Add support to print sub block registers in dpu hw catalog
cover Add support to print sub block registers in dpu hw catalog
The purpose of this patch series is to add support to print the registers of sub blocks in the dpu hardware catalog and fix the order in which all hardware blocks are dumped for a device core dump. This involves:
1. Changing data structure from stack to queue to fix the printing order of the device core dump.
2. Removing redundant suffix of sub block names.
3. Removing redundant prefix of sub block names.
4. Eliminating unused variable from relevant macros.
5. Defining names for sub blocks that have not yet been defined.
6. Implementing wrapper function that prints the registers of sub blocks when there is a need.
Sample Output of the sspp_0 block and its sub blocks for devcore dump:
======sspp_0======
...registers
...
====sspp_0_scaler====
...
...
====sspp_0_csc====
...
...
====next_block====
...
---
Ryan McCann (6):
drm/msm: Update dev core dump to not print backwards drm/msm/dpu: Drop unused num argument from relevant macros drm/msm/dpu: Define names for unnamed sblks drm/msm/dpu: Remove redundant suffix in name of sub blocks drm/msm/disp: Remove redundant prefix in name of sub blocks drm/msm/dpu: Update dev core dump to dump registers of sub blocks
drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c | 90 +++++----- drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 194 +++++++++++++++++++--- drivers/gpu/drm/msm/disp/msm_disp_snapshot_util.c | 2 +- 3 files changed, 214 insertions(+), 72 deletions(-)
---
base-commit: 710025fdedb3767655823c3a12d27d404d209f75
change-id: 20230622-devcoredump_patch-df7e8f6fd632
Best regards,
Signature · 1 line
Ryan McCann <[email protected]>
PATCH 1/6 drm/msm: Update dev core dump to not print backwards
Device core dump add block method adds hardware blocks to dumping queue with stack behavior which causes the hardware blocks to be printed in reverse order. Change the addition to dumping queue data structure from "list_add" to "list_add_tail" for FIFO queue behavior.
--- drivers/gpu/drm/msm/disp/msm_disp_snapshot_util.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/gpu/drm/msm/disp/msm_disp_snapshot_util.c +1 -1
@@ -192,5 +192,5 @@ void msm_disp_snapshot_add_block(struct msm_disp_state *disp_state, u32 len, new_blk->base_addr = base_addr;msm_disp_state_dump_regs (&new_blk->state, new_blk->size, base_addr); -list_add (&new_blk->node, &disp_state->blocks); +list_add_tail (&new_blk->node, &disp_state->blocks); }
Signature · 1 line
2.25.1
Re: [PATCH 1/6] drm/msm: Update dev core dump to not print backwards
Quoted 10 lines · Ryan McCann
Device core dump add block method adds hardware blocks to dumping queue with stack behavior which causes the hardware blocks to be printed in reverse order. Change the addition to dumping queue data structure from "
list_add" to "list_add_tail" for FIFO queue behavior.Fixes 98659487b845 ("drm/msm: add support to take dpu snapshot")Signed-off-by Ryan McCann <[email protected]>--- drivers/gpu/drm/msm/disp/msm_disp_snapshot_util.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
Signature · 2 lines
With best wishes Dmitry
PATCH 2/6 drm/msm/dpu: Drop unused num argument from relevant macros
In reply to Ryan McCannDrop unused parameter "num" from VIG_SBLK_NOSCALE and DMA sub block macros. Update calls to relevant macros to reflect change.
--- drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-)
drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c +11 -11
@@ -288,7 +288,7 @@ static const uint32_t wb2_formats[] = { .rotation_cfg = rot_cfg, \ } -#define _DMA_SBLK (num, sdma_pri) \ +#define _DMA_SBLK (sdma_pri) \ { \ .maxdwnscale = SSPP_UNITY_SCALE, \ .maxupscale = SSPP_UNITY_SCALE, \
@@ -323,10 +323,10 @@ static const struct dpu_sspp_sub_blks sdm845_vig_sblk_2 =static const struct dpu_sspp_sub_blks sdm845_vig_sblk_3 =_VIG_SBLK ("3" ,8 ,DPU_SSPP_SCALER_QSEED3 ); -static const struct dpu_sspp_sub_blks sdm845_dma_sblk_0 =_DMA_SBLK ("8" ,1 ); -static const struct dpu_sspp_sub_blks sdm845_dma_sblk_1 =_DMA_SBLK ("9" ,2 ); -static const struct dpu_sspp_sub_blks sdm845_dma_sblk_2 =_DMA_SBLK ("10" ,3 ); -static const struct dpu_sspp_sub_blks sdm845_dma_sblk_3 =_DMA_SBLK ("11" ,4 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_0 =_DMA_SBLK (1 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_1 =_DMA_SBLK (2 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_2 =_DMA_SBLK (3 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_3 =_DMA_SBLK (4 );#define SSPP_BLK (_name, _id, _base, _len, _features, \ _sblk, _xinid, _type, _clkctrl) \
@@ -365,11 +365,11 @@ static const struct dpu_sspp_sub_blks sm8550_vig_sblk_1 =static const struct dpu_sspp_sub_blks sm8550_vig_sblk_2 =_VIG_SBLK ("2" ,9 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8550_vig_sblk_3 = -_VIG_SBLK ("3" ,10 ,DPU_SSPP_SCALER_QSEED4 ); -static const struct dpu_sspp_sub_blks sm8550_dma_sblk_4 =_DMA_SBLK ("12" ,5 ); -static const struct dpu_sspp_sub_blks sm8550_dma_sblk_5 =_DMA_SBLK ("13" ,6 ); +_VIG_SBLK (10 ,DPU_SSPP_SCALER_QSEED4 ); +static const struct dpu_sspp_sub_blks sm8550_dma_sblk_4 =_DMA_SBLK (5 ); +static const struct dpu_sspp_sub_blks sm8550_dma_sblk_5 =_DMA_SBLK (6 ); -#define _VIG_SBLK_NOSCALE (num, sdma_pri) \ +#define _VIG_SBLK_NOSCALE (sdma_pri) \ { \ .maxdwnscale = SSPP_UNITY_SCALE, \ .maxupscale = SSPP_UNITY_SCALE, \
@@ -380,8 +380,8 @@ static const struct dpu_sspp_sub_blks sm8550_dma_sblk_5 = _DMA_SBLK("13", 6); .virt_num_formats = ARRAY_SIZE(plane_formats), \ } -static const struct dpu_sspp_sub_blks qcm2290_vig_sblk_0 =_VIG_SBLK_NOSCALE ("0" ,2 ); -static const struct dpu_sspp_sub_blks qcm2290_dma_sblk_0 =_DMA_SBLK ("8" ,1 ); +static const struct dpu_sspp_sub_blks qcm2290_vig_sblk_0 =_VIG_SBLK_NOSCALE (2 ); +static const struct dpu_sspp_sub_blks qcm2290_dma_sblk_0 =_DMA_SBLK (1 );/************************************************************* * MIXER sub blocks config
Signature · 1 line
2.25.1
Re: [PATCH 2/6] drm/msm/dpu: Drop unused num argument from relevant macros
Quoted 2 lines · Ryan McCann
Drop unused parameter "num" from
VIG_SBLK_NOSCALEand DMA sub block macros. Update calls to relevant macros to reflect change.
This is not relevant for this patchset.
With https://patchwork.freedesktop.org/patch/534745/?series=116851&rev=2 in place, VIG_SBLK and VIG_SBLK_ROT should also stop using num. But let's probably take another step forward and drop both arguments and use just a single instance of sblk per platform. I'll send a patch for it.
Quoted 62 lines
Signed-off-by Ryan McCann <[email protected]>--- drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c +11
-11@@ -288,7 +288,7 @@ static const uint32_t wb2_formats[] = { .rotation_cfg = rot_cfg, \ } -#define _DMA_SBLK (num, sdma_pri) \ +#define _DMA_SBLK (sdma_pri) \ { \ .maxdwnscale = SSPP_UNITY_SCALE, \ .maxupscale = SSPP_UNITY_SCALE, \@@ -323,10 +323,10 @@ static const struct dpu_sspp_sub_blks sdm845_vig_sblk_2 =static const struct dpu_sspp_sub_blks sdm845_vig_sblk_3 =_VIG_SBLK ("3" ,8 ,DPU_SSPP_SCALER_QSEED3 ); -static const struct dpu_sspp_sub_blks sdm845_dma_sblk_0 =_DMA_SBLK ("8" ,1 ); -static const struct dpu_sspp_sub_blks sdm845_dma_sblk_1 =_DMA_SBLK ("9" ,2 ); -static const struct dpu_sspp_sub_blks sdm845_dma_sblk_2 =_DMA_SBLK ("10" ,3 ); -static const struct dpu_sspp_sub_blks sdm845_dma_sblk_3 =_DMA_SBLK ("11" ,4 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_0 =_DMA_SBLK (1 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_1 =_DMA_SBLK (2 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_2 =_DMA_SBLK (3 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_3 =_DMA_SBLK (4 );#define SSPP_BLK (_name, _id, _base, _len, _features, \ _sblk, _xinid, _type, _clkctrl) \@@ -365,11 +365,11 @@ static const struct dpu_sspp_sub_blks sm8550_vig_sblk_1 =static const struct dpu_sspp_sub_blks sm8550_vig_sblk_2 =_VIG_SBLK ("2" ,9 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8550_vig_sblk_3 = -_VIG_SBLK ("3" ,10 ,DPU_SSPP_SCALER_QSEED4 ); -static const struct dpu_sspp_sub_blks sm8550_dma_sblk_4 =_DMA_SBLK ("12" ,5 ); -static const struct dpu_sspp_sub_blks sm8550_dma_sblk_5 =_DMA_SBLK ("13" ,6 ); +_VIG_SBLK (10 ,DPU_SSPP_SCALER_QSEED4 ); +static const struct dpu_sspp_sub_blks sm8550_dma_sblk_4 =_DMA_SBLK (5 ); +static const struct dpu_sspp_sub_blks sm8550_dma_sblk_5 =_DMA_SBLK (6 ); -#define _VIG_SBLK_NOSCALE (num, sdma_pri) \ +#define _VIG_SBLK_NOSCALE (sdma_pri) \ { \ .maxdwnscale = SSPP_UNITY_SCALE, \ .maxupscale = SSPP_UNITY_SCALE, \@@ -380,8 +380,8 @@ static const struct dpu_sspp_sub_blks sm8550_dma_sblk_5 = _DMA_SBLK("13", 6); .virt_num_formats = ARRAY_SIZE(plane_formats), \ } -static const struct dpu_sspp_sub_blks qcm2290_vig_sblk_0 =_VIG_SBLK_NOSCALE ("0" ,2 ); -static const struct dpu_sspp_sub_blks qcm2290_dma_sblk_0 =_DMA_SBLK ("8" ,1 ); +static const struct dpu_sspp_sub_blks qcm2290_vig_sblk_0 =_VIG_SBLK_NOSCALE (2 ); +static const struct dpu_sspp_sub_blks qcm2290_dma_sblk_0 =_DMA_SBLK (1 );/************************************************************* * MIXER sub blocks config
Signature · 2 lines
With best wishes Dmitry
Re: [PATCH 2/6] drm/msm/dpu: Drop unused num argument from relevant macros
Quoted 5 lines · Dmitry Baryshkov
Quoted 2 lines · Ryan McCann
Drop unused parameter "num" from
VIG_SBLK_NOSCALEand DMA sub block macros. Update calls to relevant macros to reflect change.This is not relevant for this patchset.
Ok, after reviewing the rest of the patches, I see that it is indeed relevant. However could you please follow another path (outlined in the review for patch 6).
Note, that for the sake of subblock dumping we can ignore the issue with the names completely and drop them in the separate patchset.
Quoted 73 lines
With https://patchwork.freedesktop.org/patch/534745/?series=116851&rev=2 in place,
VIG_SBLKandVIG_SBLK_ROTshould also stop using num. But let's probably take another step forward and drop both arguments and use just a single instance of sblk per platform. I'll send a patch for it.Quoted 69 lines
Signed-off-by Ryan McCann <[email protected]>---
drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c| 22
+++++++++++-----------1 file changed, 11 insertions(+), 11 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c +1
-1@@ -288,7 +288,7 @@ static const uint32_t wb2_formats[] = { .rotation_cfg = rot_cfg, \ } -#define _DMA_SBLK (num, sdma_pri) \ +#define _DMA_SBLK (sdma_pri) \ { \ .maxdwnscale = SSPP_UNITY_SCALE, \ .maxupscale = SSPP_UNITY_SCALE, \@@ -323,10 +323,10 @@ static const struct dpu_sspp_sub_blks
sdm845_vig_sblk_2=
static const structdpu_sspp_sub_blkssdm845_vig_sblk_3=_VIG_SBLK("3", 8, DPU_SSPP_SCALER_QSEED3); -static const structdpu_sspp_sub_blkssdm845_dma_sblk_0=_DMA_SBLK("8", 1);
-static const structdpu_sspp_sub_blkssdm845_dma_sblk_1=_DMA_SBLK("9", 2);
-static const structdpu_sspp_sub_blkssdm845_dma_sblk_2=_DMA_SBLK("10", 3);
-static const structdpu_sspp_sub_blkssdm845_dma_sblk_3=_DMA_SBLK("11", 4);drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c +4
-0+static const struct dpu_sspp_sub_blks sdm845_dma_sblk_0 =_DMA_SBLK (1 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_1 =_DMA_SBLK (2 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_2 =_DMA_SBLK (3 ); +static const struct dpu_sspp_sub_blks sdm845_dma_sblk_3 =_DMA_SBLK (4 );#define SSPP_BLK(_name,
_id,_base,_len,_features, \
_sblk,_xinid,_type,_clkctrl) \drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c +0
-0@@ -365,11 +365,11 @@ static const struct dpu_sspp_sub_blks
sm8550_vig_sblk_1=
static const structdpu_sspp_sub_blkssm8550_vig_sblk_2=_VIG_SBLK("2", 9, DPU_SSPP_SCALER_QSEED4); static const structdpu_sspp_sub_blkssm8550_vig_sblk_3= -_VIG_SBLK("3", 10, DPU_SSPP_SCALER_QSEED4); -static const structdpu_sspp_sub_blkssm8550_dma_sblk_4=_DMA_SBLK("12", 5);
-static const structdpu_sspp_sub_blkssm8550_dma_sblk_5=_DMA_SBLK("13", 6);
+_VIG_SBLK(10, DPU_SSPP_SCALER_QSEED4);drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c +3
-1+static const struct dpu_sspp_sub_blks sm8550_dma_sblk_4 =_DMA_SBLK (5 ); +static const struct dpu_sspp_sub_blks sm8550_dma_sblk_5 =_DMA_SBLK (6 ); -#define _VIG_SBLK_NOSCALE (num, sdma_pri) \ +#define _VIG_SBLK_NOSCALE (sdma_pri) \{ \
.maxdwnscale =SSPP_UNITY_SCALE, \
.maxupscale =SSPP_UNITY_SCALE, \drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c +2
-2@@ -380,8 +380,8 @@ static const struct dpu_sspp_sub_blks sm8550_dma_sblk_5 =_DMA_SBLK ("13" ,6 ); .virt_num_formats = ARRAY_SIZE(plane_formats), \ } -static const struct dpu_sspp_sub_blks qcm2290_vig_sblk_0 =_VIG_SBLK_NOSCALE ("0" ,2 ); -static const struct dpu_sspp_sub_blks qcm2290_dma_sblk_0 =_DMA_SBLK ("8" ,1 ); +static const struct dpu_sspp_sub_blks qcm2290_vig_sblk_0 =_VIG_SBLK_NOSCALE (2 ); +static const struct dpu_sspp_sub_blks qcm2290_dma_sblk_0 =_DMA_SBLK (1 );/*************************************************************
* MIXER sub blocks config
Signature · 2 lines
With best wishes Dmitry
PATCH 3/6 drm/msm/dpu: Define names for unnamed sblks
In reply to Ryan McCannSome sub blocks in the hw catalog have not been given a name, so when the registers from that block are dumped, there is no name to reference. Define names for relevant sub blocks to fix this.
--- drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c | 20 ++++++++++---------- 1 file changed, 10 insertions(+), 10 deletions(-)
drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c +10 -10
@@ -444,12 +444,12 @@ static const struct dpu_lm_sub_blks qcm2290_lm_sblk = {* DSPP sub blocks config *************************************************************/ static const struct dpu_dspp_sub_blks msm8998_dspp_sblk = { - .pcc = {.id =DPU_DSPP_PCC , .base =0x1700 , + .pcc = {.name ="pcc" , .id =DPU_DSPP_PCC , .base =0x1700 , .len =0x90 , .version =0x10007 }, };static const struct dpu_dspp_sub_blks sm8150_dspp_sblk = { - .pcc = {.id =DPU_DSPP_PCC , .base =0x1700 , + .pcc = {.name ="pcc" , .id =DPU_DSPP_PCC , .base =0x1700 , .len =0x90 , .version =0x40000 }, };
@@ -465,19 +465,19 @@ static const struct dpu_dspp_sub_blks sm8150_dspp_sblk = {* PINGPONG sub blocks config *************************************************************/ static const struct dpu_pingpong_sub_blks sdm845_pp_sblk_te = { - .te2 = {.id =DPU_PINGPONG_TE2 , .base =0x2000 , .len =0x0 , + .te2 = {.name ="te2" , .id =DPU_PINGPONG_TE2 , .base =0x2000 , .len =0x0 , .version =0x1 }, - .dither = {.id =DPU_PINGPONG_DITHER , .base =0x30e0 , + .dither = {.name ="dither" , .id =DPU_PINGPONG_DITHER , .base =0x30e0 , .len =0x20 , .version =0x10000 }, };static const struct dpu_pingpong_sub_blks sdm845_pp_sblk = { - .dither = {.id =DPU_PINGPONG_DITHER , .base =0x30e0 , + .dither = {.name ="dither" , .id =DPU_PINGPONG_DITHER , .base =0x30e0 , .len =0x20 , .version =0x10000 }, };static const struct dpu_pingpong_sub_blks sc7280_pp_sblk = { - .dither = {.id =DPU_PINGPONG_DITHER , .base =0xe0 , + .dither = {.name ="dither" , .id =DPU_PINGPONG_DITHER , .base =0xe0 , .len =0x20 , .version =0x20000 }, };
@@ -517,13 +517,13 @@ static const struct dpu_pingpong_sub_blks sc7280_pp_sblk = {* DSC sub blocks config *************************************************************/ static const struct dpu_dsc_sub_blks dsc_sblk_0 = { - .enc = {.base =0x100 , .len =0x100 }, - .ctl = {.base =0xF00 , .len =0x10 }, + .enc = {.name ="enc" , .base =0x100 , .len =0x100 }, + .ctl = {.name ="ctl" , .base =0xF00 , .len =0x10 }, };static const struct dpu_dsc_sub_blks dsc_sblk_1 = { - .enc = {.base =0x200 , .len =0x100 }, - .ctl = {.base =0xF80 , .len =0x10 }, + .enc = {.name ="enc" , .base =0x200 , .len =0x100 }, + .ctl = {.name ="ctl" , .base =0xF80 , .len =0x10 }, };#define DSC_BLK (_name, _id, _base, _features) \
Signature · 1 line
2.25.1
PATCH 4/6 drm/msm/dpu: Remove redundant suffix in name of sub blocks
In reply to Ryan McCannFor a device core dump, the registers of sub blocks are printed under the title <mainBlkName_sblkName>. For example, the csc sub block for an SSPP block "sspp_0" would be printed "sspp_0_csc0". There is a redundant 0 in the title due to a concatention done in the definition of the VIG_SBLK macro, the macro used to define the sub blocks of "sspp_0". Remove this concatenation to eliminate redundancy and remove the num parameter of relevant macros as a consequence of it no longer being used.
--- drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c | 48 +++++++++++++------------- 1 file changed, 24 insertions(+), 24 deletions(-)
drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c +24 -24
@@ -252,15 +252,15 @@ static const uint32_t wb2_formats[] = {*************************************************************/ /* SSPP common configuration */ -#define _VIG_SBLK (num, sdma_pri, qseed_ver) \ +#define _VIG_SBLK (sdma_pri, qseed_ver) \ { \ .maxdwnscale = MAX_DOWNSCALE_RATIO, \ .maxupscale = MAX_UPSCALE_RATIO, \ .smart_dma_priority = sdma_pri, \ - .scaler_blk = {.name = STRCAT("sspp_scaler", num), \ + .scaler_blk = {.name = "sspp_scaler", \ .id = qseed_ver, \ .base = 0xa00, .len = 0xa0,}, \ - .csc_blk = {.name = STRCAT("sspp_csc", num), \ + .csc_blk = {.name = "sspp_csc", \ .id = DPU_SSPP_CSC_10BIT, \ .base = 0x1a00, .len = 0x100,}, \ .format_list = plane_formats_yuv, \
@@ -270,15 +270,15 @@ static const uint32_t wb2_formats[] = { .rotation_cfg = NULL, \ } -#define _VIG_SBLK_ROT (num, sdma_pri, qseed_ver, rot_cfg) \ +#define _VIG_SBLK_ROT (sdma_pri, qseed_ver, rot_cfg) \ { \ .maxdwnscale = MAX_DOWNSCALE_RATIO, \ .maxupscale = MAX_UPSCALE_RATIO, \ .smart_dma_priority = sdma_pri, \ - .scaler_blk = {.name = STRCAT("sspp_scaler", num), \ + .scaler_blk = {.name = "sspp_scaler", \ .id = qseed_ver, \ .base = 0xa00, .len = 0xa0,}, \ - .csc_blk = {.name = STRCAT("sspp_csc", num), \ + .csc_blk = {.name = "sspp_csc", \ .id = DPU_SSPP_CSC_10BIT, \ .base = 0x1a00, .len = 0x100,}, \ .format_list = plane_formats_yuv, \
@@ -300,13 +300,13 @@ static const uint32_t wb2_formats[] = { }static const struct dpu_sspp_sub_blks msm8998_vig_sblk_0 = -_VIG_SBLK ("0" ,0 ,DPU_SSPP_SCALER_QSEED3 ); +_VIG_SBLK (0 ,DPU_SSPP_SCALER_QSEED3 );static const struct dpu_sspp_sub_blks msm8998_vig_sblk_1 = -_VIG_SBLK ("1" ,0 ,DPU_SSPP_SCALER_QSEED3 ); +_VIG_SBLK (0 ,DPU_SSPP_SCALER_QSEED3 );static const struct dpu_sspp_sub_blks msm8998_vig_sblk_2 = -_VIG_SBLK ("2" ,0 ,DPU_SSPP_SCALER_QSEED3 ); +_VIG_SBLK (0 ,DPU_SSPP_SCALER_QSEED3 );static const struct dpu_sspp_sub_blks msm8998_vig_sblk_3 = -_VIG_SBLK ("3" ,0 ,DPU_SSPP_SCALER_QSEED3 ); +_VIG_SBLK (0 ,DPU_SSPP_SCALER_QSEED3 );static const struct dpu_rotation_cfg dpu_rot_sc7280_cfg_v2 = { .rot_maxheight =1088 ,
@@ -315,13 +315,13 @@ static const struct dpu_rotation_cfg dpu_rot_sc7280_cfg_v2 = { };static const struct dpu_sspp_sub_blks sdm845_vig_sblk_0 = -_VIG_SBLK ("0" ,5 ,DPU_SSPP_SCALER_QSEED3 ); +_VIG_SBLK (5 ,DPU_SSPP_SCALER_QSEED3 );static const struct dpu_sspp_sub_blks sdm845_vig_sblk_1 = -_VIG_SBLK ("1" ,6 ,DPU_SSPP_SCALER_QSEED3 ); +_VIG_SBLK (6 ,DPU_SSPP_SCALER_QSEED3 );static const struct dpu_sspp_sub_blks sdm845_vig_sblk_2 = -_VIG_SBLK ("2" ,7 ,DPU_SSPP_SCALER_QSEED3 ); +_VIG_SBLK (7 ,DPU_SSPP_SCALER_QSEED3 );static const struct dpu_sspp_sub_blks sdm845_vig_sblk_3 = -_VIG_SBLK ("3" ,8 ,DPU_SSPP_SCALER_QSEED3 ); +_VIG_SBLK (8 ,DPU_SSPP_SCALER_QSEED3 );static const struct dpu_sspp_sub_blks sdm845_dma_sblk_0 =_DMA_SBLK (1 );static const struct dpu_sspp_sub_blks sdm845_dma_sblk_1 =_DMA_SBLK (2 );
@@ -341,29 +341,29 @@ static const struct dpu_sspp_sub_blks sdm845_dma_sblk_3 = _DMA_SBLK(4); }static const struct dpu_sspp_sub_blks sc7180_vig_sblk_0 = -_VIG_SBLK ("0" ,4 ,DPU_SSPP_SCALER_QSEED4 ); +_VIG_SBLK (4 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sc7280_vig_sblk_0 = -_VIG_SBLK_ROT ("0" ,4 ,DPU_SSPP_SCALER_QSEED4 , &dpu_rot_sc7280_cfg_v2); +_VIG_SBLK_ROT (4 ,DPU_SSPP_SCALER_QSEED4 , &dpu_rot_sc7280_cfg_v2);static const struct dpu_sspp_sub_blks sm6115_vig_sblk_0 = -_VIG_SBLK ("0" ,2 ,DPU_SSPP_SCALER_QSEED4 ); +_VIG_SBLK (2 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8250_vig_sblk_0 = -_VIG_SBLK ("0" ,5 ,DPU_SSPP_SCALER_QSEED4 ); +_VIG_SBLK (5 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8250_vig_sblk_1 = -_VIG_SBLK ("1" ,6 ,DPU_SSPP_SCALER_QSEED4 ); +_VIG_SBLK (6 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8250_vig_sblk_2 = -_VIG_SBLK ("2" ,7 ,DPU_SSPP_SCALER_QSEED4 ); +_VIG_SBLK (7 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8250_vig_sblk_3 = -_VIG_SBLK ("3" ,8 ,DPU_SSPP_SCALER_QSEED4 ); +_VIG_SBLK (8 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8550_vig_sblk_0 = -_VIG_SBLK ("0" ,7 ,DPU_SSPP_SCALER_QSEED4 ); +_VIG_SBLK (7 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8550_vig_sblk_1 = -_VIG_SBLK ("1" ,8 ,DPU_SSPP_SCALER_QSEED4 ); +_VIG_SBLK (8 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8550_vig_sblk_2 = -_VIG_SBLK ("2" ,9 ,DPU_SSPP_SCALER_QSEED4 ); +_VIG_SBLK (9 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8550_vig_sblk_3 =_VIG_SBLK (10 ,DPU_SSPP_SCALER_QSEED4 );static const struct dpu_sspp_sub_blks sm8550_dma_sblk_4 =_DMA_SBLK (5 );
Signature · 1 line
2.25.1
PATCH 5/6 drm/msm/disp: Remove redundant prefix in name of sub blocks
In reply to Ryan McCannWhen the registers of the csc and scaler sub blocks are printed during a device core dump, the sub block title is printed: <mainBlkName_sblkName>. Currently this appears as "sspp_0_sspp_csc" for a csc sub block to an SSPP main block named "sspp_0". Because the name of the sub block defined in the VIG_SBLK macro is "sspp_csc", the result is a redundant name. To avoid this redundancy, remove the duplicate prefix "sspp".
--- drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-)
drivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c +4 -4
@@ -257,10 +257,10 @@ static const uint32_t wb2_formats[] = { .maxdwnscale = MAX_DOWNSCALE_RATIO, \ .maxupscale = MAX_UPSCALE_RATIO, \ .smart_dma_priority = sdma_pri, \ - .scaler_blk = {.name = "sspp_scaler", \ + .scaler_blk = {.name = "scaler", \ .id = qseed_ver, \ .base = 0xa00, .len = 0xa0,}, \ - .csc_blk = {.name = "sspp_csc", \ + .csc_blk = {.name = "csc", \ .id = DPU_SSPP_CSC_10BIT, \ .base = 0x1a00, .len = 0x100,}, \ .format_list = plane_formats_yuv, \
@@ -275,10 +275,10 @@ static const uint32_t wb2_formats[] = { .maxdwnscale = MAX_DOWNSCALE_RATIO, \ .maxupscale = MAX_UPSCALE_RATIO, \ .smart_dma_priority = sdma_pri, \ - .scaler_blk = {.name = "sspp_scaler", \ + .scaler_blk = {.name = "scaler", \ .id = qseed_ver, \ .base = 0xa00, .len = 0xa0,}, \ - .csc_blk = {.name = "sspp_csc", \ + .csc_blk = {.name = "csc", \ .id = DPU_SSPP_CSC_10BIT, \ .base = 0x1a00, .len = 0x100,}, \ .format_list = plane_formats_yuv, \
Signature · 1 line
2.25.1
PATCH 6/6 drm/msm/dpu: Update dev core dump to dump registers of sub blocks
In reply to Ryan McCannCurrently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
--- drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 194 +++++++++++++++++++++++++++----- 1 file changed, 168 insertions(+), 26 deletions(-)
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +168 -26
@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms)return 0 ; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, +void __iomem *mmio,void *blk, +enum dpu_hw_blk_type blk_type) +{ +u32 base; + +switch (blk_type) { +case DPU_HW_BLK_TOP : + { +struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + +if (top->features &BIT (DPU_MDP_PERIPH_0_REMOVED )) { +msm_disp_snapshot_add_block (disp_state,MDP_PERIPH_TOP0 , + mmio + top->base,"top" ); +msm_disp_snapshot_add_block (disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base +MDP_PERIPH_TOP0_END , +"top_2" ); + }else { +msm_disp_snapshot_add_block (disp_state, top->len, mmio + top->base,"top" ); + } +break ; + } +case DPU_HW_BLK_LM : + { +struct dpu_lm_cfg *mixer = (struct dpu_lm_cfg *)blk; + +msm_disp_snapshot_add_block (disp_state, mixer->len, mmio + mixer->base,"%s" , + mixer->name); +break ; + } +case DPU_HW_BLK_CTL : + { +struct dpu_ctl_cfg *ctl = (struct dpu_ctl_cfg *)blk; + +msm_disp_snapshot_add_block (disp_state, ctl->len, mmio + ctl->base,"%s" , + ctl->name); +break ; + } +case DPU_HW_BLK_INTF : + { +struct dpu_intf_cfg *intf = (struct dpu_intf_cfg *)blk; + +msm_disp_snapshot_add_block (disp_state, intf->len, mmio + intf->base,"%s" , + intf->name); +break ; + } +case DPU_HW_BLK_WB : + { +struct dpu_wb_cfg *wb = (struct dpu_wb_cfg *)blk; + +msm_disp_snapshot_add_block (disp_state, wb->len, mmio + wb->base,"%s" , + wb->name); +break ; + } +case DPU_HW_BLK_SSPP : + { +struct dpu_sspp_cfg *sspp_block = (struct dpu_sspp_cfg *)blk; +const struct dpu_sspp_sub_blks *sblk = sspp_block->sblk; + + base = sspp_block->base; + +msm_disp_snapshot_add_block (disp_state, sspp_block->len, mmio + base,"%s" , + sspp_block->name); + +if (sspp_block->features &BIT (DPU_SSPP_SCALER_QSEED3 ) || + sspp_block->features &BIT (DPU_SSPP_SCALER_QSEED3LITE ) || + sspp_block->features &BIT (DPU_SSPP_SCALER_QSEED4 )) +msm_disp_snapshot_add_block (disp_state, sblk->scaler_blk.len, + mmio + base + sblk->scaler_blk.base,"%s_%s" , + sspp_block->name, sblk->scaler_blk.name); + +if (sspp_block->features &BIT (DPU_SSPP_CSC ) || sspp_block->features + &BIT (DPU_SSPP_CSC_10BIT )) +msm_disp_snapshot_add_block (disp_state, sblk->csc_blk.len, + mmio + base + sblk->csc_blk.base,"%s_%s" , + sspp_block->name, sblk->csc_blk.name); +break ; + } +case DPU_HW_BLK_DSPP : + { +struct dpu_dspp_cfg *dspp_block = (struct dpu_dspp_cfg *)blk; + + base = dspp_block->base; + +msm_disp_snapshot_add_block (disp_state, dspp_block->len, mmio + base,"%s" , + dspp_block->name); + +if (dspp_block->features &BIT (DPU_DSPP_PCC )) +msm_disp_snapshot_add_block (disp_state, dspp_block->sblk->pcc.len, + mmio + base + dspp_block->sblk->pcc.base, +"%s_%s" , dspp_block->name, + dspp_block->sblk->pcc.name); +break ; + } +case DPU_HW_BLK_PINGPONG : + { +struct dpu_pingpong_cfg *pingpong_block = (struct dpu_pingpong_cfg *)blk; +const struct dpu_pingpong_sub_blks *sblk = pingpong_block->sblk; + + base = pingpong_block->base; + +msm_disp_snapshot_add_block (disp_state, pingpong_block->len, mmio + base,"%s" , + pingpong_block->name); + +if (pingpong_block->features &BIT (DPU_PINGPONG_TE2 )) +msm_disp_snapshot_add_block (disp_state, sblk->te2.len, + mmio + base + sblk->te2.base,"%s_%s" , + pingpong_block->name, sblk->te2.name); + +if (pingpong_block->features &BIT (DPU_PINGPONG_DITHER )) +msm_disp_snapshot_add_block (disp_state, sblk->dither.len, + mmio + base + sblk->dither.base,"%s_%s" , + pingpong_block->name, sblk->dither.name); +break ; + } +case DPU_HW_BLK_DSC : + { +struct dpu_dsc_cfg *dsc_block = (struct dpu_dsc_cfg *)blk; + + base = dsc_block->base; + +if (dsc_block->features &BIT (DPU_DSC_HW_REV_1_2 )) { +struct dpu_dsc_blk enc = dsc_block->sblk->enc; +struct dpu_dsc_blk ctl = dsc_block->sblk->ctl; + +/* For now, pass in a length of 0 because the DSC_BLK register space +* overlaps with the sblks' register space. +* +* TODO: Pass in a length of 0 t0 DSC_BLK_1_2 in the HW catalog where +* applicable. +*/ +msm_disp_snapshot_add_block (disp_state,0 , mmio + base,"%s" , + dsc_block->name); +msm_disp_snapshot_add_block (disp_state, enc.len, mmio + base + enc.base, +"%s_%s" , dsc_block->name, enc.name); +msm_disp_snapshot_add_block (disp_state, ctl.len, mmio + base + ctl.base, +"%s_%s" , dsc_block->name, ctl.name); + }else { +msm_disp_snapshot_add_block (disp_state, dsc_block->len, mmio + base,"%s" , + dsc_block->name); + } +break ; + } +default : +DPU_ERROR ("Block type not supported." ); + } +} +static void dpu_kms_mdp_snapshot (struct msm_disp_state *disp_state,struct msm_kms *kms) {int i;
@@ -899,53 +1047,47 @@ static void dpu_kms_mdp_snapshot(struct msm_disp_state *disp_state, struct msm_k/* dump CTL sub-blocks HW regs info */ for (i =0 ; i < cat->ctl_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->ctl[i].len, - dpu_kms->mmio + cat->ctl[i].base,"ctl_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->ctl[i], +DPU_HW_BLK_CTL );/* dump DSPP sub-blocks HW regs info */ for (i =0 ; i < cat->dspp_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->dspp[i].len, - dpu_kms->mmio + cat->dspp[i].base,"dspp_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->dspp[i], +DPU_HW_BLK_DSPP );/* dump INTF sub-blocks HW regs info */ for (i =0 ; i < cat->intf_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->intf[i].len, - dpu_kms->mmio + cat->intf[i].base,"intf_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->intf[i], +DPU_HW_BLK_INTF );/* dump PP sub-blocks HW regs info */ for (i =0 ; i < cat->pingpong_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->pingpong[i].len, - dpu_kms->mmio + cat->pingpong[i].base,"pingpong_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->pingpong[i], +DPU_HW_BLK_PINGPONG );/* dump SSPP sub-blocks HW regs info */ for (i =0 ; i < cat->sspp_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->sspp[i].len, - dpu_kms->mmio + cat->sspp[i].base,"sspp_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->sspp[i], +DPU_HW_BLK_SSPP );/* dump LM sub-blocks HW regs info */ for (i =0 ; i < cat->mixer_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->mixer[i].len, - dpu_kms->mmio + cat->mixer[i].base,"lm_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->mixer[i], +DPU_HW_BLK_LM );/* dump WB sub-blocks HW regs info */ for (i =0 ; i < cat->wb_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->wb[i].len, - dpu_kms->mmio + cat->wb[i].base,"wb_%d" , i); - -if (cat->mdp[0 ].features &BIT (DPU_MDP_PERIPH_0_REMOVED )) { -msm_disp_snapshot_add_block (disp_state,MDP_PERIPH_TOP0 , - dpu_kms->mmio + cat->mdp[0 ].base,"top" ); -msm_disp_snapshot_add_block (disp_state, cat->mdp[0 ].len -MDP_PERIPH_TOP0_END , - dpu_kms->mmio + cat->mdp[0 ].base +MDP_PERIPH_TOP0_END ,"top_2" ); - }else { -msm_disp_snapshot_add_block (disp_state, cat->mdp[0 ].len, - dpu_kms->mmio + cat->mdp[0 ].base,"top" ); - } +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->wb[i], +DPU_HW_BLK_WB ); + +/* dump top block */ +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->mdp[0 ], +DPU_HW_BLK_TOP );/* dump DSC sub-blocks HW regs info */ for (i =0 ; i < cat->dsc_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->dsc[i].len, - dpu_kms->mmio + cat->dsc[i].base,"dsc_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->dsc[i], +DPU_HW_BLK_DSC );pm_runtime_put_sync (&dpu_kms->pdev->dev); }
Signature · 1 line
2.25.1
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
Quoted 14 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>--- drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 194 +++++++++++++++++++++++++++----- 1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms *kms) · line 890
Quoted 6 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms)return 0 ; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, +void __iomem *mmio,void *blk, +enum dpu_hw_blk_type blk_type)
No. Such multiplexers add no value to the code. Please inline it.
Not to mention that this patch is hard to review. You both move existing code and add new features. If it were to go, it should have been split into two patches: one introducing the multiplexer and another one adding subblocks.
Quoted 57 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +57
-0+{ +u32 base; + +switch (blk_type) { +case DPU_HW_BLK_TOP : + { +struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + +if (top->features &BIT (DPU_MDP_PERIPH_0_REMOVED )) { +msm_disp_snapshot_add_block (disp_state,MDP_PERIPH_TOP0 , + mmio + top->base,"top" ); +msm_disp_snapshot_add_block (disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base +MDP_PERIPH_TOP0_END , +"top_2" ); + }else { +msm_disp_snapshot_add_block (disp_state, top->len, mmio + top->base,"top" ); + } +break ; + } +case DPU_HW_BLK_LM : + { +struct dpu_lm_cfg *mixer = (struct dpu_lm_cfg *)blk; + +msm_disp_snapshot_add_block (disp_state, mixer->len, mmio + mixer->base,"%s" , + mixer->name); +break ; + } +case DPU_HW_BLK_CTL : + { +struct dpu_ctl_cfg *ctl = (struct dpu_ctl_cfg *)blk; + +msm_disp_snapshot_add_block (disp_state, ctl->len, mmio + ctl->base,"%s" , + ctl->name); +break ; + } +case DPU_HW_BLK_INTF : + { +struct dpu_intf_cfg *intf = (struct dpu_intf_cfg *)blk; + +msm_disp_snapshot_add_block (disp_state, intf->len, mmio + intf->base,"%s" , + intf->name); +break ; + } +case DPU_HW_BLK_WB : + { +struct dpu_wb_cfg *wb = (struct dpu_wb_cfg *)blk; + +msm_disp_snapshot_add_block (disp_state, wb->len, mmio + wb->base,"%s" , + wb->name); +break ; + } +case DPU_HW_BLK_SSPP : + { +struct dpu_sspp_cfg *sspp_block = (struct dpu_sspp_cfg *)blk; +const struct dpu_sspp_sub_blks *sblk = sspp_block->sblk; + + base = sspp_block->base;
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms *kms) · line 957
Quoted 10 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +10
-0+ +msm_disp_snapshot_add_block (disp_state, sspp_block->len, mmio + base,"%s" , + sspp_block->name); + +if (sspp_block->features &BIT (DPU_SSPP_SCALER_QSEED3 ) || + sspp_block->features &BIT (DPU_SSPP_SCALER_QSEED3LITE ) || + sspp_block->features &BIT (DPU_SSPP_SCALER_QSEED4 )) +msm_disp_snapshot_add_block (disp_state, sblk->scaler_blk.len, + mmio + base + sblk->scaler_blk.base,"%s_%s" , + sspp_block->name, sblk->scaler_blk.name);
Actually, it would be better to:
- drop name from all sblk instances (and use known string instead of the sblk name here)
- Use sblk->foo_blk.len to check if it should be printed or not.
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms *kms) · line 960
Quoted 3 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0+ +if (sspp_block->features &BIT (DPU_SSPP_CSC ) || sspp_block->features + &BIT (DPU_SSPP_CSC_10BIT ))
A very bad use of indentation. In future please split logically rather than just filling the line up to the line width.
Quoted 47 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +47
-0+msm_disp_snapshot_add_block (disp_state, sblk->csc_blk.len, + mmio + base + sblk->csc_blk.base,"%s_%s" , + sspp_block->name, sblk->csc_blk.name); +break ; + } +case DPU_HW_BLK_DSPP : + { +struct dpu_dspp_cfg *dspp_block = (struct dpu_dspp_cfg *)blk; + + base = dspp_block->base; + +msm_disp_snapshot_add_block (disp_state, dspp_block->len, mmio + base,"%s" , + dspp_block->name); + +if (dspp_block->features &BIT (DPU_DSPP_PCC )) +msm_disp_snapshot_add_block (disp_state, dspp_block->sblk->pcc.len, + mmio + base + dspp_block->sblk->pcc.base, +"%s_%s" , dspp_block->name, + dspp_block->sblk->pcc.name); +break ; + } +case DPU_HW_BLK_PINGPONG : + { +struct dpu_pingpong_cfg *pingpong_block = (struct dpu_pingpong_cfg *)blk; +const struct dpu_pingpong_sub_blks *sblk = pingpong_block->sblk; + + base = pingpong_block->base; + +msm_disp_snapshot_add_block (disp_state, pingpong_block->len, mmio + base,"%s" , + pingpong_block->name); + +if (pingpong_block->features &BIT (DPU_PINGPONG_TE2 )) +msm_disp_snapshot_add_block (disp_state, sblk->te2.len, + mmio + base + sblk->te2.base,"%s_%s" , + pingpong_block->name, sblk->te2.name); + +if (pingpong_block->features &BIT (DPU_PINGPONG_DITHER )) +msm_disp_snapshot_add_block (disp_state, sblk->dither.len, + mmio + base + sblk->dither.base,"%s_%s" , + pingpong_block->name, sblk->dither.name); +break ; + } +case DPU_HW_BLK_DSC : + { +struct dpu_dsc_cfg *dsc_block = (struct dpu_dsc_cfg *)blk; + + base = dsc_block->base;
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms *kms) · line 1017
Quoted 10 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +10
-0+ +if (dsc_block->features &BIT (DPU_DSC_HW_REV_1_2 )) { +struct dpu_dsc_blk enc = dsc_block->sblk->enc; +struct dpu_dsc_blk ctl = dsc_block->sblk->ctl; + +/* For now, pass in a length of 0 because the DSC_BLK register space +* overlaps with the sblks' register space. +* +* TODO: Pass in a length of 0 t0 DSC_BLK_1_2 in the HW catalog where +* applicable.
Nice catch, thank you. We should fix that.
Quoted 99 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +18
-0+*/ +msm_disp_snapshot_add_block (disp_state,0 , mmio + base,"%s" , + dsc_block->name); +msm_disp_snapshot_add_block (disp_state, enc.len, mmio + base + enc.base, +"%s_%s" , dsc_block->name, enc.name); +msm_disp_snapshot_add_block (disp_state, ctl.len, mmio + base + ctl.base, +"%s_%s" , dsc_block->name, ctl.name); + }else { +msm_disp_snapshot_add_block (disp_state, dsc_block->len, mmio + base,"%s" , + dsc_block->name); + } +break ; + } +default : +DPU_ERROR ("Block type not supported." ); + } +} +static void dpu_kms_mdp_snapshot(struct msm_disp_state *disp_state, struct msm_kms *kms) { int i;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +20
-26@@ -899,53 +1047,47 @@ static void dpu_kms_mdp_snapshot(struct msm_disp_state *disp_state, struct msm_k/* dump CTL sub-blocks HW regs info */ for (i =0 ; i < cat->ctl_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->ctl[i].len, - dpu_kms->mmio + cat->ctl[i].base,"ctl_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->ctl[i], +DPU_HW_BLK_CTL );/* dump DSPP sub-blocks HW regs info */ for (i =0 ; i < cat->dspp_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->dspp[i].len, - dpu_kms->mmio + cat->dspp[i].base,"dspp_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->dspp[i], +DPU_HW_BLK_DSPP );/* dump INTF sub-blocks HW regs info */ for (i =0 ; i < cat->intf_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->intf[i].len, - dpu_kms->mmio + cat->intf[i].base,"intf_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->intf[i], +DPU_HW_BLK_INTF );/* dump PP sub-blocks HW regs info */ for (i =0 ; i < cat->pingpong_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->pingpong[i].len, - dpu_kms->mmio + cat->pingpong[i].base,"pingpong_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->pingpong[i], +DPU_HW_BLK_PINGPONG );/* dump SSPP sub-blocks HW regs info */ for (i =0 ; i < cat->sspp_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->sspp[i].len, - dpu_kms->mmio + cat->sspp[i].base,"sspp_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->sspp[i], +DPU_HW_BLK_SSPP );/* dump LM sub-blocks HW regs info */ for (i =0 ; i < cat->mixer_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->mixer[i].len, - dpu_kms->mmio + cat->mixer[i].base,"lm_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->mixer[i], +DPU_HW_BLK_LM );/* dump WB sub-blocks HW regs info */ for (i =0 ; i < cat->wb_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->wb[i].len, - dpu_kms->mmio + cat->wb[i].base,"wb_%d" , i); - -if (cat->mdp[0 ].features &BIT (DPU_MDP_PERIPH_0_REMOVED )) { -msm_disp_snapshot_add_block (disp_state,MDP_PERIPH_TOP0 , - dpu_kms->mmio + cat->mdp[0 ].base,"top" ); -msm_disp_snapshot_add_block (disp_state, cat->mdp[0 ].len -MDP_PERIPH_TOP0_END , - dpu_kms->mmio + cat->mdp[0 ].base +MDP_PERIPH_TOP0_END ,"top_2" ); - }else { -msm_disp_snapshot_add_block (disp_state, cat->mdp[0 ].len, - dpu_kms->mmio + cat->mdp[0 ].base,"top" ); - } +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->wb[i], +DPU_HW_BLK_WB ); + +/* dump top block */ +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->mdp[0 ], +DPU_HW_BLK_TOP );/* dump DSC sub-blocks HW regs info */ for (i =0 ; i < cat->dsc_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->dsc[i].len, - dpu_kms->mmio + cat->dsc[i].base,"dsc_%d" , i); +dpu_kms_mdp_snapshot_add_block (disp_state, dpu_kms->mmio, (void *)&cat->dsc[i], +DPU_HW_BLK_DSC );pm_runtime_put_sync (&dpu_kms->pdev->dev); }
Signature · 2 lines
With best wishes Dmitry
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
Quoted 25 lines · Dmitry Baryshkov
Quoted 15 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>---
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c| 194
+++++++++++++++++++++++++++-----1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms *kms) · line 890
Quoted 6 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms) return 0; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, + void __iomem *mmio, void *blk, + enum dpu_hw_blk_type blk_type)No. Such multiplexers add no value to the code. Please inline it.
Quoted 2 lines
Not to mention that this patch is hard to review. You both move existing code and add new features. If it were to go, it should have been split into two patches: one introducing the multiplexer and another one adding subblocks.
Ok. we can split this into:
1) adding the multiplexer
2) adding sub-blk parsing support inside the multiplexer
Quoted 73 lines
Quoted 73 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +32
-0+{ + u32 base; + + switch (blk_type) { + case DPU_HW_BLK_TOP: + { + struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + + if (top->features & BIT(DPU_MDP_PERIPH_0_REMOVED)) { + msm_disp_snapshot_add_block(disp_state, MDP_PERIPH_TOP0, + mmio + top->base, "top"); + msm_disp_snapshot_add_block(disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base + MDP_PERIPH_TOP0_END, + "top_2"); + } else { + msm_disp_snapshot_add_block(disp_state, top->len, mmio + top->base,"top" ); + } + break; + } + case DPU_HW_BLK_LM: + { + struct dpu_lm_cfg *mixer = (struct dpu_lm_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, mixer->len, mmio + mixer->base,"%s" , + mixer->name); + break; + } + case DPU_HW_BLK_CTL: + { + struct dpu_ctl_cfg *ctl = (struct dpu_ctl_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, ctl->len, mmio +
ctl->base, "%s",
+ctl->name);
+ break;
+ }
+ caseDPU_HW_BLK_INTF:
+ {
+ structdpu_intf_cfg*intf = (structdpu_intf_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,intf->len, mmio +intf->base, "%s",
+intf->name);
+ break;
+ }
+ caseDPU_HW_BLK_WB:
+ {
+ structdpu_wb_cfg*wb = (structdpu_wb_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,wb->len, mmio +wb->base, "%s",
+wb->name);
+ break;
+ }
+ caseDPU_HW_BLK_SSPP:
+ {
+ structdpu_sspp_cfg*sspp_block= (structdpu_sspp_cfg*)blk; + const structdpu_sspp_sub_blks*sblk =sspp_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +11
-0+ + base = sspp_block->base; + + msm_disp_snapshot_add_block(disp_state, sspp_block->len, mmio + base, "%s", + sspp_block->name); + + if (sspp_block->features & BIT(DPU_SSPP_SCALER_QSEED3) || + sspp_block->features & BIT(DPU_SSPP_SCALER_QSEED3LITE) || + sspp_block->features & BIT(DPU_SSPP_SCALER_QSEED4)) + msm_disp_snapshot_add_block(disp_state,
sblk->scaler_blk.len,
Quoted 9 lines
Quoted 3 lines
+ mmio + base +
sblk->scaler_blk.base, "%s_%s",
+sspp_block->name,sblk->scaler_blk.name);Actually, it would be better to:
- drop name from all sblk instances (and use known string instead of the sblk name here)
- Usesblk->foo_blk.lento check if it should be printed or not.
No, I dont agree. If we drop the names from the sub_blk in the catalog, we will end up using "sub_blk_name" string here in the code to indicate which blk that is in the dump.
If we add more sub_blks in the catalog in the future we need to keep changing the code over here. Thats not how it should be.
Leaving the names in the catalog ensures that this code wont change and only catalog changes when we add a new sub_blk either for an existing or new chipset.
catalog is indicating the new blk, and dumping code just prints it.
with your approach, dumping code will or can keep changing with chipsets or sub_blks. Thats not how it should be.
Quoted 64 lines
Quoted 4 lines
+
+ if (sspp_block->features&BIT(DPU_SSPP_CSC)||sspp_block->features
+ &BIT(DPU_SSPP_CSC_10BIT))A very bad use of indentation. In future please split logically rather than just filling the line up to the line width.
Quoted 57 lines
+ msm_disp_snapshot_add_block(disp_state,
sblk->csc_blk.len, + mmio + base +sblk->csc_blk.base, "%s_%s", +sspp_block->name,sblk->csc_blk.name); + break;
+ }
+ caseDPU_HW_BLK_DSPP:
+ {
+ structdpu_dspp_cfg*dspp_block= (structdpu_dspp_cfg*)blk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +9
-0+ + base = dspp_block->base; + + msm_disp_snapshot_add_block(disp_state, dspp_block->len, mmio + base, "%s", + dspp_block->name); + + if (dspp_block->features & BIT(DPU_DSPP_PCC)) + msm_disp_snapshot_add_block(disp_state,
dspp_block->sblk->pcc.len,
+ mmio + base +dspp_block->sblk->pcc.base, + "%s_%s",dspp_block->name,
+dspp_block->sblk->pcc.name);
+ break;
+ }
+ caseDPU_HW_BLK_PINGPONG:
+ {
+ structdpu_pingpong_cfg*pingpong_block= (structdpu_pingpong_cfg*)blk;
+ const structdpu_pingpong_sub_blks*sblk =pingpong_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = pingpong_block->base; + + msm_disp_snapshot_add_block(disp_state, pingpong_block->len,mmio + base, "%s",
+pingpong_block->name);drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +23
-0+ + if (pingpong_block->features & BIT(DPU_PINGPONG_TE2)) + msm_disp_snapshot_add_block(disp_state, sblk->te2.len, + mmio + base + sblk->te2.base, "%s_%s", + pingpong_block->name, sblk->te2.name); + + if (pingpong_block->features & BIT(DPU_PINGPONG_DITHER)) + msm_disp_snapshot_add_block(disp_state, sblk->dither.len, + mmio + base + sblk->dither.base, "%s_%s", + pingpong_block->name, sblk->dither.name); + break; + } + case DPU_HW_BLK_DSC: + { + struct dpu_dsc_cfg *dsc_block = (struct dpu_dsc_cfg *)blk; + + base = dsc_block->base; + + if (dsc_block->features & BIT(DPU_DSC_HW_REV_1_2)) { + struct dpu_dsc_blk enc = dsc_block->sblk->enc; + struct dpu_dsc_blk ctl = dsc_block->sblk->ctl; + + /* For now, pass in a length of 0 because the DSC_BLK
Quoted 9 lines
Quoted 6 lines
register space
+ * overlaps with the sblks' register space. + *
+ * TODO: Pass in a length of 0 t0DSC_BLK_1_2in the HW catalog where
+ * applicable.Nice catch, thank you. We should fix that.
Yes and we would have fixed that ourself if you wanted that with this series as another patch.
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
Quoted 29 lines · Abhinav Kumar
Quoted 26 lines · Dmitry Baryshkov
Quoted 15 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>---
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c| 194
+++++++++++++++++++++++++++-----1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms · line 891
Quoted 7 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms) return 0; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, + void __iomem *mmio, void *blk, + enum dpu_hw_blk_type blk_type)No. Such multiplexers add no value to the code. Please inline it.
Quoted 88 lines
Quoted 2 lines
Not to mention that this patch is hard to review. You both move existing code and add new features. If it were to go, it should have been split into two patches: one introducing the multiplexer and another one adding subblocks.
Ok. we can split this into:
1) adding the multiplexer
2) adding sub-blk parsing support inside the multiplexerQuoted 80 lines
Quoted 68 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +24
-0+{ + u32 base; + + switch (blk_type) { + case DPU_HW_BLK_TOP: + { + struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + + if (top->features & BIT(DPU_MDP_PERIPH_0_REMOVED)) { + msm_disp_snapshot_add_block(disp_state, MDP_PERIPH_TOP0, + mmio + top->base, "top"); + msm_disp_snapshot_add_block(disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base + MDP_PERIPH_TOP0_END, + "top_2"); + } else { + msm_disp_snapshot_add_block(disp_state, top->len, mmio + top->base,"top" ); + } + break; + } + case DPU_HW_BLK_LM: + { + struct dpu_lm_cfg *mixer = (struct dpu_lm_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, mixer->len, mmio +
mixer->base, "%s",
+mixer->name);
+ break;
+ }
+ caseDPU_HW_BLK_CTL:
+ {
+ structdpu_ctl_cfg*ctl = (structdpu_ctl_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,ctl->len, mmio +ctl->base, "%s",
+ctl->name);
+ break;
+ }
+ caseDPU_HW_BLK_INTF:
+ {
+ structdpu_intf_cfg*intf = (structdpu_intf_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,intf->len, mmio +intf->base, "%s",
+intf->name);
+ break;
+ }
+ caseDPU_HW_BLK_WB:
+ {
+ structdpu_wb_cfg*wb = (structdpu_wb_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,wb->len, mmio +wb->base, "%s",
+wb->name);
+ break;
+ }
+ caseDPU_HW_BLK_SSPP:
+ {
+ structdpu_sspp_cfg*sspp_block= (structdpu_sspp_cfg*)blk; + const structdpu_sspp_sub_blks*sblk =sspp_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = sspp_block->base; + + msm_disp_snapshot_add_block(disp_state, sspp_block->len,mmio + base, "%s",
+sspp_block->name);
+Quoted 8 lines
+ if (
sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3LITE)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED4)) + msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len,
+ mmio + base +sblk->scaler_blk.base, "%s_%s",
+sspp_block->name,sblk->scaler_blk.name);Actually, it would be better to:
- drop name from all sblk instances (and use known string instead of the sblk name here)
Hey Dmitry,
FWIW, I second Abhinav's points about the sblk names. For example, if in the future we want to add a "_rot" suffix specifically to the VIG_SBLK_ROT.scaler name, it would be easier to just make that change in the HW catalog.
Quoted 1 line
Quoted 1 line
- Use
sblk->foo_blk.lento check if it should be printed or not.
From my understanding, your suggestion is to replace the feature flag checks with a sblk.len > 0 check.
I don't think that would be good because it wouldn't be correct to assume that the sblk will always be present. For example, for DPU_HW_BLK_DSC, the sblks will only be present for DSC_BLK_1_2.
In addition, it is possible for sblks, like pp_sblk_te.te2, to have a len of 0. While the register space of that specific sblk will not be printed, I'd prefer the devcore dump to reflect what is present within the HW catalog so that the user knows which pingpong blks have the TE2 sblk.
Thanks,
Jessica Zhang
Quoted 89 lines
Quoted 1 line
No, I dont agree. If we drop the names from the
sub_blkin the catalog, we will end up using "sub_blk_name" string here in the code to indicate which blk that is in the dump.If we add more
sub_blksin the catalog in the future we need to keep changing the code over here. Thats not how it should be.Leaving the names in the catalog ensures that this code wont change and only catalog changes when we add a new
sub_blkeither for an existing or new chipset.catalog is indicating the new blk, and dumping code just prints it.
with your approach, dumping code will or can keep changing with chipsets or
sub_blks. Thats not how it should be.Quoted 65 lines
Quoted 4 lines
+
+ if (sspp_block->features&BIT(DPU_SSPP_CSC)||sspp_block->features
+ &BIT(DPU_SSPP_CSC_10BIT))A very bad use of indentation. In future please split logically rather than just filling the line up to the line width.
Quoted 58 lines
+ msm_disp_snapshot_add_block(disp_state,
sblk->csc_blk.len, + mmio + base +sblk->csc_blk.base, "%s_%s", +sspp_block->name,sblk->csc_blk.name); + break;
+ }
+ caseDPU_HW_BLK_DSPP:
+ {
+ structdpu_dspp_cfg*dspp_block= (structdpu_dspp_cfg*)blk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = dspp_block->base; + + msm_disp_snapshot_add_block(disp_state, dspp_block->len,mmio + base, "%s",
+dspp_block->name);
+
+ if (dspp_block->features&BIT(DPU_DSPP_PCC))
+ msm_disp_snapshot_add_block(disp_state,dspp_block->sblk->pcc.len,
+ mmio + base +dspp_block->sblk->pcc.base, + "%s_%s",dspp_block->name,
+dspp_block->sblk->pcc.name);
+ break;
+ }
+ caseDPU_HW_BLK_PINGPONG:
+ {
+ structdpu_pingpong_cfg*pingpong_block= (structdpu_pingpong_cfg*)blk;
+ const structdpu_pingpong_sub_blks*sblk =pingpong_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = pingpong_block->base; + + msm_disp_snapshot_add_block(disp_state, pingpong_block->len,mmio + base, "%s",
+pingpong_block->name);drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +23
-0+ + if (pingpong_block->features & BIT(DPU_PINGPONG_TE2)) + msm_disp_snapshot_add_block(disp_state, sblk->te2.len, + mmio + base + sblk->te2.base, "%s_%s", + pingpong_block->name, sblk->te2.name); + + if (pingpong_block->features & BIT(DPU_PINGPONG_DITHER)) + msm_disp_snapshot_add_block(disp_state, sblk->dither.len, + mmio + base + sblk->dither.base, "%s_%s", + pingpong_block->name, sblk->dither.name); + break; + } + case DPU_HW_BLK_DSC: + { + struct dpu_dsc_cfg *dsc_block = (struct dpu_dsc_cfg *)blk; + + base = dsc_block->base; + + if (dsc_block->features & BIT(DPU_DSC_HW_REV_1_2)) { + struct dpu_dsc_blk enc = dsc_block->sblk->enc; + struct dpu_dsc_blk ctl = dsc_block->sblk->ctl; + + /* For now, pass in a length of 0 because the DSC_BLKQuoted 9 lines
Quoted 6 lines
register space
+ * overlaps with the sblks' register space. + *
+ * TODO: Pass in a length of 0 t0DSC_BLK_1_2in the HW catalog where
+ * applicable.Nice catch, thank you. We should fix that.
Yes and we would have fixed that ourself if you wanted that with this series as another patch.
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
Quoted 121 lines · Jessica Zhang
Quoted 29 lines · Abhinav Kumar
Quoted 26 lines · Dmitry Baryshkov
Quoted 15 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>---
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c| 194
+++++++++++++++++++++++++++-----1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms · line 891
Quoted 7 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms) return 0; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, + void __iomem *mmio, void *blk, + enum dpu_hw_blk_type blk_type)No. Such multiplexers add no value to the code. Please inline it.
Quoted 88 lines
Quoted 2 lines
Not to mention that this patch is hard to review. You both move existing code and add new features. If it were to go, it should have been split into two patches: one introducing the multiplexer and another one adding subblocks.
Ok. we can split this into:
1) adding the multiplexer
2) adding sub-blk parsing support inside the multiplexerQuoted 80 lines
Quoted 68 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +33
-0+{ + u32 base; + + switch (blk_type) { + case DPU_HW_BLK_TOP: + { + struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + + if (top->features & BIT(DPU_MDP_PERIPH_0_REMOVED)) { + msm_disp_snapshot_add_block(disp_state, MDP_PERIPH_TOP0, + mmio + top->base, "top"); + msm_disp_snapshot_add_block(disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base + MDP_PERIPH_TOP0_END, + "top_2"); + } else { + msm_disp_snapshot_add_block(disp_state, top->len, mmio + top->base, "top"); + } + break; + } + case DPU_HW_BLK_LM: + { + struct dpu_lm_cfg *mixer = (struct dpu_lm_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, mixer->len, mmio + mixer->base,"%s" , + mixer->name); + break; + } + case DPU_HW_BLK_CTL: + { + struct dpu_ctl_cfg *ctl = (struct dpu_ctl_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, ctl->len, mmio +
ctl->base, "%s",
+ctl->name);
+ break;
+ }
+ caseDPU_HW_BLK_INTF:
+ {
+ structdpu_intf_cfg*intf = (structdpu_intf_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,intf->len, mmio +intf->base, "%s",
+intf->name);
+ break;
+ }
+ caseDPU_HW_BLK_WB:
+ {
+ structdpu_wb_cfg*wb = (structdpu_wb_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,wb->len, mmio +wb->base, "%s",
+wb->name);
+ break;
+ }
+ caseDPU_HW_BLK_SSPP:
+ {
+ structdpu_sspp_cfg*sspp_block= (structdpu_sspp_cfg*)blk; + const structdpu_sspp_sub_blks*sblk =sspp_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = sspp_block->base; + + msm_disp_snapshot_add_block(disp_state, sspp_block->len,mmio + base, "%s",
+sspp_block->name);
+Quoted 8 lines
+ if (
sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3LITE)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED4)) + msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len,
+ mmio + base +sblk->scaler_blk.base, "%s_%s",
+sspp_block->name,sblk->scaler_blk.name);Actually, it would be better to:
- drop name from all sblk instances (and use known string instead of the sblk name here)
Quoted 3 lines
Hey Dmitry,
FWIW, I second Abhinav's points about the sblk names. For example, if in the future we want to add a "
_rot" suffix specifically to the VIG_SBLK_ROT.scaler name, it would be easier to just make that change in the HW catalog.
But why? The scaler is the same qseed3 scaler. We do not dump features, they are constant for the platform in question.
Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
- Use
sblk->foo_blk.lento check if it should be printed or not.From my understanding, your suggestion is to replace the feature flag
checks with a sblk.len > 0 check.I don't think that would be good because it wouldn't be correct to assume that the sblk will always be present. For example, for
DPU_HW_BLK_DSC, the sblks will only be present forDSC_BLK_1_2.
I don't consider sub-block as being always present. But if it present, it has non-zero length. If its length is zero, we have nothing to dump for it.
Quoted 1 line
In addition, it is possible for sblks, like pp_sblk_te.te2, to have a len of 0. While the register space of that specific sblk will not be printed, I'd prefer the devcore dump to reflect what is present within the HW catalog so that the user knows which pingpong blks have the TE2 sblk.
I'd consider this as dumping the feature instead of dumping the registers. If you think it is necessary to ease decoding of the dump, consider adding block.features to the dump instead.
Quoted 17 lines
Thanks,
Jessica Zhang
Quoted 12 lines
Quoted 1 line
No, I dont agree. If we drop the names from the
sub_blkin the catalog, we will end up using "sub_blk_name" string here in the code to indicate which blk that is in the dump.If we add more
sub_blksin the catalog in the future we need to keep changing the code over here. Thats not how it should be.Leaving the names in the catalog ensures that this code wont change and only catalog changes when we add a new
sub_blkeither for an existing or new chipset.catalog is indicating the new blk, and dumping code just prints it.
with your approach, dumping code will or can keep changing with chipsets or
sub_blks. Thats not how it should be.
Signature · 2 lines
With best wishes Dmitry
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
In reply to Abhinav KumarQuoted 29 lines · Abhinav Kumar
Quoted 26 lines · Dmitry Baryshkov
Quoted 15 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>---
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c| 194
+++++++++++++++++++++++++++-----1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms · line 891
Quoted 7 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms) return 0; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, + void __iomem *mmio, void *blk, + enum dpu_hw_blk_type blk_type)No. Such multiplexers add no value to the code. Please inline it.
Quoted 7 lines
Quoted 2 lines
Not to mention that this patch is hard to review. You both move existing code and add new features. If it were to go, it should have been split into two patches: one introducing the multiplexer and another one adding subblocks.
Ok. we can split this into:
1) adding the multiplexer
2) adding sub-blk parsing support inside the multiplexer
I'd say, drop the multiplexer completely. It adds no value here. It is only used from dpu_kms_mdp_snapshot(). If the code there was complex enough, it would have made sense to _split_ the function. But even in such case there would be no point in having multiplexer. We do not enumerate block by type.
Quoted 86 lines
Quoted 74 lines
Quoted 74 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +24
-0+{ + u32 base; + + switch (blk_type) { + case DPU_HW_BLK_TOP: + { + struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + + if (top->features & BIT(DPU_MDP_PERIPH_0_REMOVED)) { + msm_disp_snapshot_add_block(disp_state, MDP_PERIPH_TOP0, + mmio + top->base, "top"); + msm_disp_snapshot_add_block(disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base + MDP_PERIPH_TOP0_END, + "top_2"); + } else { + msm_disp_snapshot_add_block(disp_state, top->len, mmio + top->base,"top" ); + } + break; + } + case DPU_HW_BLK_LM: + { + struct dpu_lm_cfg *mixer = (struct dpu_lm_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, mixer->len, mmio +
mixer->base, "%s",
+mixer->name);
+ break;
+ }
+ caseDPU_HW_BLK_CTL:
+ {
+ structdpu_ctl_cfg*ctl = (structdpu_ctl_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,ctl->len, mmio +ctl->base, "%s",
+ctl->name);
+ break;
+ }
+ caseDPU_HW_BLK_INTF:
+ {
+ structdpu_intf_cfg*intf = (structdpu_intf_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,intf->len, mmio +intf->base, "%s",
+intf->name);
+ break;
+ }
+ caseDPU_HW_BLK_WB:
+ {
+ structdpu_wb_cfg*wb = (structdpu_wb_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,wb->len, mmio +wb->base, "%s",
+wb->name);
+ break;
+ }
+ caseDPU_HW_BLK_SSPP:
+ {
+ structdpu_sspp_cfg*sspp_block= (structdpu_sspp_cfg*)blk; + const structdpu_sspp_sub_blks*sblk =sspp_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = sspp_block->base; + + msm_disp_snapshot_add_block(disp_state, sspp_block->len,mmio + base, "%s",
+sspp_block->name);
+
+ if (sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3LITE)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED4)) + msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len,Quoted 9 lines
Quoted 3 lines
+ mmio + base +
sblk->scaler_blk.base, "%s_%s",
+sspp_block->name,sblk->scaler_blk.name);Actually, it would be better to:
- drop name from all sblk instances (and use known string instead of the sblk name here)
- Usesblk->foo_blk.lento check if it should be printed or not.No, I dont agree. If we drop the names from the
sub_blkin the catalog, we will end up using "sub_blk_name" string here in the code to indicate which blk that is in the dump.
Quoted 7 lines
If we add more
sub_blksin the catalog in the future we need to keep changing the code over here. Thats not how it should be.Leaving the names in the catalog ensures that this code wont change and only catalog changes when we add a new
sub_blkeither for an existing or new chipset.catalog is indicating the new blk, and dumping code just prints it.
with your approach, dumping code will or can keep changing with chipsets or
sub_blks. Thats not how it should be.
Well, we do not enumerate sub-blocks in any way, they are not indexed. So even with sblk->blk.name in place, adding new sub-block would require adding new code here. That's why I wrote that the calling code knows which sub-block it refers to.
Let me extract the relevant code (skipping all the conditions for now):
msm_disp_snapshot_add_block(disp_state, sspp_block->len, mmio + base, "%s",
sspp_block->name);
if (have_scaler)
msm_disp_snapshot_add_block(disp_state, sblk->scaler_blk.len, mmio + base + sblk->scaler_blk.base, "%s_%s", sspp_block->name, sblk->scaler_blk.name);
if (have_csc)
msm_disp_snapshot_add_block(disp_state, sblk->csc_blk.len, mmio + base + sblk->csc_blk.base, "%s_%s", sspp_block->name, sblk->csc_blk.name);
Consider adding new sub-block, "baz". We would still require manual addition of the following code:
msm_disp_snapshot_add_block(disp_state, sblk->baz_blk.len, mmio + base + sblk->baz_blk.base, "%s_%s", sspp_block->name, sblk->baz_blk.name);
Compare this with:
msm_disp_snapshot_add_block(disp_state, sblk->baz_blk.len, mmio + base + sblk->baz_blk.base, "%s_baz", sspp_block->name);
Moreover, if we follow the style of dpu_kms_mdp_snapshot() (which doesn't use name), it should be:
msm_disp_snapshot_add_block(disp_state, sblk->baz_blk.len, mmio + base + sblk->baz_blk.base, "sspp%d_baz", idx);
Quoted 76 lines
Quoted 66 lines
Quoted 5 lines
+
+ if (sspp_block->features&BIT(DPU_SSPP_CSC)||sspp_block->features
+ &BIT(DPU_SSPP_CSC_10BIT))A very bad use of indentation. In future please split logically rather than just filling the line up to the line width.
Quoted 58 lines
+ msm_disp_snapshot_add_block(disp_state,
sblk->csc_blk.len, + mmio + base +sblk->csc_blk.base, "%s_%s", +sspp_block->name,sblk->csc_blk.name); + break;
+ }
+ caseDPU_HW_BLK_DSPP:
+ {
+ structdpu_dspp_cfg*dspp_block= (structdpu_dspp_cfg*)blk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = dspp_block->base; + + msm_disp_snapshot_add_block(disp_state, dspp_block->len,mmio + base, "%s",
+dspp_block->name);
+
+ if (dspp_block->features&BIT(DPU_DSPP_PCC))
+ msm_disp_snapshot_add_block(disp_state,dspp_block->sblk->pcc.len,
+ mmio + base +dspp_block->sblk->pcc.base, + "%s_%s",dspp_block->name,
+dspp_block->sblk->pcc.name);
+ break;
+ }
+ caseDPU_HW_BLK_PINGPONG:
+ {
+ structdpu_pingpong_cfg*pingpong_block= (structdpu_pingpong_cfg*)blk;
+ const structdpu_pingpong_sub_blks*sblk =pingpong_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = pingpong_block->base; + + msm_disp_snapshot_add_block(disp_state, pingpong_block->len,mmio + base, "%s",
+pingpong_block->name);drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +23
-0+ + if (pingpong_block->features & BIT(DPU_PINGPONG_TE2)) + msm_disp_snapshot_add_block(disp_state, sblk->te2.len, + mmio + base + sblk->te2.base, "%s_%s", + pingpong_block->name, sblk->te2.name); + + if (pingpong_block->features & BIT(DPU_PINGPONG_DITHER)) + msm_disp_snapshot_add_block(disp_state, sblk->dither.len, + mmio + base + sblk->dither.base, "%s_%s", + pingpong_block->name, sblk->dither.name); + break; + } + case DPU_HW_BLK_DSC: + { + struct dpu_dsc_cfg *dsc_block = (struct dpu_dsc_cfg *)blk; + + base = dsc_block->base; + + if (dsc_block->features & BIT(DPU_DSC_HW_REV_1_2)) { + struct dpu_dsc_blk enc = dsc_block->sblk->enc; + struct dpu_dsc_blk ctl = dsc_block->sblk->ctl; + + /* For now, pass in a length of 0 because the DSC_BLKQuoted 9 lines
Quoted 6 lines
register space
+ * overlaps with the sblks' register space. + *
+ * TODO: Pass in a length of 0 t0DSC_BLK_1_2in the HW catalog where
+ * applicable.Nice catch, thank you. We should fix that.
Quoted 1 line
Yes and we would have fixed that ourself if you wanted that with this series as another patch.
This is not relevant to this series, so it should be fixed in a separate series.
Signature · 2 lines
With best wishes Dmitry
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
Quoted 33 lines · Dmitry Baryshkov
Quoted 32 lines · Abhinav Kumar
Quoted 28 lines · Dmitry Baryshkov
Quoted 15 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>---
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c| 194
+++++++++++++++++++++++++++-----1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms · line 891
Quoted 7 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms) return 0; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, + void __iomem *mmio, void *blk, + enum dpu_hw_blk_type blk_type)No. Such multiplexers add no value to the code. Please inline it.
Not to mention that this patch is hard to review. You both move existing code and add new features. If it were to go, it should have been split into two patches: one introducing the multiplexer and another one adding subblocks.
Quoted 7 lines
Quoted 4 lines
Ok. we can split this into:
1) adding the multiplexer
2) adding sub-blk parsing support inside the multiplexerI'd say, drop the multiplexer completely. It adds no value here. It is only used from
dpu_kms_mdp_snapshot(). If the code there was complex enough, it would have made sense to _split_ the function. But even in such case there would be no point in having multiplexer. We do not enumerate block by type.
Can you pls elaborate what you mean by enumerate blk by type?
We do have DPU_HW_BLK_***
Did you mean sub-blk?
Quoted 88 lines
Quoted 88 lines
Quoted 74 lines
Quoted 74 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +33
-0+{ + u32 base; + + switch (blk_type) { + case DPU_HW_BLK_TOP: + { + struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + + if (top->features & BIT(DPU_MDP_PERIPH_0_REMOVED)) { + msm_disp_snapshot_add_block(disp_state, MDP_PERIPH_TOP0, + mmio + top->base, "top"); + msm_disp_snapshot_add_block(disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base + MDP_PERIPH_TOP0_END, + "top_2"); + } else { + msm_disp_snapshot_add_block(disp_state, top->len, mmio + top->base, "top"); + } + break; + } + case DPU_HW_BLK_LM: + { + struct dpu_lm_cfg *mixer = (struct dpu_lm_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, mixer->len, mmio + mixer->base,"%s" , + mixer->name); + break; + } + case DPU_HW_BLK_CTL: + { + struct dpu_ctl_cfg *ctl = (struct dpu_ctl_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, ctl->len, mmio +
ctl->base, "%s",
+ctl->name);
+ break;
+ }
+ caseDPU_HW_BLK_INTF:
+ {
+ structdpu_intf_cfg*intf = (structdpu_intf_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,intf->len, mmio +intf->base, "%s",
+intf->name);
+ break;
+ }
+ caseDPU_HW_BLK_WB:
+ {
+ structdpu_wb_cfg*wb = (structdpu_wb_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,wb->len, mmio +wb->base, "%s",
+wb->name);
+ break;
+ }
+ caseDPU_HW_BLK_SSPP:
+ {
+ structdpu_sspp_cfg*sspp_block= (structdpu_sspp_cfg*)blk; + const structdpu_sspp_sub_blks*sblk =sspp_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = sspp_block->base; + + msm_disp_snapshot_add_block(disp_state, sspp_block->len,mmio + base, "%s",
+sspp_block->name);
+
+ if (sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3LITE)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED4)) + msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len,Quoted 9 lines
Quoted 3 lines
+ mmio + base +
sblk->scaler_blk.base, "%s_%s",
+sspp_block->name,sblk->scaler_blk.name);Actually, it would be better to:
- drop name from all sblk instances (and use known string instead of the sblk name here)
- Usesblk->foo_blk.lento check if it should be printed or not.No, I dont agree. If we drop the names from the
sub_blkin the catalog, we will end up using "sub_blk_name" string here in the code to indicate which blk that is in the dump.If we add more
sub_blksin the catalog in the future we need to keep changing the code over here. Thats not how it should be.
Quoted 8 lines
Quoted 5 lines
Leaving the names in the catalog ensures that this code wont change and only catalog changes when we add a new
sub_blkeither for an existing or new chipset.catalog is indicating the new blk, and dumping code just prints it.
with your approach, dumping code will or can keep changing with chipsets or
sub_blks. Thats not how it should be.Well, we do not enumerate sub-blocks in any way, they are not indexed. So even with
sblk->blk.namein place, adding new sub-block would require adding new code here. That's why I wrote that the calling code knows which sub-block it refers to.
Today, unfortunately each sub_blk type is different so we have to do this case by case.
Ideally, this should have just been
-> print main blk
-> print all sub-blks of the main blk
Without having to handle each main blk's sub-blks separately.
That way the dumping code would have remained generic without having to do even the multiplexer in the first place.
Need to explore if somehow we can come up with a generic sub-blk struct and make this possible. Then this code will become much easier and what I am saying will make total sense.
Even without that, conceptually these sub-blk names are reflecting whats in our software document. So its not a random name but reflects the actual sub-blk name from the hardware. So this belongs in the catalog.
Dumping code should not change or know whats the name of each block. It should just use whats in the catalog. thats why even conceptually I am not okay with your idea.
Quoted 22 lines
Let me extract the relevant code (skipping all the conditions for now):
msm_disp_snapshot_add_block(disp_state,
sspp_block->len, mmio + base, "%s",
sspp_block->name);if (
have_scaler)
msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len, mmio + base +sblk->scaler_blk.base, "%s_%s",sspp_block->name,sblk->scaler_blk.name);if (
have_csc)
msm_disp_snapshot_add_block(disp_state,sblk->csc_blk.len, mmio + base +sblk->csc_blk.base, "%s_%s",sspp_block->name,sblk->csc_blk.name);Consider adding new sub-block, "baz". We would still require manual addition of the following code:
msm_disp_snapshot_add_block(disp_state,
sblk->baz_blk.len, mmio + base +sblk->baz_blk.base, "%s_%s",sspp_block->name,sblk->baz_blk.name);
Quoted 6 lines
Compare this with:
msm_disp_snapshot_add_block(disp_state,
sblk->baz_blk.len, mmio + base +sblk->baz_blk.base, "%s_baz",sspp_block->name);
Basically you are saying why not make the one line change here instead of using the name from the catalog.
I think it will be better to use from the catalog for the reason I wrote above that dumping code should just "use" the catalog's information and not become a catalog itself.
You are not saving much by dropping the sub-blk name from catalog anyway.
Quoted 6 lines
Moreover, if we follow the style of
dpu_kms_mdp_snapshot()(which doesn't use name), it should be:msm_disp_snapshot_add_block(disp_state,
sblk->baz_blk.len, mmio + base +sblk->baz_blk.base, "sspp%d_baz", idx);
tbh, after looking at this series, it made me think why I didnt use the name from the catalog even for the dpu_kms_mdp_snapshot()
Quoted 2 lines
Quoted 2 lines
<snipped>
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
Quoted 151 lines · Abhinav Kumar
Quoted 33 lines · Dmitry Baryshkov
Quoted 32 lines · Abhinav Kumar
Quoted 28 lines · Dmitry Baryshkov
Quoted 15 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>---
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c| 194
+++++++++++++++++++++++++++-----1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms · line 891
Quoted 7 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms) return 0; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, + void __iomem *mmio, void *blk, + enum dpu_hw_blk_type blk_type)No. Such multiplexers add no value to the code. Please inline it.
Not to mention that this patch is hard to review. You both move existing code and add new features. If it were to go, it should have been split into two patches: one introducing the multiplexer and another one adding subblocks.
Quoted 7 lines
Quoted 4 lines
Ok. we can split this into:
1) adding the multiplexer
2) adding sub-blk parsing support inside the multiplexerI'd say, drop the multiplexer completely. It adds no value here. It is only used from
dpu_kms_mdp_snapshot(). If the code there was complex enough, it would have made sense to _split_ the function. But even in such case there would be no point in having multiplexer. We do not enumerate block by type.Can you pls elaborate what you mean by enumerate blk by type?
We do have DPU_HW_BLK_***
Did you mean sub-blk?
Quoted 88 lines
Quoted 88 lines
Quoted 74 lines
Quoted 74 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +33
-0+{ + u32 base; + + switch (blk_type) { + case DPU_HW_BLK_TOP: + { + struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + + if (top->features & BIT(DPU_MDP_PERIPH_0_REMOVED)) { + msm_disp_snapshot_add_block(disp_state, MDP_PERIPH_TOP0, + mmio + top->base, "top"); + msm_disp_snapshot_add_block(disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base + MDP_PERIPH_TOP0_END, + "top_2"); + } else { + msm_disp_snapshot_add_block(disp_state, top->len, mmio + top->base, "top"); + } + break; + } + case DPU_HW_BLK_LM: + { + struct dpu_lm_cfg *mixer = (struct dpu_lm_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, mixer->len, mmio + mixer->base,"%s" , + mixer->name); + break; + } + case DPU_HW_BLK_CTL: + { + struct dpu_ctl_cfg *ctl = (struct dpu_ctl_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, ctl->len, mmio +
ctl->base, "%s",
+ctl->name);
+ break;
+ }
+ caseDPU_HW_BLK_INTF:
+ {
+ structdpu_intf_cfg*intf = (structdpu_intf_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,intf->len, mmio +intf->base, "%s",
+intf->name);
+ break;
+ }
+ caseDPU_HW_BLK_WB:
+ {
+ structdpu_wb_cfg*wb = (structdpu_wb_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,wb->len, mmio +wb->base, "%s",
+wb->name);
+ break;
+ }
+ caseDPU_HW_BLK_SSPP:
+ {
+ structdpu_sspp_cfg*sspp_block= (structdpu_sspp_cfg*)blk; + const structdpu_sspp_sub_blks*sblk =sspp_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = sspp_block->base; + + msm_disp_snapshot_add_block(disp_state, sspp_block->len,mmio + base, "%s",
+sspp_block->name);
+
+ if (sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3LITE)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED4)) + msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len,Quoted 9 lines
Quoted 3 lines
+ mmio + base +
sblk->scaler_blk.base, "%s_%s",
+sspp_block->name,sblk->scaler_blk.name);Actually, it would be better to:
- drop name from all sblk instances (and use known string instead of the sblk name here)
- Usesblk->foo_blk.lento check if it should be printed or not.No, I dont agree. If we drop the names from the
sub_blkin the catalog, we will end up using "sub_blk_name" string here in the code to indicate which blk that is in the dump.If we add more
sub_blksin the catalog in the future we need to keep changing the code over here. Thats not how it should be.Quoted 8 lines
Quoted 5 lines
Leaving the names in the catalog ensures that this code wont change and only catalog changes when we add a new
sub_blkeither for an existing or new chipset.catalog is indicating the new blk, and dumping code just prints it.
with your approach, dumping code will or can keep changing with chipsets or
sub_blks. Thats not how it should be.Well, we do not enumerate sub-blocks in any way, they are not indexed. So even with
sblk->blk.namein place, adding new sub-block would require adding new code here. That's why I wrote that the calling code knows which sub-block it refers to.Today, unfortunately each
sub_blktype is different so we have to do this case by case.Ideally, this should have just been
Quoted 8 lines
-> print main blk
-> print all sub-blks of the main blkWithout having to handle each main blk's sub-blks separately.
That way the dumping code would have remained generic without having to do even the multiplexer in the first place.
Need to explore if somehow we can come up with a generic sub-blk struct and make this possible. Then this code will become much easier and what I am saying will make total sense.
In such case, yes. However I'd warn about having a generic array of subblocks. Having named subblock entries might complicate snapshotting, but it makes the rest of the DPU driver smaller.
Quoted 2 lines
Even without that, conceptually these sub-blk names are reflecting whats in our software document. So its not a random name but reflects the actual sub-blk name from the hardware.
Yes
Quoted 1 line
So this belongs in the catalog.
But the sub-block field already has a correct name: scaler_blk, csc_blk, etc. Having both sub-block field name and the .name inside results in kind of duplication, which seems unnecessary to me.
Quoted 1 line
Dumping code should not change or know whats the name of each block. It should just use whats in the catalog. thats why even conceptually I am not okay with your idea.
Dumping code itself (msm_disp_snapshot_*) doesn't. But the caller code knows what is the subblock.
Let me pick a definition from the patch:
static const struct dpu_dspp_sub_blks msm8998_dspp_sblk = {
.pcc = {.name = "pcc", .id = DPU_DSPP_PCC, .base = 0x1700,
.len = 0x90, .version = 0x10007},
};the "pcc" is repeated three times. When the code looks at this block, it already knows that it is a PCC block.
Compare this with:
static const struct dpu_dspp_sub_blks msm8998_dspp_sblk = {
.pcc = {
.base = 0x1700,
.len = 0x90,
.version = 0x10007,
},
};Nothing is repeated, but we still know that this is the DSPPn_PCC sub-block description.
Calling code does:
u32 base;
base = ctx->cap->sblk->pcc.base;
Quoted 47 lines
Quoted 23 lines
Let me extract the relevant code (skipping all the conditions for now):
msm_disp_snapshot_add_block(disp_state, sspp_block->len, mmio + base, "%s", sspp_block->name);if (
have_scaler) msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len, mmio + base +sblk->scaler_blk.base, "%s_%s",sspp_block->name,sblk->scaler_blk.name);if (
have_csc)
msm_disp_snapshot_add_block(disp_state,sblk->csc_blk.len, mmio + base +sblk->csc_blk.base, "%s_%s",sspp_block->name,sblk->csc_blk.name);Consider adding new sub-block, "baz". We would still require manual addition of the following code:
msm_disp_snapshot_add_block(disp_state,
sblk->baz_blk.len, mmio + base +sblk->baz_blk.base, "%s_%s",sspp_block->name,sblk->baz_blk.name);Quoted 6 lines
Compare this with:
msm_disp_snapshot_add_block(disp_state,
sblk->baz_blk.len, mmio + base +sblk->baz_blk.base, "%s_baz",sspp_block->name);Basically you are saying why not make the one line change here instead of using the name from the catalog.
I think it will be better to use from the catalog for the reason I wrote above that dumping code should just "use" the catalog's information and not become a catalog itself.
You are not saving much by dropping the sub-blk name from catalog anyway.
Quoted 6 lines
Moreover, if we follow the style of
dpu_kms_mdp_snapshot()(which doesn't use name), it should be:msm_disp_snapshot_add_block(disp_state,
sblk->baz_blk.len, mmio + base +sblk->baz_blk.base, "sspp%d_baz", idx);tbh, after looking at this series, it made me think why I didnt use the name from the catalog even for the
dpu_kms_mdp_snapshot()Quoted 2 lines
Quoted 2 lines
<snipped>
Signature · 2 lines
With best wishes Dmitry
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
Quoted 158 lines · Dmitry Baryshkov
Quoted 157 lines · Abhinav Kumar
Quoted 33 lines · Dmitry Baryshkov
Quoted 32 lines · Abhinav Kumar
Quoted 28 lines · Dmitry Baryshkov
Quoted 15 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>---
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c| 194
+++++++++++++++++++++++++++-----1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct · line 891
Quoted 7 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms) return 0; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, + void __iomem *mmio, void *blk, + enum dpu_hw_blk_type blk_type)No. Such multiplexers add no value to the code. Please inline it.
Not to mention that this patch is hard to review. You both move existing code and add new features. If it were to go, it should have been split into two patches: one introducing the multiplexer and another one adding subblocks.
Quoted 7 lines
Quoted 4 lines
Ok. we can split this into:
1) adding the multiplexer
2) adding sub-blk parsing support inside the multiplexerI'd say, drop the multiplexer completely. It adds no value here. It is only used from
dpu_kms_mdp_snapshot(). If the code there was complex enough, it would have made sense to _split_ the function. But even in such case there would be no point in having multiplexer. We do not enumerate block by type.Can you pls elaborate what you mean by enumerate blk by type?
We do have DPU_HW_BLK_***
Did you mean sub-blk?
Quoted 91 lines
Quoted 91 lines
Quoted 77 lines
Quoted 77 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +33
-0+{ + u32 base; + + switch (blk_type) { + case DPU_HW_BLK_TOP: + { + struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + + if (top->features & BIT(DPU_MDP_PERIPH_0_REMOVED)) { + msm_disp_snapshot_add_block(disp_state, MDP_PERIPH_TOP0, + mmio + top->base, "top"); + msm_disp_snapshot_add_block(disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base + MDP_PERIPH_TOP0_END, + "top_2"); + } else { + msm_disp_snapshot_add_block(disp_state, top->len, mmio + top->base,"top" ); + } + break; + } + case DPU_HW_BLK_LM: + { + struct dpu_lm_cfg *mixer = (struct dpu_lm_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, mixer->len, mmio + mixer->base,"%s" , + mixer->name); + break; + } + case DPU_HW_BLK_CTL: + { + struct dpu_ctl_cfg *ctl = (struct dpu_ctl_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, ctl->len, mmio +
ctl->base, "%s",
+ctl->name);
+ break;
+ }
+ caseDPU_HW_BLK_INTF:
+ {
+ structdpu_intf_cfg*intf = (structdpu_intf_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,intf->len, mmio +intf->base, "%s",
+intf->name);
+ break;
+ }
+ caseDPU_HW_BLK_WB:
+ {
+ structdpu_wb_cfg*wb = (structdpu_wb_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,wb->len, mmio +wb->base, "%s",
+wb->name);
+ break;
+ }
+ caseDPU_HW_BLK_SSPP:
+ {
+ structdpu_sspp_cfg*sspp_block= (structdpu_sspp_cfg*)blk;
+ const structdpu_sspp_sub_blks*sblk =sspp_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = sspp_block->base; + + msm_disp_snapshot_add_block(disp_state, sspp_block->len,mmio + base, "%s",
+sspp_block->name);
+
+ if (sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3LITE)||
+sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED4)) + msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len,
+ mmio + base +sblk->scaler_blk.base,Quoted 9 lines
Quoted 3 lines
"%s_%s",
+sspp_block->name,sblk->scaler_blk.name);Actually, it would be better to:
- drop name from all sblk instances (and use known string instead of the sblk name here)
- Usesblk->foo_blk.lento check if it should be printed or not.No, I dont agree. If we drop the names from the
sub_blkin the catalog, we will end up using "sub_blk_name" string here in the code to indicate which blk that is in the dump.If we add more
sub_blksin the catalog in the future we need to keep changing the code over here. Thats not how it should be.Quoted 8 lines
Quoted 5 lines
Leaving the names in the catalog ensures that this code wont change and only catalog changes when we add a new
sub_blkeither for an existing or new chipset.catalog is indicating the new blk, and dumping code just prints it.
with your approach, dumping code will or can keep changing with chipsets or
sub_blks. Thats not how it should be.Well, we do not enumerate sub-blocks in any way, they are not indexed. So even with
sblk->blk.namein place, adding new sub-block would require adding new code here. That's why I wrote that the calling code knows which sub-block it refers to.Today, unfortunately each
sub_blktype is different so we have to do this case by case.Ideally, this should have just been
-> print main blk
-> print all sub-blks of the main blk
Quoted 8 lines
Quoted 5 lines
Without having to handle each main blk's sub-blks separately.
That way the dumping code would have remained generic without having to do even the multiplexer in the first place.
Need to explore if somehow we can come up with a generic sub-blk struct and make this possible. Then this code will become much easier and what I am saying will make total sense.
In such case, yes. However I'd warn about having a generic array of subblocks. Having named subblock entries might complicate snapshotting, but it makes the rest of the DPU driver smaller.
Need to explore this. But not immediately.
Quoted 9 lines
Quoted 2 lines
Even without that, conceptually these sub-blk names are reflecting whats in our software document. So its not a random name but reflects the actual sub-blk name from the hardware.
Yes
Quoted 1 line
So this belongs in the catalog.
But the sub-block field already has a correct name:
scaler_blk,csc_blk, etc. Having both sub-block field name and the .name inside results in kind of duplication, which seems unnecessary to me.
No, there is a difference and not duplicated. One is the name of the struct so it can really be anything and doesnt need to match the hw doc name. But the other is the string name which we can give exactly to match software interface doc and makes parsing such a dump much much easier.
One point I dont see you have considered is the block index of the sub_blk.
Today, yes I see only a "pcc" or a "dither" etc
What if there are two PCCs or two dithers.
Then their names can just be "pcc_0" and "pcc_1" or "dither_0" and "dither_1".
Having name gives us the ability to easily incorporate even unsequential indices.
For example, every sspp's name today is not sequential. it can be "sspp_3" then "sspp_8" etc
By having names reflect the correct indices, dumping code becomes less complex as the catalog will still have the right names as dumping code will just use that.
Quoted 4 lines
Quoted 1 line
Dumping code should not change or know whats the name of each block. It should just use whats in the catalog. thats why even conceptually I am not okay with your idea.
Dumping code itself (msm_disp_snapshot_*) doesn't. But the caller code knows what is the subblock.
But this is the dumping code because we are adding which blocks to dump.
Quoted 9 lines
Let me pick a definition from the patch:
static const struct
dpu_dspp_sub_blksmsm8998_dspp_sblk= {
.pcc = {.name = "pcc", .id =DPU_DSPP_PCC, .base =0x1700, .len =0x90, .version =0x10007},
};the "pcc" is repeated three times. When the code looks at this block, it already knows that it is a PCC block.
I agree with you about the "id" though. From what I can see in this patch, "id" is not used and we can drop that. So that change from your series is fine with me.
But the pcc is the struct name which doesnt really need to reflect the hw name but the name field of the PCC should IMO.
Quoted 66 lines
Compare this with:
static const struct
dpu_dspp_sub_blksmsm8998_dspp_sblk= {
.pcc = {
.base =0x1700,
.len =0x90,
.version =0x10007,
},
};Nothing is repeated, but we still know that this is the
DSPPn_PCCsub-block description.Calling code does:
u32 base;
base =ctx->cap->sblk->pcc.base;Quoted 47 lines
Quoted 23 lines
Let me extract the relevant code (skipping all the conditions for now):
msm_disp_snapshot_add_block(disp_state, sspp_block->len, mmio + base, "%s", sspp_block->name);if (
have_scaler) msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len, mmio + base +sblk->scaler_blk.base, "%s_%s",sspp_block->name,sblk->scaler_blk.name);if (
have_csc)
msm_disp_snapshot_add_block(disp_state,sblk->csc_blk.len, mmio + base +sblk->csc_blk.base, "%s_%s",sspp_block->name,sblk->csc_blk.name);Consider adding new sub-block, "baz". We would still require manual addition of the following code:
msm_disp_snapshot_add_block(disp_state,
sblk->baz_blk.len, mmio + base +sblk->baz_blk.base, "%s_%s",sspp_block->name,sblk->baz_blk.name);Quoted 6 lines
Compare this with:
msm_disp_snapshot_add_block(disp_state,
sblk->baz_blk.len, mmio + base +sblk->baz_blk.base, "%s_baz",sspp_block->name);Basically you are saying why not make the one line change here instead of using the name from the catalog.
I think it will be better to use from the catalog for the reason I wrote above that dumping code should just "use" the catalog's information and not become a catalog itself.
You are not saving much by dropping the sub-blk name from catalog anyway.
Quoted 6 lines
Moreover, if we follow the style of
dpu_kms_mdp_snapshot()(which doesn't use name), it should be:msm_disp_snapshot_add_block(disp_state,
sblk->baz_blk.len, mmio + base +sblk->baz_blk.base, "sspp%d_baz", idx);tbh, after looking at this series, it made me think why I didnt use the name from the catalog even for the
dpu_kms_mdp_snapshot()Quoted 2 lines
Quoted 2 lines
<snipped>
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
Quoted 191 lines · Abhinav Kumar
Quoted 159 lines · Dmitry Baryshkov
Quoted 158 lines · Abhinav Kumar
Quoted 33 lines · Dmitry Baryshkov
Quoted 32 lines · Abhinav Kumar
Quoted 28 lines · Dmitry Baryshkov
Quoted 15 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>---
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c| 194
+++++++++++++++++++++++++++-----1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct · line 891
Quoted 7 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms) return 0; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, + void __iomem *mmio, void *blk, + enum dpu_hw_blk_type blk_type)No. Such multiplexers add no value to the code. Please inline it.
Not to mention that this patch is hard to review. You both move existing code and add new features. If it were to go, it should have been split into two patches: one introducing the multiplexer and another one adding subblocks.
Quoted 7 lines
Quoted 4 lines
Ok. we can split this into:
1) adding the multiplexer
2) adding sub-blk parsing support inside the multiplexerI'd say, drop the multiplexer completely. It adds no value here. It is only used from
dpu_kms_mdp_snapshot(). If the code there was complex enough, it would have made sense to _split_ the function. But even in such case there would be no point in having multiplexer. We do not enumerate block by type.Can you pls elaborate what you mean by enumerate blk by type?
We do have DPU_HW_BLK_***
Did you mean sub-blk?
Quoted 92 lines
Quoted 92 lines
Quoted 78 lines
Quoted 78 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +16
-0+{ + u32 base; + + switch (blk_type) { + case DPU_HW_BLK_TOP: + { + struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + + if (top->features & BIT(DPU_MDP_PERIPH_0_REMOVED)) { + msm_disp_snapshot_add_block(disp_state,MDP_PERIPH_TOP0 , + mmio + top->base, "top"); + msm_disp_snapshot_add_block(disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base + MDP_PERIPH_TOP0_END, + "top_2"); + } else { + msm_disp_snapshot_add_block(disp_state, top->len,mmio +
top->base, "top");
+ }
+ break;
+ }
+ caseDPU_HW_BLK_LM:
+ {
+ structdpu_lm_cfg*mixer = (structdpu_lm_cfg*)blk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +11
-0+ + msm_disp_snapshot_add_block(disp_state, mixer->len, mmio + mixer->base,"%s" , + mixer->name); + break; + } + case DPU_HW_BLK_CTL: + { + struct dpu_ctl_cfg *ctl = (struct dpu_ctl_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, ctl->len, mmio +
ctl->base, "%s",
+ctl->name);
+ break;
+ }
+ caseDPU_HW_BLK_INTF:
+ {
+ structdpu_intf_cfg*intf = (structdpu_intf_cfg*)blk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +11
-0+ + msm_disp_snapshot_add_block(disp_state, intf->len, mmio + intf->base,"%s" , + intf->name); + break; + } + case DPU_HW_BLK_WB: + { + struct dpu_wb_cfg *wb = (struct dpu_wb_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, wb->len, mmio +
wb->base, "%s",
+wb->name);
+ break;
+ }
+ caseDPU_HW_BLK_SSPP:
+ {
+ structdpu_sspp_cfg*sspp_block= (structdpu_sspp_cfg*)blk;
+ const structdpu_sspp_sub_blks*sblk =sspp_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = sspp_block->base; + + msm_disp_snapshot_add_block(disp_state, sspp_block->len,mmio + base, "%s",
+sspp_block->name);
+
+ if (sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3LITE)||
+sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED4)) + msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len,
+ mmio + base +sblk->scaler_blk.base,Quoted 9 lines
Quoted 3 lines
"%s_%s",
+sspp_block->name,sblk->scaler_blk.name);Actually, it would be better to:
- drop name from all sblk instances (and use known string instead of the sblk name here)
- Usesblk->foo_blk.lento check if it should be printed or not.No, I dont agree. If we drop the names from the
sub_blkin the catalog, we will end up using "sub_blk_name" string here in the code to indicate which blk that is in the dump.If we add more
sub_blksin the catalog in the future we need to keep changing the code over here. Thats not how it should be.Quoted 8 lines
Quoted 5 lines
Leaving the names in the catalog ensures that this code wont change and only catalog changes when we add a new
sub_blkeither for an existing or new chipset.catalog is indicating the new blk, and dumping code just prints it.
with your approach, dumping code will or can keep changing with chipsets or
sub_blks. Thats not how it should be.Well, we do not enumerate sub-blocks in any way, they are not indexed. So even with
sblk->blk.namein place, adding new sub-block would require adding new code here. That's why I wrote that the calling code knows which sub-block it refers to.Today, unfortunately each
sub_blktype is different so we have to do this case by case.Ideally, this should have just been
-> print main blk
-> print all sub-blks of the main blkQuoted 8 lines
Quoted 5 lines
Without having to handle each main blk's sub-blks separately.
That way the dumping code would have remained generic without having to do even the multiplexer in the first place.
Need to explore if somehow we can come up with a generic sub-blk struct and make this possible. Then this code will become much easier and what I am saying will make total sense.
In such case, yes. However I'd warn about having a generic array of subblocks. Having named subblock entries might complicate snapshotting, but it makes the rest of the DPU driver smaller.
Need to explore this. But not immediately.
Quoted 9 lines
Quoted 2 lines
Even without that, conceptually these sub-blk names are reflecting whats in our software document. So its not a random name but reflects the actual sub-blk name from the hardware.
Yes
Quoted 1 line
So this belongs in the catalog.
But the sub-block field already has a correct name:
scaler_blk,csc_blk, etc. Having both sub-block field name and the .name inside results in kind of duplication, which seems unnecessary to me.No, there is a difference and not duplicated. One is the name of the struct so it can really be anything and doesnt need to match the hw doc name. But the other is the string name which we can give exactly to match software interface doc and makes parsing such a dump much much easier.
One point I dont see you have considered is the block index of the
sub_blk.Today, yes I see only a "pcc" or a "dither" etc
What if there are two PCCs or two dithers.
Quoted 8 lines
Then their names can just be "
pcc_0" and "pcc_1" or "dither_0" and "dither_1".Having name gives us the ability to easily incorporate even unsequential indices.
For example, every sspp's name today is not sequential. it can be "
sspp_3" then "sspp_8" etcBy having names reflect the correct indices, dumping code becomes less complex as the catalog will still have the right names as dumping code will just use that.
The QC team is in agreement that we would like to go ahead with the names from the catalog and not drop them.
Hence we will post the next revision with the name still from the catalog and drop the multiplexer completely.
Since the intern has a short period of time to finish development on this task, we would like to go ahead with this approach and post the next rev.
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
Quoted 199 lines · Abhinav Kumar
Quoted 196 lines · Abhinav Kumar
Quoted 160 lines · Dmitry Baryshkov
Quoted 159 lines · Abhinav Kumar
Quoted 33 lines · Dmitry Baryshkov
Quoted 32 lines · Abhinav Kumar
Quoted 28 lines · Dmitry Baryshkov
Quoted 15 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>---
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c| 194
+++++++++++++++++++++++++++-----1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct · line 891
Quoted 7 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +3
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms) return 0; } +static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, + void __iomem *mmio, void *blk, + enum dpu_hw_blk_type blk_type)No. Such multiplexers add no value to the code. Please inline it.
Not to mention that this patch is hard to review. You both move existing code and add new features. If it were to go, it should have been split into two patches: one introducing the multiplexer and another one adding subblocks.
Quoted 7 lines
Quoted 4 lines
Ok. we can split this into:
1) adding the multiplexer
2) adding sub-blk parsing support inside the multiplexerI'd say, drop the multiplexer completely. It adds no value here. It is only used from
dpu_kms_mdp_snapshot(). If the code there was complex enough, it would have made sense to _split_ the function. But even in such case there would be no point in having multiplexer. We do not enumerate block by type.Can you pls elaborate what you mean by enumerate blk by type?
We do have DPU_HW_BLK_***
Did you mean sub-blk?
Quoted 93 lines
Quoted 93 lines
Quoted 79 lines
Quoted 79 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +13
-0+{ + u32 base; + + switch (blk_type) { + case DPU_HW_BLK_TOP: + { + struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk; + + if (top->features & BIT(DPU_MDP_PERIPH_0_REMOVED)) { + msm_disp_snapshot_add_block(disp_state,MDP_PERIPH_TOP0 , + mmio + top->base, "top"); + msm_disp_snapshot_add_block(disp_state, top->len - MDP_PERIPH_TOP0_END, + mmio + top->base +
MDP_PERIPH_TOP0_END,
+ "top_2");
+ } else {
+msm_disp_snapshot_add_block(disp_state, top->len, mmio + top->base, "top");
+ }
+ break;
+ }
+ caseDPU_HW_BLK_LM:
+ {
+ structdpu_lm_cfg*mixer = (structdpu_lm_cfg*)blk; +
+ msm_disp_snapshot_add_block(disp_state,mixer->len, mmio +mixer->base, "%s",
+mixer->name);
+ break;
+ }
+ caseDPU_HW_BLK_CTL:
+ {
+ structdpu_ctl_cfg*ctl = (structdpu_ctl_cfg*)blk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +20
-0+ + msm_disp_snapshot_add_block(disp_state, ctl->len, mmio + ctl->base, "%s", + ctl->name); + break; + } + case DPU_HW_BLK_INTF: + { + struct dpu_intf_cfg *intf = (struct dpu_intf_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, intf->len, mmio + intf->base,"%s" , + intf->name); + break; + } + case DPU_HW_BLK_WB: + { + struct dpu_wb_cfg *wb = (struct dpu_wb_cfg *)blk; + + msm_disp_snapshot_add_block(disp_state, wb->len, mmio +
wb->base, "%s",
+wb->name);
+ break;
+ }
+ caseDPU_HW_BLK_SSPP:
+ {
+ structdpu_sspp_cfg*sspp_block= (structdpu_sspp_cfg*)blk;
+ const structdpu_sspp_sub_blks*sblk =sspp_block->sblk;drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-0+ + base = sspp_block->base; + + msm_disp_snapshot_add_block(disp_state,
sspp_block->len, mmio + base, "%s",
+sspp_block->name);
+
+ if (sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3)|| +sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED3LITE)||
+sspp_block->features&BIT(DPU_SSPP_SCALER_QSEED4)) + msm_disp_snapshot_add_block(disp_state,sblk->scaler_blk.len,
+ mmio + base +Quoted 9 lines
Quoted 3 lines
sblk->scaler_blk.base, "%s_%s",
+sspp_block->name,sblk->scaler_blk.name);Actually, it would be better to:
- drop name from all sblk instances (and use known string instead of the sblk name here)
- Usesblk->foo_blk.lento check if it should be printed or not.No, I dont agree. If we drop the names from the
sub_blkin the catalog, we will end up using "sub_blk_name" string here in the code to indicate which blk that is in the dump.If we add more
sub_blksin the catalog in the future we need to keep changing the code over here. Thats not how it should be.Quoted 8 lines
Quoted 5 lines
Leaving the names in the catalog ensures that this code wont change and only catalog changes when we add a new
sub_blkeither for an existing or new chipset.catalog is indicating the new blk, and dumping code just prints it.
with your approach, dumping code will or can keep changing with chipsets or
sub_blks. Thats not how it should be.Well, we do not enumerate sub-blocks in any way, they are not indexed. So even with
sblk->blk.namein place, adding new sub-block would require adding new code here. That's why I wrote that the calling code knows which sub-block it refers to.Today, unfortunately each
sub_blktype is different so we have to do this case by case.Ideally, this should have just been
-> print main blk
-> print all sub-blks of the main blkQuoted 8 lines
Quoted 5 lines
Without having to handle each main blk's sub-blks separately.
That way the dumping code would have remained generic without having to do even the multiplexer in the first place.
Need to explore if somehow we can come up with a generic sub-blk struct and make this possible. Then this code will become much easier and what I am saying will make total sense.
In such case, yes. However I'd warn about having a generic array of subblocks. Having named subblock entries might complicate snapshotting, but it makes the rest of the DPU driver smaller.
Need to explore this. But not immediately.
Quoted 9 lines
Quoted 2 lines
Even without that, conceptually these sub-blk names are reflecting whats in our software document. So its not a random name but reflects the actual sub-blk name from the hardware.
Yes
Quoted 1 line
So this belongs in the catalog.
But the sub-block field already has a correct name:
scaler_blk,csc_blk, etc. Having both sub-block field name and the .name inside results in kind of duplication, which seems unnecessary to me.No, there is a difference and not duplicated. One is the name of the struct so it can really be anything and doesnt need to match the hw doc name. But the other is the string name which we can give exactly to match software interface doc and makes parsing such a dump much much easier.
One point I dont see you have considered is the block index of the
sub_blk.Today, yes I see only a "pcc" or a "dither" etc
What if there are two PCCs or two dithers.
Then their names can just be "
pcc_0" and "pcc_1" or "dither_0" and "dither_1".Having name gives us the ability to easily incorporate even unsequential indices.
Quoted 8 lines
Quoted 4 lines
For example, every sspp's name today is not sequential. it can be "
sspp_3" then "sspp_8" etcBy having names reflect the correct indices, dumping code becomes less complex as the catalog will still have the right names as dumping code will just use that.
The QC team is in agreement that we would like to go ahead with the names from the catalog and not drop them.
Hence we will post the next revision with the name still from the catalog and drop the multiplexer completely.
Ack, let's see how it goes.
Quoted 2 lines
Since the intern has a short period of time to finish development on this task, we would like to go ahead with this approach and post the next rev.
This is a bad argument.
Signature · 2 lines
With best wishes Dmitry
Re: [PATCH 6/6] drm/msm/dpu: Update dev core dump to dump registers of sub blocks
In reply to Ryan McCannQuoted 17 lines · Ryan McCann
Currently, the device core dump mechanism does not dump registers of sub blocks within the DSPP, SSPP, DSC, and PINGPONG blocks. Add wrapper function to dump hardware blocks that contain sub blocks.
Signed-off-by Ryan McCann <[email protected]>--- drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 194 +++++++++++++++++++++++++++----- 1 file changed, 168 insertions(+), 26 deletions(-)drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +0
-0@@ -885,6 +885,154 @@ static int dpu_irq_postinstall(struct msm_kms *kms)return 0 ; }
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms *kms) · line 897
Quoted 10 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +10
-0+static void dpu_kms_mdp_snapshot_add_block (struct msm_disp_state *disp_state, +void __iomem *mmio,void *blk, +enum dpu_hw_blk_type blk_type) +{ +u32 base; + +switch (blk_type) { +case DPU_HW_BLK_TOP : + { +struct dpu_mdp_cfg *top = (struct dpu_mdp_cfg *)blk;
Block configuration is constant, so these should be constant pointers. Not to mention that such type conversions are usually considered to be a bad idea. We loose type information, so compiler can not catch any errors here.
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms *kms) · line 908
Quoted 11 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +11
-0+ +if (top->features &BIT (DPU_MDP_PERIPH_0_REMOVED )) { +msm_disp_snapshot_add_block (disp_state,MDP_PERIPH_TOP0 , + mmio + top->base,"top" ); +msm_disp_snapshot_add_block (disp_state, top->len -MDP_PERIPH_TOP0_END , + mmio + top->base +MDP_PERIPH_TOP0_END , +"top_2" ); + }else { +msm_disp_snapshot_add_block (disp_state, top->len, mmio + top->base,"top" ); + } +break ;
[skipped]
default:
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static int dpu_irq_postinstall(struct msm_kms *kms) · line 1032
Quoted 1 line
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +1
-0+DPU_ERROR ("Block type not supported." );
Which block type?
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c static void dpu_kms_mdp_snapshot(struct msm_disp_state *disp_state, struct msm_k · line 1050
Quoted 13 lines
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +4
-2+ } +} +static void dpu_kms_mdp_snapshot (struct msm_disp_state *disp_state,struct msm_kms *kms) {int i;@@ -899,53 +1047,47 @@ static void dpu_kms_mdp_snapshot(struct msm_disp_state *disp_state, struct msm_k/* dump CTL sub-blocks HW regs info */ for (i =0 ; i < cat->ctl_count; i++) -msm_disp_snapshot_add_block (disp_state, cat->ctl[i].len, - dpu_kms->mmio + cat->ctl[i].base,"ctl_%d" , i); + dpu_kms_mdp_snapshot_add_block(disp_state, dpu_kms->mmio, (void *)&cat->ctl[i],
Here we are loosing the const property, which is a bad idea.
Quoted 1 line
drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c +1
-0+ DPU_HW_BLK_CTL);
Signature · 2 lines
With best wishes Dmitry
Marijn Suijten
reviewerIt is nice if cover letters also include the subsystem:
drm/msm: Add support to print DPU sub-block registers
Quoted 2 lines · Ryan McCann
The purpose of this patch series is to add support to print the registers of sub blocks in the dpu hardware catalog and fix the order in which all
Global nit: I think we previously settled on "sblk" or "sub-block(s)", not "sub blocks".
And capitalize DPU.
Quoted 10 lines
hardware blocks are dumped for a device core dump. This involves:
1. Changing data structure from stack to queue to fix the printing order of the device core dump.
2. Removing redundant suffix of sub block names.
3. Removing redundant prefix of sub block names.
4. Eliminating unused variable from relevant macros.
Dmitry has been doing that in one of his DPU catalog reworks.
Quoted 1 line
5. Defining names for sub blocks that have not yet been defined.
Let's see what this means, because the code logic should already be able to figure this out (and in some places we can perhaps delete the name entirely).
Quoted 2 lines
6. Implementing wrapper function that prints the registers of sub blocks when there is a need.
Thought this could be rather automated, but let me see what it means in the patches.
Quoted 5 lines
Sample Output of the
sspp_0block and its sub blocks for devcore dump:
======sspp_0======
...registers
...
====sspp_0_scaler====
My suggestion would be to put less emphasis on this header with:
----sspp_0_scaler----
So that it becomes obvious that this is a sblk of the ====sspp_0==== above.
Quoted 9 lines
...
...
====sspp_0_csc====
...
...
====next_block====
...Signed-off-by Ryan McCann <[email protected]>
No need for sign-off in cover letters.
- Marijn
Quoted 21 lines
---
Ryan McCann (6):
drm/msm: Update dev core dump to not print backwards drm/msm/dpu: Drop unused num argument from relevant macros drm/msm/dpu: Define names for unnamed sblks drm/msm/dpu: Remove redundant suffix in name of sub blocks drm/msm/disp: Remove redundant prefix in name of sub blocks drm/msm/dpu: Update dev core dump to dump registers of sub blocksdrivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c | 90 +++++----- drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 194 +++++++++++++++++++--- drivers/gpu/drm/msm/disp/msm_disp_snapshot_util.c | 2 +- 3 files changed, 214 insertions(+), 72 deletions(-)---
base-commit:710025fdedb3767655823c3a12d27d404d209f75
change-id: 20230622-devcoredump_patch-df7e8f6fd632Best regards,
Signature · 1 line
Ryan McCann <[email protected]>
Quoted 7 lines · Marijn Suijten
It is nice if cover letters also include the subsystem:
drm/msm: Add support to print DPU sub-block registers
Quoted 2 lines · Ryan McCann
The purpose of this patch series is to add support to print the registers of sub blocks in the dpu hardware catalog and fix the order in which all
Hey Marijn,
Sorry for the late response -- had to shift focus onto another feature for a bit.
Quoted 5 lines
Global nit: I think we previously settled on "sblk" or "sub-block(s)", not "sub blocks".
And capitalize DPU.
Acked.
Quoted 6 lines
Quoted 6 lines
hardware blocks are dumped for a device core dump. This involves:
1. Changing data structure from stack to queue to fix the printing order of the device core dump.
Quoted 7 lines
Quoted 5 lines
2. Removing redundant suffix of sub block names.
3. Removing redundant prefix of sub block names.
4. Eliminating unused variable from relevant macros.
Dmitry has been doing that in one of his DPU catalog reworks.
Got it. Is there a specific one that's making similar changes? IIRC, I didn't see one that was changing the *_SBLK macro.
Quoted 19 lines
Quoted 2 lines
5. Defining names for sub blocks that have not yet been defined.
Let's see what this means, because the code logic should already be able to figure this out (and in some places we can perhaps delete the name entirely).
Quoted 2 lines
6. Implementing wrapper function that prints the registers of sub blocks when there is a need.
Thought this could be rather automated, but let me see what it means in the patches.
Quoted 5 lines
Sample Output of the
sspp_0block and its sub blocks for devcore dump:
======sspp_0======
...registers
...
====sspp_0_scaler====
Quoted 6 lines
My suggestion would be to put less emphasis on this header with:
----sspp_0_scaler----So that it becomes obvious that this is a sblk of the ====sspp_0==== above.
FWIW, I think having the main block name prefix in the sblk header makes it clear which blocks are main block and which ones are sblks.
In addition, I'd like to keep this change within DPU (aside from the fix in the first patch) since implementing this change would require changing the *_add_block() parameters, and DSI/DP don't seem to have a need for conditional header formatting.
Thanks,
Jessica Zhang
Quoted 36 lines
Quoted 10 lines
...
...
====sspp_0_csc====
...
...
====next_block====
...Signed-off-by Ryan McCann <[email protected]>No need for sign-off in cover letters.
- Marijn
Quoted 21 lines
---
Ryan McCann (6):
drm/msm: Update dev core dump to not print backwards drm/msm/dpu: Drop unused num argument from relevant macros drm/msm/dpu: Define names for unnamed sblks drm/msm/dpu: Remove redundant suffix in name of sub blocks drm/msm/disp: Remove redundant prefix in name of sub blocks drm/msm/dpu: Update dev core dump to dump registers of sub blocksdrivers/gpu/drm/msm/disp/dpu1/dpu_hw_catalog.c | 90 +++++----- drivers/gpu/drm/msm/disp/dpu1/dpu_kms.c | 194 +++++++++++++++++++--- drivers/gpu/drm/msm/disp/msm_disp_snapshot_util.c | 2 +- 3 files changed, 214 insertions(+), 72 deletions(-)---
base-commit:710025fdedb3767655823c3a12d27d404d209f75
change-id: 20230622-devcoredump_patch-df7e8f6fd632Best regards,
Signature · 1 line
Ryan McCann <[email protected]>