Commit a85a7da4 authored by Alex Deucher's avatar Alex Deucher Committed by Christian König

drm/radeon: update ib_execute for SI (v2)

When submitting a CONST_IB, emit a SWITCH_BUFFER
packet before the CONST_IB.  This isn't strictly necessary
(the driver will work fine without it), but is good practice
and allows for more flexible DE/CE sychronization options
in the future.  Current userspace drivers do not take
advantage of the CE yet.

v2: - clean up code flow a bit
    - no need to flush caches for CONST IB
Signed-off-by: default avatarAlex Deucher <alexander.deucher@amd.com>
Reviewed-by: default avatarChristian König <christian.koenig@amd.com>
parent 4ef72566
...@@ -1765,18 +1765,23 @@ void si_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib) ...@@ -1765,18 +1765,23 @@ void si_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
struct radeon_ring *ring = &rdev->ring[ib->ring]; struct radeon_ring *ring = &rdev->ring[ib->ring];
u32 header; u32 header;
if (ring->rptr_save_reg) { if (ib->is_const_ib) {
uint32_t next_rptr = ring->wptr + 3 + 4 + 8; /* set switch buffer packet before const IB */
radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); radeon_ring_write(ring, PACKET3(PACKET3_SWITCH_BUFFER, 0));
radeon_ring_write(ring, ((ring->rptr_save_reg - radeon_ring_write(ring, 0);
PACKET3_SET_CONFIG_REG_START) >> 2));
radeon_ring_write(ring, next_rptr);
}
if (ib->is_const_ib)
header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2); header = PACKET3(PACKET3_INDIRECT_BUFFER_CONST, 2);
else } else {
if (ring->rptr_save_reg) {
uint32_t next_rptr = ring->wptr + 3 + 4 + 8;
radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
radeon_ring_write(ring, ((ring->rptr_save_reg -
PACKET3_SET_CONFIG_REG_START) >> 2));
radeon_ring_write(ring, next_rptr);
}
header = PACKET3(PACKET3_INDIRECT_BUFFER, 2); header = PACKET3(PACKET3_INDIRECT_BUFFER, 2);
}
radeon_ring_write(ring, header); radeon_ring_write(ring, header);
radeon_ring_write(ring, radeon_ring_write(ring,
...@@ -1787,18 +1792,20 @@ void si_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib) ...@@ -1787,18 +1792,20 @@ void si_ring_ib_execute(struct radeon_device *rdev, struct radeon_ib *ib)
radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF); radeon_ring_write(ring, upper_32_bits(ib->gpu_addr) & 0xFFFF);
radeon_ring_write(ring, ib->length_dw | (ib->vm_id << 24)); radeon_ring_write(ring, ib->length_dw | (ib->vm_id << 24));
/* flush read cache over gart for this vmid */ if (!ib->is_const_ib) {
radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1)); /* flush read cache over gart for this vmid */
radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2); radeon_ring_write(ring, PACKET3(PACKET3_SET_CONFIG_REG, 1));
radeon_ring_write(ring, ib->vm_id); radeon_ring_write(ring, (CP_COHER_CNTL2 - PACKET3_SET_CONFIG_REG_START) >> 2);
radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3)); radeon_ring_write(ring, ib->vm_id);
radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA | radeon_ring_write(ring, PACKET3(PACKET3_SURFACE_SYNC, 3));
PACKET3_TC_ACTION_ENA | radeon_ring_write(ring, PACKET3_TCL1_ACTION_ENA |
PACKET3_SH_KCACHE_ACTION_ENA | PACKET3_TC_ACTION_ENA |
PACKET3_SH_ICACHE_ACTION_ENA); PACKET3_SH_KCACHE_ACTION_ENA |
radeon_ring_write(ring, 0xFFFFFFFF); PACKET3_SH_ICACHE_ACTION_ENA);
radeon_ring_write(ring, 0); radeon_ring_write(ring, 0xFFFFFFFF);
radeon_ring_write(ring, 10); /* poll interval */ radeon_ring_write(ring, 0);
radeon_ring_write(ring, 10); /* poll interval */
}
} }
/* /*
......
...@@ -901,5 +901,6 @@ ...@@ -901,5 +901,6 @@
#define PACKET3_WAIT_ON_DE_COUNTER_DIFF 0x88 #define PACKET3_WAIT_ON_DE_COUNTER_DIFF 0x88
#define PACKET3_SET_CE_DE_COUNTERS 0x89 #define PACKET3_SET_CE_DE_COUNTERS 0x89
#define PACKET3_WAIT_ON_AVAIL_BUFFER 0x8A #define PACKET3_WAIT_ON_AVAIL_BUFFER 0x8A
#define PACKET3_SWITCH_BUFFER 0x8B
#endif #endif
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment