gpuid given
+ * by the attribute value
+ * @KFD_IOCTL_SVM_ATTR_SET_FLAGS: bitmask of flags to set (see
+ * KFD_IOCTL_SVM_FLAG_...)
+ * @KFD_IOCTL_SVM_ATTR_CLR_FLAGS: bitmask of flags to clear
+ * @KFD_IOCTL_SVM_ATTR_GRANULARITY: migration granularity
+ * (log2 num pages)
+ */
+enum kfd_ioctl_svm_attr_type {
+ KFD_IOCTL_SVM_ATTR_PREFERRED_LOC,
+ KFD_IOCTL_SVM_ATTR_PREFETCH_LOC,
+ KFD_IOCTL_SVM_ATTR_ACCESS,
+ KFD_IOCTL_SVM_ATTR_ACCESS_IN_PLACE,
+ KFD_IOCTL_SVM_ATTR_NO_ACCESS,
+ KFD_IOCTL_SVM_ATTR_SET_FLAGS,
+ KFD_IOCTL_SVM_ATTR_CLR_FLAGS,
+ KFD_IOCTL_SVM_ATTR_GRANULARITY
+};
+
+/**
+ * kfd_ioctl_svm_attribute - Attributes as pairs of type and value
+ *
+ * The meaning of the @value depends on the attribute type.
+ *
+ * @type: attribute type (see enum @kfd_ioctl_svm_attr_type)
+ * @value: attribute value
+ */
+struct kfd_ioctl_svm_attribute {
+ __u32 type;
+ __u32 value;
+};
+
+/**
+ * kfd_ioctl_svm_args - Arguments for SVM ioctl
+ *
+ * @op specifies the operation to perform (see enum
+ * @kfd_ioctl_svm_op). @start_addr and @size are common for all
+ * operations.
+ *
+ * A variable number of attributes can be given in @attrs.
+ * @nattr specifies the number of attributes. New attributes can be
+ * added in the future without breaking the ABI. If unknown attributes
+ * are given, the function returns -EINVAL.
+ *
+ * @KFD_IOCTL_SVM_OP_SET_ATTR sets attributes for a virtual address
+ * range. It may overlap existing virtual address ranges. If it does,
+ * the existing ranges will be split such that the attribute changes
+ * only apply to the specified address range.
+ *
+ * @KFD_IOCTL_SVM_OP_GET_ATTR returns the intersection of attributes
+ * over all memory in the given range and returns the result as the
+ * attribute value. If different pages have different preferred or
+ * prefetch locations, 0xffffffff will be returned for
+ * @KFD_IOCTL_SVM_ATTR_PREFERRED_LOC or
+ * @KFD_IOCTL_SVM_ATTR_PREFETCH_LOC resepctively. For
+ * @KFD_IOCTL_SVM_ATTR_SET_FLAGS, flags of all pages will be
+ * aggregated by bitwise AND. The minimum migration granularity
+ * throughout the range will be returned for
+ * @KFD_IOCTL_SVM_ATTR_GRANULARITY.
+ *
+ * Querying of accessibility attributes works by initializing the
+ * attribute type to @KFD_IOCTL_SVM_ATTR_ACCESS and the value to the
+ * GPUID being queried. Multiple attributes can be given to allow
+ * querying multiple GPUIDs. The ioctl function overwrites the
+ * attribute type to indicate the access for the specified GPU.
+ *
+ * @KFD_IOCTL_SVM_ATTR_CLR_FLAGS is invalid for
+ * @KFD_IOCTL_SVM_OP_GET_ATTR.
+ */
+struct kfd_ioctl_svm_args {
+ __u64 start_addr;
+ __u64 size;
+ __u32 op;
+ __u32 nattr;
+ /* Variable length array of attributes */
+ struct kfd_ioctl_svm_attribute attrs[0];
+};
+
#define AMDKFD_IOCTL_BASE 'K'
#define AMDKFD_IO(nr) _IO(AMDKFD_IOCTL_BASE, nr)
#define AMDKFD_IOR(nr, type) _IOR(AMDKFD_IOCTL_BASE, nr, type)
@@ -573,7 +697,9 @@ enum kfd_mmio_remap {
#define AMDKFD_IOC_SMI_EVENTS \
AMDKFD_IOWR(0x1F, struct kfd_ioctl_smi_events_args)
+#define AMDKFD_IOC_SVM AMDKFD_IOWR(0x20, struct kfd_ioctl_svm_args)
+
#define AMDKFD_COMMAND_START 0x01
-#define AMDKFD_COMMAND_END 0x20
+#define AMDKFD_COMMAND_END 0x21
#endif
--
2.31.0
More information about the dri-devel
mailing list