dmaengine: memset clarifications and fixes
Subsystems (5)
v3 cover dmaengine: memset clarifications and fixes
The following contains a clarification for the behavior of the 'value' parameter in the memset operation. It is intended to be a single byte pattern as laid out here:
https://lore.kernel.org/dmaengine/YejrA5ZWZ3lTRO%2F1@matsya/
Then I'm attempting to fix all places it is currently used. But note that I do not have access to this hardware and cannot test it. We'll really need a maintainer to take a look at each of these to verify that the changes are correct.
v3:
- Using signed byte for pattern everywhere
Ben Walker (4):
dmaengine: Document dmaengine_prep_dma_memset
dmaengine: at_hdmac: In atc_prep_dma_memset, treat value as a single
byte
dmaengine: at_xdmac: In at_xdmac_prep_dma_memset, treat value as a
single byte
dmaengine: hidma: In hidma_prep_dma_memset treat value as a single
byte
drivers/dma/at_hdmac.c | 10 +++++++++- drivers/dma/at_xdmac.c | 9 ++++++++- drivers/dma/qcom/hidma.c | 13 ++++++++++++- include/linux/dmaengine.h | 8 ++++++++ 4 files changed, 37 insertions(+), 3 deletions(-)
Signature · 1 line
2.33.1
PATCH v3 1/4 dmaengine: Document dmaengine_prep_dma_memset
Document this function to make clear the expected behavior of the 'value' parameter. It was intended to match the behavior of POSIX memset as laid out here:
https://lore.kernel.org/dmaengine/YejrA5ZWZ3lTRO%2F1@matsya/
--- Documentation/driver-api/dmaengine/provider.rst | 6 ++++++ include/linux/dmaengine.h | 8 ++++++++ 2 files changed, 14 insertions(+)
Documentation/driver-api/dmaengine/provider.rst +6 -0
@@ -206,6 +206,12 @@ Currently, the types available are: - The device is able to perform parity check using RAID6 P+Q algorithm against a memory buffer. +- DMA_MEMSET + + - The device is able to fill memory with the provided pattern + + - The pattern is treated as a single byte signed value. + - DMA_INTERRUPT - The device is able to trigger a dummy transfer that will
include/linux/dmaengine.h +8 -0
@@ -1031,6 +1031,14 @@ static inline struct dma_async_tx_descriptor *dmaengine_prep_interleaved_dma(return chan->device->device_prep_interleaved_dma (chan, xt, flags); } +/** +* dmaengine_prep_dma_memset() - Prepare a DMA memset descriptor. +* @chan: The channel to be used for this descriptor +* @dest: Address of buffer to be set +* @value: Treated as a single byte value that fills the destination buffer +* @len: The total size of dest +* @flags: DMA engine flags +*/ static inline struct dma_async_tx_descriptor *dmaengine_prep_dma_memset (struct dma_chan *chan,dma_addr_t dest,int value,size_t len,unsigned long flags)
Signature · 1 line
2.35.1
PATCH v3 2/4 dmaengine: at_hdmac: In atc_prep_dma_memset, treat value as a single byte
In reply to Ben WalkerThe value passed in to .prep_dma_memset is to be treated as a single byte repeating pattern.
--- drivers/dma/at_hdmac.c | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-)
drivers/dma/at_hdmac.c +9 -1
@@ -942,6 +942,7 @@ atc_prep_dma_memset(struct dma_chan *chan, dma_addr_t dest, int value,struct at_desc *desc;void __iomem *vaddr;dma_addr_t paddr; +char fill_pattern;dev_vdbg (chan2dev (chan),"%s: d%pad v0x%x l0x%zx f0x%lx\n" , __func__, &dest, value, len, flags);
@@ -963,7 +964,14 @@ atc_prep_dma_memset(struct dma_chan *chan, dma_addr_t dest, int value, __func__);return NULL ; } - *(u32 *)vaddr = value; + +/* Only the first byte of value is to be used according to dmaengine */ + fill_pattern = (char )value; + + *(u32 *)vaddr = (fill_pattern <<24 ) | + (fill_pattern <<16 ) | + (fill_pattern <<8 ) | + fill_pattern; desc =atc_create_memset_desc (chan, paddr, dest, len);if (!desc) {
Signature · 1 line
2.35.1
PATCH v3 3/4 dmaengine: at_xdmac: In at_xdmac_prep_dma_memset, treat value as a single byte
In reply to Ben WalkerThe value passed in to .prep_dma_memset is to be treated as a single byte repeating pattern.
--- drivers/dma/at_xdmac.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)
drivers/dma/at_xdmac.c +8 -1
@@ -1202,6 +1202,7 @@ static struct at_xdmac_desc *at_xdmac_memset_create_desc(struct dma_chan *chan,unsigned long flags;size_t ublen;u32 dwidth; +char pattern;/* * WARNING: The channel configuration is set here since there is no * dmaengine_slave_config call in this case. Moreover we don't know the
@@ -1244,10 +1245,16 @@ static struct at_xdmac_desc *at_xdmac_memset_create_desc(struct dma_chan *chan, chan_cc |=AT_XDMAC_CC_DWIDTH (dwidth); +/* Only the first byte of value is to be used according to dmaengine */ + pattern = (char )value; + ublen = len >> dwidth; desc->lld.mbr_da = dst_addr; - desc->lld.mbr_ds = value; + desc->lld.mbr_ds = (pattern <<24 ) | + (pattern <<16 ) | + (pattern <<8 ) | + pattern; desc->lld.mbr_ubc =AT_XDMAC_MBR_UBC_NDV3 |AT_XDMAC_MBR_UBC_NDEN |AT_XDMAC_MBR_UBC_NSEN
Signature · 1 line
2.35.1
Re: [PATCH v3 3/4] dmaengine: at_xdmac: In at_xdmac_prep_dma_memset, treat value as a single byte
Sorry for the delay.
Just wanted to inform you that I plan to test this patch set this week.
Cheers,
ta
PATCH v3 4/4 dmaengine: hidma: In hidma_prep_dma_memset treat value as a single byte
In reply to Ben WalkerThe value parameter is a single byte, so duplicate it to the 8 byte range that is used as the pattern.
--- drivers/dma/qcom/hidma.c | 13 ++++++++++++- 1 file changed, 12 insertions(+), 1 deletion(-)
drivers/dma/qcom/hidma.c +12 -1
@@ -431,6 +431,7 @@ hidma_prep_dma_memset(struct dma_chan *dmach, dma_addr_t dest, int value,struct hidma_desc *mdesc =NULL ;struct hidma_dev *mdma = mchan->dmadev;unsigned long irqflags; +u64 byte_pattern, fill_pattern;/* Get free descriptor */ spin_lock_irqsave (&mchan->lock, irqflags);
@@ -443,9 +444,19 @@ hidma_prep_dma_memset(struct dma_chan *dmach, dma_addr_t dest, int value,if (!mdesc)return NULL ; + byte_pattern = (char )value; + fill_pattern = (byte_pattern <<56 ) | + (byte_pattern <<48 ) | + (byte_pattern <<40 ) | + (byte_pattern <<32 ) | + (byte_pattern <<24 ) | + (byte_pattern <<16 ) | + (byte_pattern <<8 ) | + byte_pattern; + mdesc->desc.flags = flags;hidma_ll_set_transfer_params (mdma->lldev, mdesc->tre_ch, - value, dest, len, flags, + fill_pattern, dest, len, flags,HIDMA_TRE_MEMSET );/* Place descriptor in prepared list */
Signature · 1 line
2.35.1
Quoted 10 lines · Ben Walker
The following contains a clarification for the behavior of the 'value' parameter in the memset operation. It is intended to be a single byte pattern as laid out here:
https://lore.kernel.org/dmaengine/YejrA5ZWZ3lTRO%2F1@matsya/
Then I'm attempting to fix all places it is currently used. But note that I do not have access to this hardware and cannot test it. We'll really need a maintainer to take a look at each of these to verify that the changes are correct.
Applied, thanks
Signature · 1 line
~Vinod
+ Maxime.
Hi!
Quoted 7 lines · Ben Walker
The following contains a clarification for the behavior of the 'value' parameter in the memset operation. It is intended to be a single byte pattern as laid out here:
https://lore.kernel.org/dmaengine/YejrA5ZWZ3lTRO%2F1@matsya/
Then I'm attempting to fix all places it is currently used. But note
Sorry for being late. I see there are no in kernel users of dma_memset/dma_memset_sg methods (other than dmatest.c). Any idea why do we still keep them?
Signature · 2 lines
Cheers, ta