Skip to contentSkip to conversation

drm/amdkfd: Simplify the mmap offset related bit operations

PATCH In review Zhao, Yong 2 participants 3 messages Nov 5, 2019 – Nov 7, 2019 2 messages not in the archive

Subsystems (3)
lore
Share & tools
lore

Zhao, Yong

author
#0 lore raw

PATCH 1/1 drm/amdkfd: Simplify the mmap offset related bit operations

The new code uses straightforward bit shifts and thus has better readability.

Change-Id I0c1f7cca7e24ddb7b4ffe1cb0fa71943828ae373
---
 drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 10 ++++------
 drivers/gpu/drm/amd/amdkfd/kfd_events.c  |  1 -
 drivers/gpu/drm/amd/amdkfd/kfd_priv.h    |  9 +++------
 drivers/gpu/drm/amd/amdkfd/kfd_process.c |  3 +--
 4 files changed, 8 insertions(+), 15 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +4 -6
index b91993753b82..34078df36621 100644
@@ -298,7 +298,6 @@ static int kfd_ioctl_create_queue(struct file *filep, struct kfd_process *p,
 	/* Return gpu_id as doorbell offset for mmap usage */
 	args->doorbell_offset = KFD_MMAP_TYPE_DOORBELL;
 	args->doorbell_offset |= KFD_MMAP_GPU_ID(args->gpu_id);
-	args->doorbell_offset <<= PAGE_SHIFT;
 	if (KFD_IS_SOC15(dev->device_info->asic_family))
 		/* On SOC15 ASICs, include the doorbell offset within the
 		 * process doorbell frame, which could be 1 page or 2 pages.
@@ -1938,20 +1937,19 @@ static int kfd_mmap(struct file *filp, struct vm_area_struct *vma)
 {
 	struct kfd_process *process;
 	struct kfd_dev *dev = NULL;
-	unsigned long vm_pgoff;
+	unsigned long mmap_offset;
 	unsigned int gpu_id;
 
 	process = kfd_get_process(current);
 	if (IS_ERR(process))
 		return PTR_ERR(process);
 
-	vm_pgoff = vma->vm_pgoff;
-	vma->vm_pgoff = KFD_MMAP_OFFSET_VALUE_GET(vm_pgoff);
-	gpu_id = KFD_MMAP_GPU_ID_GET(vm_pgoff);
+	mmap_offset = vma->vm_pgoff << PAGE_SHIFT;
+	gpu_id = KFD_MMAP_GET_GPU_ID(mmap_offset);
 	if (gpu_id)
 		dev = kfd_device_by_id(gpu_id);
 
-	switch (vm_pgoff & KFD_MMAP_TYPE_MASK) {
+	switch (mmap_offset & KFD_MMAP_TYPE_MASK) {
 	case KFD_MMAP_TYPE_DOORBELL:
 		if (!dev)
 			return -ENODEV;
drivers/gpu/drm/amd/amdkfd/kfd_events.c +0 -1
index 908081c85de1..1f8365575b12 100644
@@ -346,7 +346,6 @@ int kfd_event_create(struct file *devkfd, struct kfd_process *p,
 		ret = create_signal_event(devkfd, p, ev);
 		if (!ret) {
 			*event_page_offset = KFD_MMAP_TYPE_EVENTS;
-			*event_page_offset <<= PAGE_SHIFT;
 			*event_slot_index = ev->event_id;
 		}
 		break;
drivers/gpu/drm/amd/amdkfd/kfd_priv.h +3 -6
index 66bae8f2dad1..8eecd2cd1fd2 100644
@@ -59,24 +59,21 @@
  * NOTE: struct vm_area_struct.vm_pgoff uses offset in pages. Hence, these
  *  defines are w.r.t to PAGE_SIZE
  */
-#define KFD_MMAP_TYPE_SHIFT	(62 - PAGE_SHIFT)
+#define KFD_MMAP_TYPE_SHIFT	(62)
 #define KFD_MMAP_TYPE_MASK	(0x3ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_DOORBELL	(0x3ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_EVENTS	(0x2ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_RESERVED_MEM	(0x1ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_MMIO	(0x0ULL << KFD_MMAP_TYPE_SHIFT)
 
-#define KFD_MMAP_GPU_ID_SHIFT (46 - PAGE_SHIFT)
+#define KFD_MMAP_GPU_ID_SHIFT (46)
 #define KFD_MMAP_GPU_ID_MASK (((1ULL << KFD_GPU_ID_HASH_WIDTH) - 1) \
 				<< KFD_MMAP_GPU_ID_SHIFT)
 #define KFD_MMAP_GPU_ID(gpu_id) ((((uint64_t)gpu_id) << KFD_MMAP_GPU_ID_SHIFT)\
 				& KFD_MMAP_GPU_ID_MASK)
-#define KFD_MMAP_GPU_ID_GET(offset)    ((offset & KFD_MMAP_GPU_ID_MASK) \
+#define KFD_MMAP_GET_GPU_ID(offset)    ((offset & KFD_MMAP_GPU_ID_MASK) \
 				>> KFD_MMAP_GPU_ID_SHIFT)
 
-#define KFD_MMAP_OFFSET_VALUE_MASK	(0x3FFFFFFFFFFFULL >> PAGE_SHIFT)
-#define KFD_MMAP_OFFSET_VALUE_GET(offset) (offset & KFD_MMAP_OFFSET_VALUE_MASK)
-
 /*
  * When working with cp scheduler we should assign the HIQ manually or via
  * the amdgpu driver to a fixed hqd slot, here are the fixed HIQ hqd slot
drivers/gpu/drm/amd/amdkfd/kfd_process.c +1 -2
index 6abfb77ae540..39dc49b8fd85 100644
@@ -554,8 +554,7 @@ static int kfd_process_init_cwsr_apu(struct kfd_process *p, struct file *filep)
 		if (!dev->cwsr_enabled || qpd->cwsr_kaddr || qpd->cwsr_base)
 			continue;
 
-		offset = (KFD_MMAP_TYPE_RESERVED_MEM | KFD_MMAP_GPU_ID(dev->id))
-			<< PAGE_SHIFT;
+		offset = KFD_MMAP_TYPE_RESERVED_MEM | KFD_MMAP_GPU_ID(dev->id);
 		qpd->tba_addr = (int64_t)vm_mmap(filep, 0,
 			KFD_CWSR_TBA_TMA_SIZE, PROT_READ | PROT_EXEC,
 			MAP_SHARED, offset);
Signature · 6 lines
2.17.1

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): [email protected]

#2 lore raw
Quoted 2 lines · Yong

The new code uses straightforward bit shifts and thus has better readability.

You're missing the MMAP-related code for mmio remapping. In kfd_ioctl_alloc_memory_of_gpu:

        /* MMIO is mapped through kfd device
         * Generate a kfd mmap offset
         */
        if (flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP) { args->mmap_offset = KFD_MMAP_TYPE_MMIO |
KFD_MMAP_GPU_ID(args->gpu_id);
                args->mmap_offset <<= PAGE_SHIFT; }

Regards,
  Felix

Quoted 100 lines
Change-Id I0c1f7cca7e24ddb7b4ffe1cb0fa71943828ae373

---
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 10 ++++------ drivers/gpu/drm/amd/amdkfd/kfd_events.c | 1 - drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 9 +++------ drivers/gpu/drm/amd/amdkfd/kfd_process.c | 3 +--

4 files changed, 8 insertions(+), 15 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +0 -1
index b91993753b82..34078df36621 100644
@@ -298,7 +298,6 @@ static int kfd_ioctl_create_queue(struct file 
 *filep, struct kfd_process *p,
 /* Return gpu_id as doorbell offset for mmap usage */
 args->doorbell_offset = KFD_MMAP_TYPE_DOORBELL;
 args->doorbell_offset |= KFD_MMAP_GPU_ID(args->gpu_id);
- args->doorbell_offset <<= PAGE_SHIFT;
 if (KFD_IS_SOC15(dev->device_info->asic_family))
 /* On SOC15 ASICs, include the doorbell offset within the

* process doorbell frame, which could be 1 page or 2 pages.

drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +4 -5
@@ -1938,20 +1937,19 @@ static int kfd_mmap(struct file *filp, struct 
 vm_area_struct *vma)
 {
 struct kfd_process *process;
 struct kfd_dev *dev = NULL;
- unsigned long vm_pgoff;
+ unsigned long mmap_offset;
 unsigned int gpu_id;
 process = kfd_get_process(current);
 if (IS_ERR(process))
 return PTR_ERR(process);
- vm_pgoff = vma->vm_pgoff;
- vma->vm_pgoff = KFD_MMAP_OFFSET_VALUE_GET(vm_pgoff);
- gpu_id = KFD_MMAP_GPU_ID_GET(vm_pgoff);
+ mmap_offset = vma->vm_pgoff << PAGE_SHIFT;
+ gpu_id = KFD_MMAP_GET_GPU_ID(mmap_offset);
 if (gpu_id)
 dev = kfd_device_by_id(gpu_id);
- switch (vm_pgoff & KFD_MMAP_TYPE_MASK) {
+ switch (mmap_offset & KFD_MMAP_TYPE_MASK) {

case KFD_MMAP_TYPE_DOORBELL:
if (!dev)
return -ENODEV;

drivers/gpu/drm/amd/amdkfd/kfd_events.c +0 -1
index 908081c85de1..1f8365575b12 100644
@@ -346,7 +346,6 @@ int kfd_event_create(struct file *devkfd, struct 
 kfd_process *p,
 ret = create_signal_event(devkfd, p, ev);
 if (!ret) {
 *event_page_offset = KFD_MMAP_TYPE_EVENTS;
- *event_page_offset <<= PAGE_SHIFT;
 *event_slot_index = ev->event_id;
 }

break;

drivers/gpu/drm/amd/amdkfd/kfd_priv.h +3 -6
index 66bae8f2dad1..8eecd2cd1fd2 100644
@@ -59,24 +59,21 @@
 * NOTE: struct vm_area_struct.vm_pgoff uses offset in pages. Hence, these
 * defines are w.r.t to PAGE_SIZE
 */
-#define KFD_MMAP_TYPE_SHIFT (62 - PAGE_SHIFT)
+#define KFD_MMAP_TYPE_SHIFT (62)
 #define KFD_MMAP_TYPE_MASK (0x3ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_DOORBELL (0x3ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_EVENTS (0x2ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_RESERVED_MEM (0x1ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_MMIO (0x0ULL << KFD_MMAP_TYPE_SHIFT)
-#define KFD_MMAP_GPU_ID_SHIFT (46 - PAGE_SHIFT)
+#define KFD_MMAP_GPU_ID_SHIFT (46)
 #define KFD_MMAP_GPU_ID_MASK (((1ULL << KFD_GPU_ID_HASH_WIDTH) - 1) \
 << KFD_MMAP_GPU_ID_SHIFT)
 #define KFD_MMAP_GPU_ID(gpu_id) ((((uint64_t)gpu_id) << 
 KFD_MMAP_GPU_ID_SHIFT)\
 & KFD_MMAP_GPU_ID_MASK)
-#define KFD_MMAP_GPU_ID_GET(offset) ((offset & KFD_MMAP_GPU_ID_MASK) \
+#define KFD_MMAP_GET_GPU_ID(offset) ((offset & KFD_MMAP_GPU_ID_MASK) \
 >> KFD_MMAP_GPU_ID_SHIFT)
-#define KFD_MMAP_OFFSET_VALUE_MASK (0x3FFFFFFFFFFFULL >> PAGE_SHIFT)
-#define KFD_MMAP_OFFSET_VALUE_GET(offset) (offset & 
 KFD_MMAP_OFFSET_VALUE_MASK)
-
 /*
 * When working with cp scheduler we should assign the HIQ manually or via
 * the amdgpu driver to a fixed hqd slot, here are the fixed HIQ hqd slot
drivers/gpu/drm/amd/amdkfd/kfd_process.c +0 -0
index 6abfb77ae540..39dc49b8fd85 100644

drivers/gpu/drm/amd/amdkfd/kfd_process.c static int kfd_process_init_cwsr_apu(struct · line 560

Quoted 9 lines
drivers/gpu/drm/amd/amdkfd/kfd_process.c +1 -2
@@ -554,8 +554,7 @@ static int kfd_process_init_cwsr_apu(struct 
 kfd_process *p, struct file *filep)
 if (!dev->cwsr_enabled || qpd->cwsr_kaddr || qpd->cwsr_base)
 continue;
- offset = (KFD_MMAP_TYPE_RESERVED_MEM | KFD_MMAP_GPU_ID(dev->id))
- << PAGE_SHIFT;
+ offset = KFD_MMAP_TYPE_RESERVED_MEM | KFD_MMAP_GPU_ID(dev->id);
 qpd->tba_addr = (int64_t)vm_mmap(filep, 0,
 KFD_CWSR_TBA_TMA_SIZE, PROT_READ | PROT_EXEC,
 MAP_SHARED, offset);

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx

A message that is not in the archive (sent off-list or not yet received): 38503488-c676-9e61-3e09-c27b1dd18e26-5C7GfCeVMHo@public.gmane.org

author
#4 lore raw

True. Thank you for spotting that. MMAP-related code was added after I inititally drafted this change earlier this year.

Regards,
Yong
________________________________
From: Kuehling, Felix <[email protected]>
Sent: Thursday, November 7, 2019 12:05 AM
To: Zhao, Yong <[email protected]>; [email protected] <[email protected]>
Subject: Re: [PATCH] drm/amdkfd: Simplify the mmap offset related bit operations

Quoted 2 lines · Yong

The new code uses straightforward bit shifts and thus has better readability.

You're missing the MMAP-related code for mmio remapping. In kfd_ioctl_alloc_memory_of_gpu:

         /* MMIO is mapped through kfd device
          * Generate a kfd mmap offset
          */
         if (flags & KFD_IOC_ALLOC_MEM_FLAGS_MMIO_REMAP) {
                 args->mmap_offset = KFD_MMAP_TYPE_MMIO |
KFD_MMAP_GPU_ID(args->gpu_id);
                 args->mmap_offset <<= PAGE_SHIFT;
         }

Regards,
Felix

Quoted 100 lines
Change-Id I0c1f7cca7e24ddb7b4ffe1cb0fa71943828ae373

---
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c | 10 ++++------ drivers/gpu/drm/amd/amdkfd/kfd_events.c | 1 - drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 9 +++------ drivers/gpu/drm/amd/amdkfd/kfd_process.c | 3 +--

4 files changed, 8 insertions(+), 15 deletions(-)
drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +0 -1
index b91993753b82..34078df36621 100644
@@ -298,7 +298,6 @@ static int kfd_ioctl_create_queue(struct file
 *filep, struct kfd_process *p,
 /* Return gpu_id as doorbell offset for mmap usage */
 args->doorbell_offset = KFD_MMAP_TYPE_DOORBELL;
 args->doorbell_offset |= KFD_MMAP_GPU_ID(args->gpu_id);
- args->doorbell_offset <<= PAGE_SHIFT;
 if (KFD_IS_SOC15(dev->device_info->asic_family))
 /* On SOC15 ASICs, include the doorbell offset within the

* process doorbell frame, which could be 1 page or 2 pages.

drivers/gpu/drm/amd/amdkfd/kfd_chardev.c +4 -5
@@ -1938,20 +1937,19 @@ static int kfd_mmap(struct file *filp, struct
 vm_area_struct *vma)
 {
 struct kfd_process *process;
 struct kfd_dev *dev = NULL;
- unsigned long vm_pgoff;
+ unsigned long mmap_offset;
 unsigned int gpu_id;
 process = kfd_get_process(current);
 if (IS_ERR(process))
 return PTR_ERR(process);
- vm_pgoff = vma->vm_pgoff;
- vma->vm_pgoff = KFD_MMAP_OFFSET_VALUE_GET(vm_pgoff);
- gpu_id = KFD_MMAP_GPU_ID_GET(vm_pgoff);
+ mmap_offset = vma->vm_pgoff << PAGE_SHIFT;
+ gpu_id = KFD_MMAP_GET_GPU_ID(mmap_offset);
 if (gpu_id)
 dev = kfd_device_by_id(gpu_id);
- switch (vm_pgoff & KFD_MMAP_TYPE_MASK) {
+ switch (mmap_offset & KFD_MMAP_TYPE_MASK) {

case KFD_MMAP_TYPE_DOORBELL:
if (!dev)
return -ENODEV;

drivers/gpu/drm/amd/amdkfd/kfd_events.c +0 -1
index 908081c85de1..1f8365575b12 100644
@@ -346,7 +346,6 @@ int kfd_event_create(struct file *devkfd, struct
 kfd_process *p,
 ret = create_signal_event(devkfd, p, ev);
 if (!ret) {
 *event_page_offset = KFD_MMAP_TYPE_EVENTS;
- *event_page_offset <<= PAGE_SHIFT;
 *event_slot_index = ev->event_id;
 }

break;

drivers/gpu/drm/amd/amdkfd/kfd_priv.h +3 -6
index 66bae8f2dad1..8eecd2cd1fd2 100644
@@ -59,24 +59,21 @@
 * NOTE: struct vm_area_struct.vm_pgoff uses offset in pages. Hence, these
 * defines are w.r.t to PAGE_SIZE
 */
-#define KFD_MMAP_TYPE_SHIFT (62 - PAGE_SHIFT)
+#define KFD_MMAP_TYPE_SHIFT (62)
 #define KFD_MMAP_TYPE_MASK (0x3ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_DOORBELL (0x3ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_EVENTS (0x2ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_RESERVED_MEM (0x1ULL << KFD_MMAP_TYPE_SHIFT)
 #define KFD_MMAP_TYPE_MMIO (0x0ULL << KFD_MMAP_TYPE_SHIFT)
-#define KFD_MMAP_GPU_ID_SHIFT (46 - PAGE_SHIFT)
+#define KFD_MMAP_GPU_ID_SHIFT (46)
 #define KFD_MMAP_GPU_ID_MASK (((1ULL << KFD_GPU_ID_HASH_WIDTH) - 1) \
 << KFD_MMAP_GPU_ID_SHIFT)
 #define KFD_MMAP_GPU_ID(gpu_id) ((((uint64_t)gpu_id) <<
 KFD_MMAP_GPU_ID_SHIFT)\
 & KFD_MMAP_GPU_ID_MASK)
-#define KFD_MMAP_GPU_ID_GET(offset) ((offset & KFD_MMAP_GPU_ID_MASK) \
+#define KFD_MMAP_GET_GPU_ID(offset) ((offset & KFD_MMAP_GPU_ID_MASK) \
 >> KFD_MMAP_GPU_ID_SHIFT)
-#define KFD_MMAP_OFFSET_VALUE_MASK (0x3FFFFFFFFFFFULL >> PAGE_SHIFT)
-#define KFD_MMAP_OFFSET_VALUE_GET(offset) (offset &
 KFD_MMAP_OFFSET_VALUE_MASK)
-
 /*
 * When working with cp scheduler we should assign the HIQ manually or via
 * the amdgpu driver to a fixed hqd slot, here are the fixed HIQ hqd slot
drivers/gpu/drm/amd/amdkfd/kfd_process.c +0 -0
index 6abfb77ae540..39dc49b8fd85 100644

drivers/gpu/drm/amd/amdkfd/kfd_process.c static int kfd_process_init_cwsr_apu(struct · line 560

Quoted 9 lines
drivers/gpu/drm/amd/amdkfd/kfd_process.c +1 -2
@@ -554,8 +554,7 @@ static int kfd_process_init_cwsr_apu(struct
 kfd_process *p, struct file *filep)
 if (!dev->cwsr_enabled || qpd->cwsr_kaddr || qpd->cwsr_base)
 continue;
- offset = (KFD_MMAP_TYPE_RESERVED_MEM | KFD_MMAP_GPU_ID(dev->id))
- << PAGE_SHIFT;
+ offset = KFD_MMAP_TYPE_RESERVED_MEM | KFD_MMAP_GPU_ID(dev->id);
 qpd->tba_addr = (int64_t)vm_mmap(filep, 0,
 KFD_CWSR_TBA_TMA_SIZE, PROT_READ | PROT_EXEC,
 MAP_SHARED, offset);

_______________________________________________
amd-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/amd-gfx