summaryrefslogtreecommitdiffstats
diff options
context:
space:
mode:
authorAndrew Lewycky <Andrew.Lewycky@amd.com>2013-11-13 15:54:17 -0500
committerAlex Deucher <alexander.deucher@amd.com>2013-11-15 15:56:22 -0500
commitd5754ab8f9b0f36ab3d6825a3555db48f28145d4 (patch)
tree0c942f43eacfaee5c1b56ea0c5330cfa790f738a
parent4cc948b94a222c310ae089c36718aac7a03aec90 (diff)
downloadop-kernel-dev-d5754ab8f9b0f36ab3d6825a3555db48f28145d4.zip
op-kernel-dev-d5754ab8f9b0f36ab3d6825a3555db48f28145d4.tar.gz
drm/radeon: use a single doorbell for cik kms compute
A single doorbell page is plenty for cik kms compute. Use a single page and manage doorbell allocation by individual doorbells rather than pages. Identify doorbells by their index rather than byte offset. Signed-off-by: Andrew Lewycky <Andrew.Lewycky@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
-rw-r--r--drivers/gpu/drm/radeon/cik.c36
-rw-r--r--drivers/gpu/drm/radeon/radeon.h23
-rw-r--r--drivers/gpu/drm/radeon/radeon_device.c47
3 files changed, 48 insertions, 58 deletions
diff --git a/drivers/gpu/drm/radeon/cik.c b/drivers/gpu/drm/radeon/cik.c
index c169897..08aa58e 100644
--- a/drivers/gpu/drm/radeon/cik.c
+++ b/drivers/gpu/drm/radeon/cik.c
@@ -1560,17 +1560,17 @@ u32 cik_get_xclk(struct radeon_device *rdev)
* cik_mm_rdoorbell - read a doorbell dword
*
* @rdev: radeon_device pointer
- * @offset: byte offset into the aperture
+ * @index: doorbell index
*
* Returns the value in the doorbell aperture at the
- * requested offset (CIK).
+ * requested doorbell index (CIK).
*/
-u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 offset)
+u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 index)
{
- if (offset < rdev->doorbell.size) {
- return readl(((void __iomem *)rdev->doorbell.ptr) + offset);
+ if (index < rdev->doorbell.num_doorbells) {
+ return readl(rdev->doorbell.ptr + index);
} else {
- DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", offset);
+ DRM_ERROR("reading beyond doorbell aperture: 0x%08x!\n", index);
return 0;
}
}
@@ -1579,18 +1579,18 @@ u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 offset)
* cik_mm_wdoorbell - write a doorbell dword
*
* @rdev: radeon_device pointer
- * @offset: byte offset into the aperture
+ * @index: doorbell index
* @v: value to write
*
* Writes @v to the doorbell aperture at the
- * requested offset (CIK).
+ * requested doorbell index (CIK).
*/
-void cik_mm_wdoorbell(struct radeon_device *rdev, u32 offset, u32 v)
+void cik_mm_wdoorbell(struct radeon_device *rdev, u32 index, u32 v)
{
- if (offset < rdev->doorbell.size) {
- writel(v, ((void __iomem *)rdev->doorbell.ptr) + offset);
+ if (index < rdev->doorbell.num_doorbells) {
+ writel(v, rdev->doorbell.ptr + index);
} else {
- DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", offset);
+ DRM_ERROR("writing beyond doorbell aperture: 0x%08x!\n", index);
}
}
@@ -4054,7 +4054,7 @@ void cik_compute_ring_set_wptr(struct radeon_device *rdev,
struct radeon_ring *ring)
{
rdev->wb.wb[ring->wptr_offs/4] = cpu_to_le32(ring->wptr);
- WDOORBELL32(ring->doorbell_offset, ring->wptr);
+ WDOORBELL32(ring->doorbell_index, ring->wptr);
}
/**
@@ -4395,10 +4395,6 @@ static int cik_cp_compute_resume(struct radeon_device *rdev)
return r;
}
- /* doorbell offset */
- rdev->ring[idx].doorbell_offset =
- (rdev->ring[idx].doorbell_page_num * PAGE_SIZE) + 0;
-
/* init the mqd struct */
memset(buf, 0, sizeof(struct bonaire_mqd));
@@ -4510,7 +4506,7 @@ static int cik_cp_compute_resume(struct radeon_device *rdev)
RREG32(CP_HQD_PQ_DOORBELL_CONTROL);
mqd->queue_state.cp_hqd_pq_doorbell_control &= ~DOORBELL_OFFSET_MASK;
mqd->queue_state.cp_hqd_pq_doorbell_control |=
- DOORBELL_OFFSET(rdev->ring[idx].doorbell_offset / 4);
+ DOORBELL_OFFSET(rdev->ring[idx].doorbell_index);
mqd->queue_state.cp_hqd_pq_doorbell_control |= DOORBELL_EN;
mqd->queue_state.cp_hqd_pq_doorbell_control &=
~(DOORBELL_SOURCE | DOORBELL_HIT);
@@ -7841,14 +7837,14 @@ int cik_init(struct radeon_device *rdev)
ring = &rdev->ring[CAYMAN_RING_TYPE_CP1_INDEX];
ring->ring_obj = NULL;
r600_ring_init(rdev, ring, 1024 * 1024);
- r = radeon_doorbell_get(rdev, &ring->doorbell_page_num);
+ r = radeon_doorbell_get(rdev, &ring->doorbell_index);
if (r)
return r;
ring = &rdev->ring[CAYMAN_RING_TYPE_CP2_INDEX];
ring->ring_obj = NULL;
r600_ring_init(rdev, ring, 1024 * 1024);
- r = radeon_doorbell_get(rdev, &ring->doorbell_page_num);
+ r = radeon_doorbell_get(rdev, &ring->doorbell_index);
if (r)
return r;
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h
index 2fe2f63..4970ac0 100644
--- a/drivers/gpu/drm/radeon/radeon.h
+++ b/drivers/gpu/drm/radeon/radeon.h
@@ -649,13 +649,15 @@ void radeon_scratch_free(struct radeon_device *rdev, uint32_t reg);
/*
* GPU doorbell structures, functions & helpers
*/
+#define RADEON_MAX_DOORBELLS 1024 /* Reserve at most 1024 doorbell slots for radeon-owned rings. */
+
struct radeon_doorbell {
- u32 num_pages;
- bool free[1024];
/* doorbell mmio */
- resource_size_t base;
- resource_size_t size;
- void __iomem *ptr;
+ resource_size_t base;
+ resource_size_t size;
+ u32 __iomem *ptr;
+ u32 num_doorbells; /* Number of doorbells actually reserved for radeon. */
+ unsigned long used[DIV_ROUND_UP(RADEON_MAX_DOORBELLS, BITS_PER_LONG)];
};
int radeon_doorbell_get(struct radeon_device *rdev, u32 *page);
@@ -802,8 +804,7 @@ struct radeon_ring {
u32 pipe;
u32 queue;
struct radeon_bo *mqd_obj;
- u32 doorbell_page_num;
- u32 doorbell_offset;
+ u32 doorbell_index;
unsigned wptr_offs;
};
@@ -2241,8 +2242,8 @@ void r100_mm_wreg(struct radeon_device *rdev, uint32_t reg, uint32_t v,
u32 r100_io_rreg(struct radeon_device *rdev, u32 reg);
void r100_io_wreg(struct radeon_device *rdev, u32 reg, u32 v);
-u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 offset);
-void cik_mm_wdoorbell(struct radeon_device *rdev, u32 offset, u32 v);
+u32 cik_mm_rdoorbell(struct radeon_device *rdev, u32 index);
+void cik_mm_wdoorbell(struct radeon_device *rdev, u32 index, u32 v);
/*
* Cast helper
@@ -2305,8 +2306,8 @@ void cik_mm_wdoorbell(struct radeon_device *rdev, u32 offset, u32 v);
#define RREG32_IO(reg) r100_io_rreg(rdev, (reg))
#define WREG32_IO(reg, v) r100_io_wreg(rdev, (reg), (v))
-#define RDOORBELL32(offset) cik_mm_rdoorbell(rdev, (offset))
-#define WDOORBELL32(offset, v) cik_mm_wdoorbell(rdev, (offset), (v))
+#define RDOORBELL32(index) cik_mm_rdoorbell(rdev, (index))
+#define WDOORBELL32(index, v) cik_mm_wdoorbell(rdev, (index), (v))
/*
* Indirect registers accessor
diff --git a/drivers/gpu/drm/radeon/radeon_device.c b/drivers/gpu/drm/radeon/radeon_device.c
index b9234c4..39b033b4 100644
--- a/drivers/gpu/drm/radeon/radeon_device.c
+++ b/drivers/gpu/drm/radeon/radeon_device.c
@@ -251,28 +251,23 @@ void radeon_scratch_free(struct radeon_device *rdev, uint32_t reg)
*/
int radeon_doorbell_init(struct radeon_device *rdev)
{
- int i;
-
/* doorbell bar mapping */
rdev->doorbell.base = pci_resource_start(rdev->pdev, 2);
rdev->doorbell.size = pci_resource_len(rdev->pdev, 2);
- /* limit to 4 MB for now */
- if (rdev->doorbell.size > (4 * 1024 * 1024))
- rdev->doorbell.size = 4 * 1024 * 1024;
+ rdev->doorbell.num_doorbells = min_t(u32, rdev->doorbell.size / sizeof(u32), RADEON_MAX_DOORBELLS);
+ if (rdev->doorbell.num_doorbells == 0)
+ return -EINVAL;
- rdev->doorbell.ptr = ioremap(rdev->doorbell.base, rdev->doorbell.size);
+ rdev->doorbell.ptr = ioremap(rdev->doorbell.base, rdev->doorbell.num_doorbells * sizeof(u32));
if (rdev->doorbell.ptr == NULL) {
return -ENOMEM;
}
DRM_INFO("doorbell mmio base: 0x%08X\n", (uint32_t)rdev->doorbell.base);
DRM_INFO("doorbell mmio size: %u\n", (unsigned)rdev->doorbell.size);
- rdev->doorbell.num_pages = rdev->doorbell.size / PAGE_SIZE;
+ memset(&rdev->doorbell.used, 0, sizeof(rdev->doorbell.used));
- for (i = 0; i < rdev->doorbell.num_pages; i++) {
- rdev->doorbell.free[i] = true;
- }
return 0;
}
@@ -290,40 +285,38 @@ void radeon_doorbell_fini(struct radeon_device *rdev)
}
/**
- * radeon_doorbell_get - Allocate a doorbell page
+ * radeon_doorbell_get - Allocate a doorbell entry
*
* @rdev: radeon_device pointer
- * @doorbell: doorbell page number
+ * @doorbell: doorbell index
*
- * Allocate a doorbell page for use by the driver (all asics).
+ * Allocate a doorbell for use by the driver (all asics).
* Returns 0 on success or -EINVAL on failure.
*/
int radeon_doorbell_get(struct radeon_device *rdev, u32 *doorbell)
{
- int i;
-
- for (i = 0; i < rdev->doorbell.num_pages; i++) {
- if (rdev->doorbell.free[i]) {
- rdev->doorbell.free[i] = false;
- *doorbell = i;
- return 0;
- }
+ unsigned long offset = find_first_zero_bit(rdev->doorbell.used, rdev->doorbell.num_doorbells);
+ if (offset < rdev->doorbell.num_doorbells) {
+ __set_bit(offset, rdev->doorbell.used);
+ *doorbell = offset;
+ return 0;
+ } else {
+ return -EINVAL;
}
- return -EINVAL;
}
/**
- * radeon_doorbell_free - Free a doorbell page
+ * radeon_doorbell_free - Free a doorbell entry
*
* @rdev: radeon_device pointer
- * @doorbell: doorbell page number
+ * @doorbell: doorbell index
*
- * Free a doorbell page allocated for use by the driver (all asics)
+ * Free a doorbell allocated for use by the driver (all asics)
*/
void radeon_doorbell_free(struct radeon_device *rdev, u32 doorbell)
{
- if (doorbell < rdev->doorbell.num_pages)
- rdev->doorbell.free[doorbell] = true;
+ if (doorbell < rdev->doorbell.num_doorbells)
+ __clear_bit(doorbell, rdev->doorbell.used);
}
/*
OpenPOWER on IntegriCloud