[PATCH 1/8] drm/msm/dpu: support allocating multiple CWB pairs for quad-pipe

From: Jun Nie

Date: Thu Nov 12 2026 - 03:00:32 EST


Currently, CWB allocation only supports a single pair of CWB muxes and
dedicated pingpong blocks. In quad-pipe topologies, two pairs of CWB blocks
are required to feed concurrent writeback paths.

Because each Layer Mixer (LM) iteration previously restarted the search
from CWB index 0, allocating a second pair would attempt to re-evaluate
already considered or reserved CWB blocks.

Track the start search index (`cwb_mux_idx_start`) and advance it past the
most recently assigned CWB block. This ensures subsequent LMs only scan
unallocated CWB blocks, enabling support for multiple CWB pairs.

Signed-off-by: Jun Nie <jun.nie@xxxxxxxxxx>
---
drivers/gpu/drm/msm/disp/dpu1/dpu_rm.c | 10 ++++++++--
1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/drivers/gpu/drm/msm/disp/dpu1/dpu_rm.c b/drivers/gpu/drm/msm/disp/dpu1/dpu_rm.c
index 7e77d88f8959..4ada16cf8610 100644
--- a/drivers/gpu/drm/msm/disp/dpu1/dpu_rm.c
+++ b/drivers/gpu/drm/msm/disp/dpu1/dpu_rm.c
@@ -244,6 +244,7 @@ static int _dpu_rm_reserve_cwb_mux_and_pingpongs(struct dpu_rm *rm,
int cwb_pp_start_idx = PINGPONG_CWB_0 - PINGPONG_0;
int cwb_pp_idx[MAX_BLOCKS];
int cwb_mux_idx[MAX_BLOCKS];
+ int cwb_mux_idx_start = 0;

/*
* Reserve additional dedicated CWB PINGPONG blocks and muxes for each
@@ -252,9 +253,9 @@ static int _dpu_rm_reserve_cwb_mux_and_pingpongs(struct dpu_rm *rm,
* TODO: add support reserving resources for platforms with no
* PINGPONG_CWB
*/
- for (int i = 0; i < ARRAY_SIZE(rm->mixer_blks) &&
+ for (int i = 0; i < ARRAY_SIZE(rm->cwb_blks) &&
cwb_mux_count < num_cwb_mux; i++) {
- for (int j = 0; j < ARRAY_SIZE(rm->cwb_blks); j++) {
+ for (int j = cwb_mux_idx_start; j < ARRAY_SIZE(rm->cwb_blks); j++) {
/*
* Odd LMs must be assigned to odd CWB muxes and even
* LMs with even CWB muxes.
@@ -268,6 +269,11 @@ static int _dpu_rm_reserve_cwb_mux_and_pingpongs(struct dpu_rm *rm,
i % 2 != j % 2)
continue;

+ /*
+ * Advance search offset past the allocated block so
+ * subsequent mixers allocate the next available CWB block.
+ */
+ cwb_mux_idx_start = j + 1;
cwb_mux_idx[cwb_mux_count] = j;
cwb_pp_idx[cwb_mux_count] = j + cwb_pp_start_idx;
cwb_mux_count++;

--
2.53.0