2021-12-09 18:07:58

by Vignesh Raghavendra

Subject: [PATCH] dma: ti: k3-udma: Workaround errata i2234

Per [1], UDMA TR15 transactions may hang if ICNT0 is less than 64B
Work around is to set EOL flag is to 1 for ICNT0.

Since, there is no performance penalty / side effects of setting EOL
flag event ICNTO > 64B, just set the flag for all UDMAP TR15
descriptors.

[1] https://www.ti.com/lit/er/sprz455a/sprz455a.pdf
Errata doc for J721E DRA829/TDA4VM Processors Silicon Revision 1.1/1.0 (Rev. A)

Signed-off-by: Vignesh Raghavendra <[email protected]>
---
drivers/dma/ti/k3-udma.c | 48 +++++++++++++++++++-----------------
include/linux/dma/ti-cppi5.h | 1 +
2 files changed, 27 insertions(+), 22 deletions(-)

diff --git a/drivers/dma/ti/k3-udma.c b/drivers/dma/ti/k3-udma.c
index 6e56d1cef5ee..d65c4cc5fbf7 100644
--- a/drivers/dma/ti/k3-udma.c
+++ b/drivers/dma/ti/k3-udma.c
@@ -2863,6 +2863,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
struct udma_desc *d;
struct cppi5_tr_type1_t *tr_req = NULL;
u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
+ u32 csf = CPPI5_TR_CSF_SUPR_EVT;
unsigned int i;
size_t tr_size;
int num_tr = 0;
@@ -2885,10 +2886,12 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,

d->sglen = sglen;

- if (uc->ud->match_data->type == DMA_TYPE_UDMA)
+ if (uc->ud->match_data->type == DMA_TYPE_UDMA) {
asel = 0;
- else
+ csf |= CPPI5_TR_CSF_EOL_ICNT0;
+ } else {
asel = (u64)uc->config.asel << K3_ADDRESS_ASEL_SHIFT;
+ }

tr_req = d->hwdesc[0].tr_req_base;
for_each_sg(sgl, sgent, sglen, i) {
@@ -2906,7 +2909,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,

cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1, false,
false, CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
- cppi5_tr_csf_set(&tr_req[tr_idx].flags, CPPI5_TR_CSF_SUPR_EVT);
+ cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);

sg_addr |= asel;
tr_req[tr_idx].addr = sg_addr;
@@ -2919,8 +2922,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
false, false,
CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
- cppi5_tr_csf_set(&tr_req[tr_idx].flags,
- CPPI5_TR_CSF_SUPR_EVT);
+ cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);

tr_req[tr_idx].addr = sg_addr + tr0_cnt1 * tr0_cnt0;
tr_req[tr_idx].icnt0 = tr1_cnt0;
@@ -2932,8 +2934,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
d->residue += sg_dma_len(sgent);
}

- cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags,
- CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
+ cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags, csf | CPPI5_TR_CSF_EOP);

return d;
}
@@ -2947,6 +2948,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
struct scatterlist *sgent;
struct cppi5_tr_type15_t *tr_req = NULL;
enum dma_slave_buswidth dev_width;
+ u32 csf = CPPI5_TR_CSF_SUPR_EVT;
u16 tr_cnt0, tr_cnt1;
dma_addr_t dev_addr;
struct udma_desc *d;
@@ -3017,6 +3019,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,

if (uc->ud->match_data->type == DMA_TYPE_UDMA) {
asel = 0;
+ csf |= CPPI5_TR_CSF_EOL_ICNT0;
} else {
asel = (u64)uc->config.asel << K3_ADDRESS_ASEL_SHIFT;
dev_addr |= asel;
@@ -3040,7 +3043,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,

cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE15, false,
true, CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
- cppi5_tr_csf_set(&tr_req[tr_idx].flags, CPPI5_TR_CSF_SUPR_EVT);
+ cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
cppi5_tr_set_trigger(&tr_req[tr_idx].flags,
uc->config.tr_trigger_type,
CPPI5_TR_TRIGGER_TYPE_ICNT2_DEC, 0, 0);
@@ -3086,8 +3089,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE15,
false, true,
CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
- cppi5_tr_csf_set(&tr_req[tr_idx].flags,
- CPPI5_TR_CSF_SUPR_EVT);
+ cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
cppi5_tr_set_trigger(&tr_req[tr_idx].flags,
uc->config.tr_trigger_type,
CPPI5_TR_TRIGGER_TYPE_ICNT2_DEC,
@@ -3131,8 +3133,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
d->residue += sg_len;
}

- cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags,
- CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
+ cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags, csf | CPPI5_TR_CSF_EOP);

return d;
}
@@ -3454,6 +3455,7 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
struct cppi5_tr_type1_t *tr_req;
unsigned int periods = buf_len / period_len;
u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
+ u32 csf = CPPI5_TR_CSF_SUPR_EVT;
unsigned int i;
int num_tr;

@@ -3472,11 +3474,13 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
return NULL;

tr_req = d->hwdesc[0].tr_req_base;
- if (uc->ud->match_data->type == DMA_TYPE_UDMA)
+ if (uc->ud->match_data->type == DMA_TYPE_UDMA) {
period_addr = buf_addr;
- else
+ csf |= CPPI5_TR_CSF_EOL_ICNT0;
+ } else {
period_addr = buf_addr |
((u64)uc->config.asel << K3_ADDRESS_ASEL_SHIFT);
+ }

for (i = 0; i < periods; i++) {
int tr_idx = i * num_tr;
@@ -3490,8 +3494,7 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
tr_req[tr_idx].dim1 = tr0_cnt0;

if (num_tr == 2) {
- cppi5_tr_csf_set(&tr_req[tr_idx].flags,
- CPPI5_TR_CSF_SUPR_EVT);
+ cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
tr_idx++;

cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
@@ -3505,8 +3508,7 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
}

if (!(flags & DMA_PREP_INTERRUPT))
- cppi5_tr_csf_set(&tr_req[tr_idx].flags,
- CPPI5_TR_CSF_SUPR_EVT);
+ cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);

period_addr += period_len;
}
@@ -3659,6 +3661,7 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
int num_tr;
size_t tr_size = sizeof(struct cppi5_tr_type15_t);
u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
+ u32 csf = CPPI5_TR_CSF_SUPR_EVT;

if (uc->config.dir != DMA_MEM_TO_MEM) {
dev_err(chan->device->dev,
@@ -3689,13 +3692,15 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
if (uc->ud->match_data->type != DMA_TYPE_UDMA) {
src |= (u64)uc->ud->asel << K3_ADDRESS_ASEL_SHIFT;
dest |= (u64)uc->ud->asel << K3_ADDRESS_ASEL_SHIFT;
+ } else {
+ csf |= CPPI5_TR_CSF_EOL_ICNT0;
}

tr_req = d->hwdesc[0].tr_req_base;

cppi5_tr_init(&tr_req[0].flags, CPPI5_TR_TYPE15, false, true,
CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
- cppi5_tr_csf_set(&tr_req[0].flags, CPPI5_TR_CSF_SUPR_EVT);
+ cppi5_tr_csf_set(&tr_req[0].flags, csf);

tr_req[0].addr = src;
tr_req[0].icnt0 = tr0_cnt0;
@@ -3714,7 +3719,7 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
if (num_tr == 2) {
cppi5_tr_init(&tr_req[1].flags, CPPI5_TR_TYPE15, false, true,
CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
- cppi5_tr_csf_set(&tr_req[1].flags, CPPI5_TR_CSF_SUPR_EVT);
+ cppi5_tr_csf_set(&tr_req[1].flags, csf);

tr_req[1].addr = src + tr0_cnt1 * tr0_cnt0;
tr_req[1].icnt0 = tr1_cnt0;
@@ -3729,8 +3734,7 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
tr_req[1].dicnt3 = 1;
}

- cppi5_tr_csf_set(&tr_req[num_tr - 1].flags,
- CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
+ cppi5_tr_csf_set(&tr_req[num_tr - 1].flags, csf | CPPI5_TR_CSF_EOP);

if (uc->config.metadata_size)
d->vd.tx.metadata_ops = &metadata_ops;
diff --git a/include/linux/dma/ti-cppi5.h b/include/linux/dma/ti-cppi5.h
index efa2f0309f00..c53c0f6e3b1a 100644
--- a/include/linux/dma/ti-cppi5.h
+++ b/include/linux/dma/ti-cppi5.h
@@ -616,6 +616,7 @@ static inline void *cppi5_hdesc_get_swdata(struct cppi5_host_desc_t *desc)
#define CPPI5_TR_CSF_SUPR_EVT BIT(2)
#define CPPI5_TR_CSF_EOL_ADV_SHIFT (4U)
#define CPPI5_TR_CSF_EOL_ADV_MASK GENMASK(6, 4)
+#define CPPI5_TR_CSF_EOL_ICNT0 BIT(4)
#define CPPI5_TR_CSF_EOP BIT(7)

/**
--
2.34.1

2021-12-10 15:22:06

by Péter Ujfalusi

Subject: Re: [PATCH] dma: ti: k3-udma: Workaround errata i2234

Hi Vignesh,

On 09/12/2021 20:07, Vignesh Raghavendra wrote:
> Per [1], UDMA TR15 transactions may hang if ICNT0 is less than 64B
> Work around is to set EOL flag is to 1 for ICNT0.
>
> Since, there is no performance penalty / side effects of setting EOL
> flag event ICNTO > 64B, just set the flag for all UDMAP TR15
> descriptors.

PDMAs and CSI does not send EOL? If you set it the EOL to one then when
it arrives the remaining icnt0 is skipped...

>
> [1] https://www.ti.com/lit/er/sprz455a/sprz455a.pdf
> Errata doc for J721E DRA829/TDA4VM Processors Silicon Revision 1.1/1.0 (Rev. A)
>
> Signed-off-by: Vignesh Raghavendra <[email protected]>
> ---
> drivers/dma/ti/k3-udma.c | 48 +++++++++++++++++++-----------------
> include/linux/dma/ti-cppi5.h | 1 +
> 2 files changed, 27 insertions(+), 22 deletions(-)
>
> diff --git a/drivers/dma/ti/k3-udma.c b/drivers/dma/ti/k3-udma.c
> index 6e56d1cef5ee..d65c4cc5fbf7 100644
> --- a/drivers/dma/ti/k3-udma.c
> +++ b/drivers/dma/ti/k3-udma.c
> @@ -2863,6 +2863,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
> struct udma_desc *d;
> struct cppi5_tr_type1_t *tr_req = NULL;
> u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
> + u32 csf = CPPI5_TR_CSF_SUPR_EVT;
> unsigned int i;
> size_t tr_size;
> int num_tr = 0;
> @@ -2885,10 +2886,12 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
>
> d->sglen = sglen;
>
> - if (uc->ud->match_data->type == DMA_TYPE_UDMA)
> + if (uc->ud->match_data->type == DMA_TYPE_UDMA) {
> asel = 0;
> - else
> + csf |= CPPI5_TR_CSF_EOL_ICNT0;
> + } else {
> asel = (u64)uc->config.asel << K3_ADDRESS_ASEL_SHIFT;
> + }
>
> tr_req = d->hwdesc[0].tr_req_base;
> for_each_sg(sgl, sgent, sglen, i) {
> @@ -2906,7 +2909,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
>
> cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1, false,
> false, CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags, CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
>
> sg_addr |= asel;
> tr_req[tr_idx].addr = sg_addr;
> @@ -2919,8 +2922,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
> cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
> false, false,
> CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags,
> - CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
>
> tr_req[tr_idx].addr = sg_addr + tr0_cnt1 * tr0_cnt0;
> tr_req[tr_idx].icnt0 = tr1_cnt0;
> @@ -2932,8 +2934,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
> d->residue += sg_dma_len(sgent);
> }
>
> - cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags,
> - CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
> + cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags, csf | CPPI5_TR_CSF_EOP);
>
> return d;
> }
> @@ -2947,6 +2948,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
> struct scatterlist *sgent;
> struct cppi5_tr_type15_t *tr_req = NULL;
> enum dma_slave_buswidth dev_width;
> + u32 csf = CPPI5_TR_CSF_SUPR_EVT;
> u16 tr_cnt0, tr_cnt1;
> dma_addr_t dev_addr;
> struct udma_desc *d;
> @@ -3017,6 +3019,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
>
> if (uc->ud->match_data->type == DMA_TYPE_UDMA) {
> asel = 0;
> + csf |= CPPI5_TR_CSF_EOL_ICNT0;
> } else {
> asel = (u64)uc->config.asel << K3_ADDRESS_ASEL_SHIFT;
> dev_addr |= asel;
> @@ -3040,7 +3043,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
>
> cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE15, false,
> true, CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags, CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
> cppi5_tr_set_trigger(&tr_req[tr_idx].flags,
> uc->config.tr_trigger_type,
> CPPI5_TR_TRIGGER_TYPE_ICNT2_DEC, 0, 0);
> @@ -3086,8 +3089,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
> cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE15,
> false, true,
> CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags,
> - CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
> cppi5_tr_set_trigger(&tr_req[tr_idx].flags,
> uc->config.tr_trigger_type,
> CPPI5_TR_TRIGGER_TYPE_ICNT2_DEC,
> @@ -3131,8 +3133,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
> d->residue += sg_len;
> }
>
> - cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags,
> - CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
> + cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags, csf | CPPI5_TR_CSF_EOP);
>
> return d;
> }
> @@ -3454,6 +3455,7 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
> struct cppi5_tr_type1_t *tr_req;
> unsigned int periods = buf_len / period_len;
> u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
> + u32 csf = CPPI5_TR_CSF_SUPR_EVT;
> unsigned int i;
> int num_tr;
>
> @@ -3472,11 +3474,13 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
> return NULL;
>
> tr_req = d->hwdesc[0].tr_req_base;
> - if (uc->ud->match_data->type == DMA_TYPE_UDMA)
> + if (uc->ud->match_data->type == DMA_TYPE_UDMA) {
> period_addr = buf_addr;
> - else
> + csf |= CPPI5_TR_CSF_EOL_ICNT0;
> + } else {
> period_addr = buf_addr |
> ((u64)uc->config.asel << K3_ADDRESS_ASEL_SHIFT);
> + }
>
> for (i = 0; i < periods; i++) {
> int tr_idx = i * num_tr;
> @@ -3490,8 +3494,7 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
> tr_req[tr_idx].dim1 = tr0_cnt0;
>
> if (num_tr == 2) {
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags,
> - CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
> tr_idx++;
>
> cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
> @@ -3505,8 +3508,7 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
> }
>
> if (!(flags & DMA_PREP_INTERRUPT))
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags,
> - CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
>
> period_addr += period_len;
> }
> @@ -3659,6 +3661,7 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
> int num_tr;
> size_t tr_size = sizeof(struct cppi5_tr_type15_t);
> u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
> + u32 csf = CPPI5_TR_CSF_SUPR_EVT;
>
> if (uc->config.dir != DMA_MEM_TO_MEM) {
> dev_err(chan->device->dev,
> @@ -3689,13 +3692,15 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
> if (uc->ud->match_data->type != DMA_TYPE_UDMA) {
> src |= (u64)uc->ud->asel << K3_ADDRESS_ASEL_SHIFT;
> dest |= (u64)uc->ud->asel << K3_ADDRESS_ASEL_SHIFT;
> + } else {
> + csf |= CPPI5_TR_CSF_EOL_ICNT0;
> }
>
> tr_req = d->hwdesc[0].tr_req_base;
>
> cppi5_tr_init(&tr_req[0].flags, CPPI5_TR_TYPE15, false, true,
> CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[0].flags, CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[0].flags, csf);
>
> tr_req[0].addr = src;
> tr_req[0].icnt0 = tr0_cnt0;
> @@ -3714,7 +3719,7 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
> if (num_tr == 2) {
> cppi5_tr_init(&tr_req[1].flags, CPPI5_TR_TYPE15, false, true,
> CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[1].flags, CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[1].flags, csf);
>
> tr_req[1].addr = src + tr0_cnt1 * tr0_cnt0;
> tr_req[1].icnt0 = tr1_cnt0;
> @@ -3729,8 +3734,7 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
> tr_req[1].dicnt3 = 1;
> }
>
> - cppi5_tr_csf_set(&tr_req[num_tr - 1].flags,
> - CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
> + cppi5_tr_csf_set(&tr_req[num_tr - 1].flags, csf | CPPI5_TR_CSF_EOP);
>
> if (uc->config.metadata_size)
> d->vd.tx.metadata_ops = &metadata_ops;
> diff --git a/include/linux/dma/ti-cppi5.h b/include/linux/dma/ti-cppi5.h
> index efa2f0309f00..c53c0f6e3b1a 100644
> --- a/include/linux/dma/ti-cppi5.h
> +++ b/include/linux/dma/ti-cppi5.h
> @@ -616,6 +616,7 @@ static inline void *cppi5_hdesc_get_swdata(struct cppi5_host_desc_t *desc)
> #define CPPI5_TR_CSF_SUPR_EVT BIT(2)
> #define CPPI5_TR_CSF_EOL_ADV_SHIFT (4U)
> #define CPPI5_TR_CSF_EOL_ADV_MASK GENMASK(6, 4)
> +#define CPPI5_TR_CSF_EOL_ICNT0 BIT(4)

the correct expression is: (1 << CPPI5_TR_CSF_EOL_ADV_SHIFT)
as EOL = 1 is what you want to set.
EOL = 2 will clear icnt0 and 1 on EOL.
3 will do the same for icnt 0, 1 and 2
4 will skip the remainin tr.

> #define CPPI5_TR_CSF_EOP BIT(7)
>
> /**
>

--
Péter

2021-12-13 05:36:45

by Vinod Koul

Subject: Re: [PATCH] dma: ti: k3-udma: Workaround errata i2234

On 09-12-21, 23:37, Vignesh Raghavendra wrote:
> Per [1], UDMA TR15 transactions may hang if ICNT0 is less than 64B
> Work around is to set EOL flag is to 1 for ICNT0.
>
> Since, there is no performance penalty / side effects of setting EOL
> flag event ICNTO > 64B, just set the flag for all UDMAP TR15
> descriptors.

s/dma/dmaengine

>
> [1] https://www.ti.com/lit/er/sprz455a/sprz455a.pdf
> Errata doc for J721E DRA829/TDA4VM Processors Silicon Revision 1.1/1.0 (Rev. A)
>
> Signed-off-by: Vignesh Raghavendra <[email protected]>
> ---
> drivers/dma/ti/k3-udma.c | 48 +++++++++++++++++++-----------------
> include/linux/dma/ti-cppi5.h | 1 +
> 2 files changed, 27 insertions(+), 22 deletions(-)
>
> diff --git a/drivers/dma/ti/k3-udma.c b/drivers/dma/ti/k3-udma.c
> index 6e56d1cef5ee..d65c4cc5fbf7 100644
> --- a/drivers/dma/ti/k3-udma.c
> +++ b/drivers/dma/ti/k3-udma.c
> @@ -2863,6 +2863,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
> struct udma_desc *d;
> struct cppi5_tr_type1_t *tr_req = NULL;
> u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
> + u32 csf = CPPI5_TR_CSF_SUPR_EVT;
> unsigned int i;
> size_t tr_size;
> int num_tr = 0;
> @@ -2885,10 +2886,12 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
>
> d->sglen = sglen;
>
> - if (uc->ud->match_data->type == DMA_TYPE_UDMA)
> + if (uc->ud->match_data->type == DMA_TYPE_UDMA) {
> asel = 0;
> - else
> + csf |= CPPI5_TR_CSF_EOL_ICNT0;
> + } else {
> asel = (u64)uc->config.asel << K3_ADDRESS_ASEL_SHIFT;
> + }
>
> tr_req = d->hwdesc[0].tr_req_base;
> for_each_sg(sgl, sgent, sglen, i) {
> @@ -2906,7 +2909,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
>
> cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1, false,
> false, CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags, CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
>
> sg_addr |= asel;
> tr_req[tr_idx].addr = sg_addr;
> @@ -2919,8 +2922,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
> cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
> false, false,
> CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags,
> - CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
>
> tr_req[tr_idx].addr = sg_addr + tr0_cnt1 * tr0_cnt0;
> tr_req[tr_idx].icnt0 = tr1_cnt0;
> @@ -2932,8 +2934,7 @@ udma_prep_slave_sg_tr(struct udma_chan *uc, struct scatterlist *sgl,
> d->residue += sg_dma_len(sgent);
> }
>
> - cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags,
> - CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
> + cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags, csf | CPPI5_TR_CSF_EOP);
>
> return d;
> }
> @@ -2947,6 +2948,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
> struct scatterlist *sgent;
> struct cppi5_tr_type15_t *tr_req = NULL;
> enum dma_slave_buswidth dev_width;
> + u32 csf = CPPI5_TR_CSF_SUPR_EVT;
> u16 tr_cnt0, tr_cnt1;
> dma_addr_t dev_addr;
> struct udma_desc *d;
> @@ -3017,6 +3019,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
>
> if (uc->ud->match_data->type == DMA_TYPE_UDMA) {
> asel = 0;
> + csf |= CPPI5_TR_CSF_EOL_ICNT0;
> } else {
> asel = (u64)uc->config.asel << K3_ADDRESS_ASEL_SHIFT;
> dev_addr |= asel;
> @@ -3040,7 +3043,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
>
> cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE15, false,
> true, CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags, CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
> cppi5_tr_set_trigger(&tr_req[tr_idx].flags,
> uc->config.tr_trigger_type,
> CPPI5_TR_TRIGGER_TYPE_ICNT2_DEC, 0, 0);
> @@ -3086,8 +3089,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
> cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE15,
> false, true,
> CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags,
> - CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
> cppi5_tr_set_trigger(&tr_req[tr_idx].flags,
> uc->config.tr_trigger_type,
> CPPI5_TR_TRIGGER_TYPE_ICNT2_DEC,
> @@ -3131,8 +3133,7 @@ udma_prep_slave_sg_triggered_tr(struct udma_chan *uc, struct scatterlist *sgl,
> d->residue += sg_len;
> }
>
> - cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags,
> - CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
> + cppi5_tr_csf_set(&tr_req[tr_idx - 1].flags, csf | CPPI5_TR_CSF_EOP);
>
> return d;
> }
> @@ -3454,6 +3455,7 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
> struct cppi5_tr_type1_t *tr_req;
> unsigned int periods = buf_len / period_len;
> u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
> + u32 csf = CPPI5_TR_CSF_SUPR_EVT;
> unsigned int i;
> int num_tr;
>
> @@ -3472,11 +3474,13 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
> return NULL;
>
> tr_req = d->hwdesc[0].tr_req_base;
> - if (uc->ud->match_data->type == DMA_TYPE_UDMA)
> + if (uc->ud->match_data->type == DMA_TYPE_UDMA) {
> period_addr = buf_addr;
> - else
> + csf |= CPPI5_TR_CSF_EOL_ICNT0;
> + } else {
> period_addr = buf_addr |
> ((u64)uc->config.asel << K3_ADDRESS_ASEL_SHIFT);
> + }
>
> for (i = 0; i < periods; i++) {
> int tr_idx = i * num_tr;
> @@ -3490,8 +3494,7 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
> tr_req[tr_idx].dim1 = tr0_cnt0;
>
> if (num_tr == 2) {
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags,
> - CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
> tr_idx++;
>
> cppi5_tr_init(&tr_req[tr_idx].flags, CPPI5_TR_TYPE1,
> @@ -3505,8 +3508,7 @@ udma_prep_dma_cyclic_tr(struct udma_chan *uc, dma_addr_t buf_addr,
> }
>
> if (!(flags & DMA_PREP_INTERRUPT))
> - cppi5_tr_csf_set(&tr_req[tr_idx].flags,
> - CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[tr_idx].flags, csf);
>
> period_addr += period_len;
> }
> @@ -3659,6 +3661,7 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
> int num_tr;
> size_t tr_size = sizeof(struct cppi5_tr_type15_t);
> u16 tr0_cnt0, tr0_cnt1, tr1_cnt0;
> + u32 csf = CPPI5_TR_CSF_SUPR_EVT;
>
> if (uc->config.dir != DMA_MEM_TO_MEM) {
> dev_err(chan->device->dev,
> @@ -3689,13 +3692,15 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
> if (uc->ud->match_data->type != DMA_TYPE_UDMA) {
> src |= (u64)uc->ud->asel << K3_ADDRESS_ASEL_SHIFT;
> dest |= (u64)uc->ud->asel << K3_ADDRESS_ASEL_SHIFT;
> + } else {
> + csf |= CPPI5_TR_CSF_EOL_ICNT0;
> }
>
> tr_req = d->hwdesc[0].tr_req_base;
>
> cppi5_tr_init(&tr_req[0].flags, CPPI5_TR_TYPE15, false, true,
> CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[0].flags, CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[0].flags, csf);
>
> tr_req[0].addr = src;
> tr_req[0].icnt0 = tr0_cnt0;
> @@ -3714,7 +3719,7 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
> if (num_tr == 2) {
> cppi5_tr_init(&tr_req[1].flags, CPPI5_TR_TYPE15, false, true,
> CPPI5_TR_EVENT_SIZE_COMPLETION, 0);
> - cppi5_tr_csf_set(&tr_req[1].flags, CPPI5_TR_CSF_SUPR_EVT);
> + cppi5_tr_csf_set(&tr_req[1].flags, csf);
>
> tr_req[1].addr = src + tr0_cnt1 * tr0_cnt0;
> tr_req[1].icnt0 = tr1_cnt0;
> @@ -3729,8 +3734,7 @@ udma_prep_dma_memcpy(struct dma_chan *chan, dma_addr_t dest, dma_addr_t src,
> tr_req[1].dicnt3 = 1;
> }
>
> - cppi5_tr_csf_set(&tr_req[num_tr - 1].flags,
> - CPPI5_TR_CSF_SUPR_EVT | CPPI5_TR_CSF_EOP);
> + cppi5_tr_csf_set(&tr_req[num_tr - 1].flags, csf | CPPI5_TR_CSF_EOP);
>
> if (uc->config.metadata_size)
> d->vd.tx.metadata_ops = &metadata_ops;
> diff --git a/include/linux/dma/ti-cppi5.h b/include/linux/dma/ti-cppi5.h
> index efa2f0309f00..c53c0f6e3b1a 100644
> --- a/include/linux/dma/ti-cppi5.h
> +++ b/include/linux/dma/ti-cppi5.h
> @@ -616,6 +616,7 @@ static inline void *cppi5_hdesc_get_swdata(struct cppi5_host_desc_t *desc)
> #define CPPI5_TR_CSF_SUPR_EVT BIT(2)
> #define CPPI5_TR_CSF_EOL_ADV_SHIFT (4U)
> #define CPPI5_TR_CSF_EOL_ADV_MASK GENMASK(6, 4)
> +#define CPPI5_TR_CSF_EOL_ICNT0 BIT(4)
> #define CPPI5_TR_CSF_EOP BIT(7)
>
> /**
> --
> 2.34.1

--
~Vinod

2023-03-23 11:58:11

by Jayesh Choudhary

Subject: Re: [PATCH] dma: ti: k3-udma: Workaround errata i2234

Hello Peter,

On 10/12/21 20:52, Péter Ujfalusi wrote:
> Hi Vignesh,
>
> On 09/12/2021 20:07, Vignesh Raghavendra wrote:
>> Per [1], UDMA TR15 transactions may hang if ICNT0 is less than 64B
>> Work around is to set EOL flag is to 1 for ICNT0.
>>
>> Since, there is no performance penalty / side effects of setting EOL
>> flag event ICNTO > 64B, just set the flag for all UDMAP TR15
>> descriptors.
>
> PDMAs and CSI does not send EOL? If you set it the EOL to one then when
> it arrives the remaining icnt0 is skipped...

I am planning to respin v2 for this.
Will fix this in v2.

>
>>
>> [1] https://www.ti.com/lit/er/sprz455a/sprz455a.pdf
>> Errata doc for J721E DRA829/TDA4VM Processors Silicon Revision 1.1/1.0 (Rev. A)
>>
>> Signed-off-by: Vignesh Raghavendra <[email protected]>
>> ---
>> drivers/dma/ti/k3-udma.c | 48 +++++++++++++++++++-----------------
>> include/linux/dma/ti-cppi5.h | 1 +
>> 2 files changed, 27 insertions(+), 22 deletions(-)
>>

[...]

>> diff --git a/include/linux/dma/ti-cppi5.h b/include/linux/dma/ti-cppi5.h
>> index efa2f0309f00..c53c0f6e3b1a 100644
>> --- a/include/linux/dma/ti-cppi5.h
>> +++ b/include/linux/dma/ti-cppi5.h
>> @@ -616,6 +616,7 @@ static inline void *cppi5_hdesc_get_swdata(struct cppi5_host_desc_t *desc)
>> #define CPPI5_TR_CSF_SUPR_EVT BIT(2)
>> #define CPPI5_TR_CSF_EOL_ADV_SHIFT (4U)
>> #define CPPI5_TR_CSF_EOL_ADV_MASK GENMASK(6, 4)
>> +#define CPPI5_TR_CSF_EOL_ICNT0 BIT(4)
>
> the correct expression is: (1 << CPPI5_TR_CSF_EOL_ADV_SHIFT)

Both these expressions expands to the same value
(CPPI5_TR_CSF_EOL_ADV_SHIFT = 4U)

And according to the linux checkpatch, the usage of BIT macro
is preferred so I will keep it the same.

Warm regards,
-Jayesh

> as EOL = 1 is what you want to set.
> EOL = 2 will clear icnt0 and 1 on EOL.
> 3 will do the same for icnt 0, 1 and 2
> 4 will skip the remainin tr.
>
>> #define CPPI5_TR_CSF_EOP BIT(7)
>>
>> /**
>>
>