[PATCH] Revert "spi: zynq_qspi: Use dummy buswidth in dummy byte calculation"
From: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
This reverts commit e09784728689de7949d4cdd559a9590e0bfcc702. The commit wrongly divides the dummy bytes by dummy bus width to calculate the dummy bytes. The framework already converts the dummy cycles to the number of bytes and the controller use the SPI flash command to determine the dummy cycles via the address width.
Signed-off-by: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
---
drivers/spi/zynq_qspi.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)
diff --git a/drivers/spi/zynq_qspi.c b/drivers/spi/zynq_qspi.c index 00e3ffcd1d..d1d4048966 100644 --- a/drivers/spi/zynq_qspi.c +++ b/drivers/spi/zynq_qspi.c @@ -676,7 +676,6 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, const struct spi_mem_op *op) { int op_len, pos = 0, ret, i; - u32 dummy_bytes = 0; unsigned int flag = 0; const u8 *tx_buf = NULL; u8 *rx_buf = NULL; @@ -689,11 +688,6 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, }
op_len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes; - if (op->dummy.nbytes) { - op_len = op->cmd.nbytes + op->addr.nbytes + - op->dummy.nbytes / op->dummy.buswidth; - dummy_bytes = op->dummy.nbytes / op->dummy.buswidth; - }
u8 op_buf[op_len];
@@ -707,8 +701,8 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, pos += op->addr.nbytes; }
- if (dummy_bytes) - memset(op_buf + pos, 0xff, dummy_bytes); + if (op->dummy.nbytes) + memset(op_buf + pos, 0xff, op->dummy.nbytes);
/* 1st transfer: opcode + address + dummy cycles */ /* Make sure to set END bit if no tx or rx data messages follow */ -- 2.30.2
________________________________ Kommanditgesellschaft - Sitz: Detmold - Amtsgericht Lemgo HRA 2790 - Komplementärin: Weidmüller Interface Führungsgesellschaft mbH - Sitz: Detmold - Amtsgericht Lemgo HRB 3924; Geschäftsführer: Dr. Timo Berger, Volker Bibelhausen, Dr. Sebastian Durst, André Sombecki; USt-ID-Nr. DE124599660
On 3/31/23 16:44, Stefan Herbrechtsmeier wrote:
From: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
This reverts commit e09784728689de7949d4cdd559a9590e0bfcc702. The commit wrongly divides the dummy bytes by dummy bus width to calculate the dummy bytes. The framework already converts the dummy cycles to the number of bytes and the controller use the SPI flash command to determine the dummy cycles via the address width.
Signed-off-by: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
drivers/spi/zynq_qspi.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)
diff --git a/drivers/spi/zynq_qspi.c b/drivers/spi/zynq_qspi.c index 00e3ffcd1d..d1d4048966 100644 --- a/drivers/spi/zynq_qspi.c +++ b/drivers/spi/zynq_qspi.c @@ -676,7 +676,6 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, const struct spi_mem_op *op) { int op_len, pos = 0, ret, i;
u32 dummy_bytes = 0; unsigned int flag = 0; const u8 *tx_buf = NULL; u8 *rx_buf = NULL;@@ -689,11 +688,6 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, }
op_len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
if (op->dummy.nbytes) {op_len = op->cmd.nbytes + op->addr.nbytes +op->dummy.nbytes / op->dummy.buswidth;dummy_bytes = op->dummy.nbytes / op->dummy.buswidth;} u8 op_buf[op_len];@@ -707,8 +701,8 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, pos += op->addr.nbytes; }
if (dummy_bytes)memset(op_buf + pos, 0xff, dummy_bytes);
if (op->dummy.nbytes)memset(op_buf + pos, 0xff, op->dummy.nbytes); /* 1st transfer: opcode + address + dummy cycles */ /* Make sure to set END bit if no tx or rx data messages follow */-- 2.30.2
Ashok: Can you please comment on this one?
Thanks, Michal
Hi Stefan,
-----Original Message----- From: Simek, Michal michal.simek@amd.com Sent: Monday, April 17, 2023 3:47 PM To: Stefan Herbrechtsmeier stefan.herbrechtsmeier-oss@weidmueller.com; u-boot@lists.denx.de; Soma, Ashok Reddy ashok.reddy.soma@amd.com Cc: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com; Jagan Teki jagan@amarulasolutions.com Subject: Re: [PATCH] Revert "spi: zynq_qspi: Use dummy buswidth in dummy byte calculation"
On 3/31/23 16:44, Stefan Herbrechtsmeier wrote:
From: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
This reverts commit e09784728689de7949d4cdd559a9590e0bfcc702. The commit wrongly divides the dummy bytes by dummy bus width to calculate the dummy bytes. The framework already converts the dummy cycles to the number of bytes and the controller use the SPI flash command to determine the dummy cycles via the address width.
As per my understanding dummy bus width should be equal to data buswidth and not equal to address bus width. Please let me know if this understanding in incorrect.
Based on the above understanding we have changed in Xilinx repo, at below code https://github.com/Xilinx/u-boot-xlnx/blob/024eb37c1e38ab811abe5408d42069fbd... from op->dummy.buswidth = spi_nor_get_protocol_addr_nbits(proto); to op->dummy.buswidth = spi_nor_get_protocol_data_nbits(proto);
I have sent the same as RFC. Please see the attached patch.
Currently this patch is present as part of Xilinx repo, and hence based on this below implementation dymmy_bytes are recalculated. I have sent controller driver code to upstream but, I have not yet sent spi-nor-core.c code, which I will be sending soon with some other changes.
Please let me know your thoughts about this patch and the changes.
Thanks, Ashok
Signed-off-by: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
drivers/spi/zynq_qspi.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)
diff --git a/drivers/spi/zynq_qspi.c b/drivers/spi/zynq_qspi.c index 00e3ffcd1d..d1d4048966 100644 --- a/drivers/spi/zynq_qspi.c +++ b/drivers/spi/zynq_qspi.c @@ -676,7 +676,6 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, const struct spi_mem_op *op) { int op_len, pos = 0, ret, i;
u32 dummy_bytes = 0; unsigned int flag = 0; const u8 *tx_buf = NULL; u8 *rx_buf = NULL;@@ -689,11 +688,6 @@ static int zynq_qspi_exec_op(struct spi_slave
*slave,
} op_len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
if (op->dummy.nbytes) {op_len = op->cmd.nbytes + op->addr.nbytes +op->dummy.nbytes / op->dummy.buswidth;dummy_bytes = op->dummy.nbytes / op->dummy.buswidth;} u8 op_buf[op_len];@@ -707,8 +701,8 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, pos += op->addr.nbytes; }
if (dummy_bytes)memset(op_buf + pos, 0xff, dummy_bytes);
if (op->dummy.nbytes)memset(op_buf + pos, 0xff, op->dummy.nbytes); /* 1st transfer: opcode + address + dummy cycles */ /* Make sure to set END bit if no tx or rx data messagesfollow */
2.30.2
Ashok: Can you please comment on this one?
Thanks, Michal
Hi Ashok,
Am 18.04.2023 um 10:43 schrieb Soma, Ashok Reddy:
-----Original Message----- From: Simek, Michal michal.simek@amd.com Sent: Monday, April 17, 2023 3:47 PM To: Stefan Herbrechtsmeier stefan.herbrechtsmeier-oss@weidmueller.com; u-boot@lists.denx.de; Soma, Ashok Reddy ashok.reddy.soma@amd.com Cc: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com; Jagan Teki jagan@amarulasolutions.com Subject: Re: [PATCH] Revert "spi: zynq_qspi: Use dummy buswidth in dummy byte calculation"
On 3/31/23 16:44, Stefan Herbrechtsmeier wrote:
From: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
This reverts commit e09784728689de7949d4cdd559a9590e0bfcc702. The commit wrongly divides the dummy bytes by dummy bus width to calculate the dummy bytes. The framework already converts the dummy cycles to the number of bytes and the controller use the SPI flash command to determine the dummy cycles via the address width.
As per my understanding dummy bus width should be equal to data buswidth and not equal to address bus width. Please let me know if this understanding in incorrect.
Why? The kernel use the addr bus width too.
Independent of the correct answer the core should handle this because it forward the dummy *bytes* to the driver:
/* convert the dummy cycles to the number of bytes */ op.dummy.nbytes = (nor->read_dummy * op.dummy.buswidth) / 8;
Furthermore, the driver send the dummy as part of the address:
/* 1st transfer: opcode + address + dummy cycles */
Based on the above understanding we have changed in Xilinx repo, at below code https://github.com/Xilinx/u-boot-xlnx/blob/024eb37c1e38ab811abe5408d42069fbd... from op->dummy.buswidth = spi_nor_get_protocol_addr_nbits(proto); to op->dummy.buswidth = spi_nor_get_protocol_data_nbits(proto);
I have sent the same as RFC. Please see the attached patch.
Currently this patch is present as part of Xilinx repo, and hence based on this below implementation dymmy_bytes are recalculated.
In this case the patch should be reverted until the other patch is accepted.
I have sent controller driver code to upstream but, I have not yet sent spi-nor-core.c code, which I will be sending soon with some other changes.
Please let me know your thoughts about this patch and the changes.
It is unclear to me why the dummy should be part of the data. The current implementation is wrong because it doesn't respects the address width and send the dummy bytes as part of the addr.
Regards Stefan
________________________________ Kommanditgesellschaft - Sitz: Detmold - Amtsgericht Lemgo HRA 2790 - Komplementärin: Weidmüller Interface Führungsgesellschaft mbH - Sitz: Detmold - Amtsgericht Lemgo HRB 3924; Geschäftsführer: Dr. Timo Berger, Volker Bibelhausen, Dr. Sebastian Durst, André Sombecki; USt-ID-Nr. DE124599660
Hi Ashok,
have you test your patches with 1-4-4 mode?
I think your patches are a nop for 1-1-4 mode and break other modes:
spi-nor-core op.dummy.nbytes = (nor->read_dummy * op.dummy.buswidth) / 8;
zynq_qspi: dummy_bytes = op->dummy.nbytes / op->dummy.buswidth;
=> dummy_bytes = (nor->read_dummy * op.dummy.buswidth) / 8) / op->dummy.buswidth => dummy_bytes = nor->read_dummy / 8
You changes ignores the bus width and breaks if the addr bus width is not 1!
Regards Stefan
Am 18.04.2023 um 12:27 schrieb Stefan Herbrechtsmeier:
Hi Ashok,
Am 18.04.2023 um 10:43 schrieb Soma, Ashok Reddy:
-----Original Message----- From: Simek, Michal michal.simek@amd.com Sent: Monday, April 17, 2023 3:47 PM To: Stefan Herbrechtsmeier stefan.herbrechtsmeier-oss@weidmueller.com; u-boot@lists.denx.de; Soma, Ashok Reddy ashok.reddy.soma@amd.com Cc: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com; Jagan Teki jagan@amarulasolutions.com Subject: Re: [PATCH] Revert "spi: zynq_qspi: Use dummy buswidth in dummy byte calculation"
On 3/31/23 16:44, Stefan Herbrechtsmeier wrote:
From: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
This reverts commit e09784728689de7949d4cdd559a9590e0bfcc702. The commit wrongly divides the dummy bytes by dummy bus width to calculate the dummy bytes. The framework already converts the dummy cycles to the number of bytes and the controller use the SPI flash command to determine the dummy cycles via the address width.
As per my understanding dummy bus width should be equal to data buswidth and not equal to address bus width. Please let me know if this understanding in incorrect.
Why? The kernel use the addr bus width too.
Independent of the correct answer the core should handle this because it forward the dummy *bytes* to the driver:
/* convert the dummy cycles to the number of bytes */ op.dummy.nbytes = (nor->read_dummy * op.dummy.buswidth) / 8;Furthermore, the driver send the dummy as part of the address:
/* 1st transfer: opcode + address + dummy cycles */Based on the above understanding we have changed in Xilinx repo, at below code https://github.com/Xilinx/u-boot-xlnx/blob/024eb37c1e38ab811abe5408d42069fbd...
from op->dummy.buswidth = spi_nor_get_protocol_addr_nbits(proto); to op->dummy.buswidth = spi_nor_get_protocol_data_nbits(proto);
I have sent the same as RFC. Please see the attached patch.
Currently this patch is present as part of Xilinx repo, and hence based on this below implementation dymmy_bytes are recalculated.
In this case the patch should be reverted until the other patch is accepted.
I have sent controller driver code to upstream but, I have not yet sent spi-nor-core.c code, which I will be sending soon with some other changes.
Please let me know your thoughts about this patch and the changes.
It is unclear to me why the dummy should be part of the data. The current implementation is wrong because it doesn't respects the address width and send the dummy bytes as part of the addr.
Regards Stefan
________________________________ Kommanditgesellschaft - Sitz: Detmold - Amtsgericht Lemgo HRA 2790 - Komplementärin: Weidmüller Interface Führungsgesellschaft mbH - Sitz: Detmold - Amtsgericht Lemgo HRB 3924; Geschäftsführer: Dr. Timo Berger, Volker Bibelhausen, Dr. Sebastian Durst, André Sombecki; USt-ID-Nr. DE124599660
Acked-by: Ashok Reddy Soma ashok.reddy.soma@amd.com
Hi Stefan,
-----Original Message----- From: Stefan Herbrechtsmeier <stefan.herbrechtsmeier- oss@weidmueller.com> Sent: Tuesday, April 18, 2023 3:58 PM To: Soma, Ashok Reddy ashok.reddy.soma@amd.com; Simek, Michal michal.simek@amd.com; u-boot@lists.denx.de Cc: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com; Jagan Teki jagan@amarulasolutions.com Subject: Re: [PATCH] Revert "spi: zynq_qspi: Use dummy buswidth in dummy byte calculation"
Hi Ashok,
Am 18.04.2023 um 10:43 schrieb Soma, Ashok Reddy:
-----Original Message----- From: Simek, Michal michal.simek@amd.com Sent: Monday, April 17, 2023 3:47 PM To: Stefan Herbrechtsmeier stefan.herbrechtsmeier-oss@weidmueller.com; u-boot@lists.denx.de; Soma, Ashok Reddy
Cc: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com; Jagan Teki jagan@amarulasolutions.com Subject: Re: [PATCH] Revert "spi: zynq_qspi: Use dummy buswidth in dummy byte calculation"
On 3/31/23 16:44, Stefan Herbrechtsmeier wrote:
From: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
This reverts commit e09784728689de7949d4cdd559a9590e0bfcc702.
The
commit wrongly divides the dummy bytes by dummy bus width to calculate the dummy bytes. The framework already converts the dummy cycles to the number of bytes and the controller use the SPI flash command to determine the dummy cycles via the address width.
As per my understanding dummy bus width should be equal to data
buswidth and not equal to address bus width.
Please let me know if this understanding in incorrect.
Why? The kernel use the addr bus width too.
Independent of the correct answer the core should handle this because it forward the dummy *bytes* to the driver:
/* convert the dummy cycles to the number of bytes */ op.dummy.nbytes = (nor->read_dummy * op.dummy.buswidth) / 8;Furthermore, the driver send the dummy as part of the address:
/* 1st transfer: opcode + address + dummy cycles */Based on the above understanding we have changed in Xilinx repo, at below code https://github.com/Xilinx/u-boot-
xlnx/blob/024eb37c1e38ab811abe5408d42
069fbd7901824/drivers/mtd/spi/spi-nor-core.c#L262 from op->dummy.buswidth = spi_nor_get_protocol_addr_nbits(proto); to op->dummy.buswidth = spi_nor_get_protocol_data_nbits(proto);
I have sent the same as RFC. Please see the attached patch.
Currently this patch is present as part of Xilinx repo, and hence based on this
below implementation dymmy_bytes are recalculated.
In this case the patch should be reverted until the other patch is accepted.
Agree with you, we can revert until I send the other patch. Ack'ing this patch.
Thanks, Ashok
I have sent controller driver code to upstream but, I have not yet sent spi-
nor-core.c code, which I will be sending soon with some other changes.
Please let me know your thoughts about this patch and the changes.
It is unclear to me why the dummy should be part of the data. The current implementation is wrong because it doesn't respects the address width and send the dummy bytes as part of the addr.
Regards Stefan
Kommanditgesellschaft - Sitz: Detmold - Amtsgericht Lemgo HRA 2790 - Komplementärin: Weidmüller Interface Führungsgesellschaft mbH - Sitz: Detmold - Amtsgericht Lemgo HRB 3924; Geschäftsführer: Dr. Timo Berger, Volker Bibelhausen, Dr. Sebastian Durst, André Sombecki; USt-ID-Nr. DE124599660
On 3/31/23 16:44, Stefan Herbrechtsmeier wrote:
From: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
This reverts commit e09784728689de7949d4cdd559a9590e0bfcc702. The commit wrongly divides the dummy bytes by dummy bus width to calculate the dummy bytes. The framework already converts the dummy cycles to the number of bytes and the controller use the SPI flash command to determine the dummy cycles via the address width.
Signed-off-by: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
drivers/spi/zynq_qspi.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)
diff --git a/drivers/spi/zynq_qspi.c b/drivers/spi/zynq_qspi.c index 00e3ffcd1d..d1d4048966 100644 --- a/drivers/spi/zynq_qspi.c +++ b/drivers/spi/zynq_qspi.c @@ -676,7 +676,6 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, const struct spi_mem_op *op) { int op_len, pos = 0, ret, i;
u32 dummy_bytes = 0; unsigned int flag = 0; const u8 *tx_buf = NULL; u8 *rx_buf = NULL;@@ -689,11 +688,6 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, }
op_len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes;
if (op->dummy.nbytes) {op_len = op->cmd.nbytes + op->addr.nbytes +op->dummy.nbytes / op->dummy.buswidth;dummy_bytes = op->dummy.nbytes / op->dummy.buswidth;} u8 op_buf[op_len];@@ -707,8 +701,8 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, pos += op->addr.nbytes; }
if (dummy_bytes)memset(op_buf + pos, 0xff, dummy_bytes);
if (op->dummy.nbytes)memset(op_buf + pos, 0xff, op->dummy.nbytes); /* 1st transfer: opcode + address + dummy cycles */ /* Make sure to set END bit if no tx or rx data messages follow */-- 2.30.2
Kommanditgesellschaft - Sitz: Detmold - Amtsgericht Lemgo HRA 2790 - Komplementärin: Weidmüller Interface Führungsgesellschaft mbH - Sitz: Detmold - Amtsgericht Lemgo HRB 3924; Geschäftsführer: Dr. Timo Berger, Volker Bibelhausen, Dr. Sebastian Durst, André Sombecki; USt-ID-Nr. DE124599660
Can you please rebase it on the top of the latest version?
$ git am -s ./20230331_stefan_herbrechtsmeier_oss_revert_spi_zynq_qspi_use_dummy_buswidth_in_dummy_byte_calculatio.mbx Adding link to lore.kernel.org Applying: Revert "spi: zynq_qspi: Use dummy buswidth in dummy byte calculation" error: patch failed: drivers/spi/zynq_qspi.c:676 error: drivers/spi/zynq_qspi.c: patch does not apply Patch failed at 0001 Revert "spi: zynq_qspi: Use dummy buswidth in dummy byte calculation" hint: Use 'git am --show-current-patch=diff' to see the failed patch When you have resolved this problem, run "git am --continue". If you prefer to skip this patch, run "git am --skip" instead. To restore the original branch and stop patching, run "git am --abort".
Thanks, Michal
Hi Michal,
Am 26.04.2023 um 16:37 schrieb Michal Simek:
On 3/31/23 16:44, Stefan Herbrechtsmeier wrote:
From: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
This reverts commit e09784728689de7949d4cdd559a9590e0bfcc702. The commit wrongly divides the dummy bytes by dummy bus width to calculate the dummy bytes. The framework already converts the dummy cycles to the number of bytes and the controller use the SPI flash command to determine the dummy cycles via the address width.
Signed-off-by: Stefan Herbrechtsmeier stefan.herbrechtsmeier@weidmueller.com
drivers/spi/zynq_qspi.c | 10 ++-------- 1 file changed, 2 insertions(+), 8 deletions(-)
diff --git a/drivers/spi/zynq_qspi.c b/drivers/spi/zynq_qspi.c index 00e3ffcd1d..d1d4048966 100644 --- a/drivers/spi/zynq_qspi.c +++ b/drivers/spi/zynq_qspi.c @@ -676,7 +676,6 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, const struct spi_mem_op *op) { int op_len, pos = 0, ret, i; - u32 dummy_bytes = 0; unsigned int flag = 0; const u8 *tx_buf = NULL; u8 *rx_buf = NULL; @@ -689,11 +688,6 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, }
op_len = op->cmd.nbytes + op->addr.nbytes + op->dummy.nbytes; - if (op->dummy.nbytes) { - op_len = op->cmd.nbytes + op->addr.nbytes + - op->dummy.nbytes / op->dummy.buswidth; - dummy_bytes = op->dummy.nbytes / op->dummy.buswidth; - }
u8 op_buf[op_len];
@@ -707,8 +701,8 @@ static int zynq_qspi_exec_op(struct spi_slave *slave, pos += op->addr.nbytes; }
- if (dummy_bytes) - memset(op_buf + pos, 0xff, dummy_bytes); + if (op->dummy.nbytes) + memset(op_buf + pos, 0xff, op->dummy.nbytes);
/* 1st transfer: opcode + address + dummy cycles */ /* Make sure to set END bit if no tx or rx data messages follow */ -- 2.30.2
Kommanditgesellschaft - Sitz: Detmold - Amtsgericht Lemgo HRA 2790 - Komplementärin: Weidmüller Interface Führungsgesellschaft mbH - Sitz: Detmold - Amtsgericht Lemgo HRB 3924; Geschäftsführer: Dr. Timo Berger, Volker Bibelhausen, Dr. Sebastian Durst, André Sombecki; USt-ID-Nr. DE124599660
Can you please rebase it on the top of the latest version?
I have resend the patch hopefully without reformation.
Regards Stefan
participants (3)
-
Michal Simek -
Soma, Ashok Reddy -
Stefan Herbrechtsmeier