Skip to contentSkip to conversation

staging: pi433: remove need to recompile code to debug fifo content

PATCH In review Paulo Miguel Almeida 3 participants 5 messages Feb 7, 2022 – Feb 8, 2022 Staging

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 staging: pi433: remove need to recompile code to debug fifo content

Debugging content present in the FIFO register is tricky as when we read the FIFO register that changes the content of fifo struct which reduces number of possible ways of debugging it. Rf69 uC has the possibility of triggering certain IRQs depending on how many items are in the FIFO queue, so being able to know what's in there is an important way to troubleshoot certain problems.

This patch removes the requirement of having to compile pi433 driver with DEBUG_FIFO_ACCESS set and let that be driven by printk verbositity level and/or dynamic debug config instead.

---
Meta-comments:

#1
In my mind, I didn't like the idea of having to change the code and then echo "module pi433 +p" > <debugfs>/dynamic_debug/control to only then be able to read stuff being sent/retrieved from fifo. It felt somewhat redundant at a certain level. On the other hand, I understand that removing the conditional compilation will force a for-loop to iterate for no real reason most of the time (max 66 iterations)... so I made a trade-off and in case anyone disagrees with that, just let me know and I will be happy to change to a different approach.

#2
In the past, it's been pointed out to me during code review that I tend to add code comments which could be omitted. In this case, the for-loop seemed a bit odd without explaining why it's in there. Let me know if you think I should keep/remove it.

Patch dependency:

This patch depends on the following patches being applied first:
- https://lore.kernel.org/lkml/[email protected]/
- https://lore.kernel.org/lkml/[email protected]/

---
 drivers/staging/pi433/rf69.c | 15 ++-------------
 1 file changed, 2 insertions(+), 13 deletions(-)
drivers/staging/pi433/rf69.c +2 -13
index a8def70808d6..901f8db3e3ce 100644
@@ -6,11 +6,6 @@
  *	Marcus Wolf <[email protected]>
  */
 
-/* enable prosa debug info */
-#undef DEBUG
-/* enable print of values on fifo access */
-#undef DEBUG_FIFO_ACCESS
-
 #include <linux/types.h>
 #include <linux/spi/spi.h>
 
@@ -829,9 +824,7 @@ int rf69_set_dagc(struct spi_device *spi, enum dagc dagc)
 
 int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size)
 {
-#ifdef DEBUG_FIFO_ACCESS
 	int i;
-#endif
 	struct spi_transfer transfer;
 	u8 local_buffer[FIFO_SIZE + 1];
 	int retval;
@@ -851,10 +844,9 @@ int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size)
 
 	retval = spi_sync_transfer(spi, &transfer, 1);
 
-#ifdef DEBUG_FIFO_ACCESS
+	/* print content read from fifo for debugging purposes */
 	for (i = 0; i < size; i++)
 		dev_dbg(&spi->dev, "%d - 0x%x\n", i, local_buffer[i + 1]);
-#endif
 
 	memcpy(buffer, &local_buffer[1], size);
 
@@ -863,9 +855,7 @@ int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size)
 
 int rf69_write_fifo(struct spi_device *spi, u8 *buffer, unsigned int size)
 {
-#ifdef DEBUG_FIFO_ACCESS
 	int i;
-#endif
 	u8 local_buffer[FIFO_SIZE + 1];
 
 	if (size > FIFO_SIZE) {
@@ -877,10 +867,9 @@ int rf69_write_fifo(struct spi_device *spi, u8 *buffer, unsigned int size)
 	local_buffer[0] = REG_FIFO | WRITE_BIT;
 	memcpy(&local_buffer[1], buffer, size);
 
-#ifdef DEBUG_FIFO_ACCESS
+	/* print content written from fifo for debugging purposes */
 	for (i = 0; i < size; i++)
 		dev_dbg(&spi->dev, "0x%x\n", buffer[i]);
-#endif
 
 	return spi_write(spi, local_buffer, size + 1);
 }
Signature · 1 line
2.34.1
#1 lore raw
Quoted 15 lines · Paulo Miguel Almeida

Debugging content present in the FIFO register is tricky as when we read the FIFO register that changes the content of fifo struct which reduces number of possible ways of debugging it. Rf69 uC has the possibility of triggering certain IRQs depending on how many items are in the FIFO queue, so being able to know what's in there is an important way to troubleshoot certain problems.

This patch removes the requirement of having to compile pi433 driver with DEBUG_FIFO_ACCESS set and let that be driven by printk verbositity level and/or dynamic debug config instead.

---
Meta-comments:

Quoted 10 lines

#1
In my mind, I didn't like the idea of having to change the code and then echo "module pi433 +p" > <debugfs>/dynamic_debug/control to only then be able to read stuff being sent/retrieved from fifo. It felt somewhat redundant at a certain level. On the other hand, I understand that removing the conditional compilation will force a for-loop to iterate for no real reason most of the time (max 66 iterations)... so I made a trade-off and in case anyone disagrees with that, just let me know and I will be happy to change to a different approach.

This is fine. It's useful information to you. It's makes the code nicer by removing ifdefs. It's not going to show up in benchmarking.

Quoted 5 lines

#2
In the past, it's been pointed out to me during code review that I tend to add code comments which could be omitted. In this case, the for-loop seemed a bit odd without explaining why it's in there. Let me know if you think I should keep/remove it.

Remove. Everyone knows what dev_dbg() does and the "read from fifo" vs "written from[sic] fifo" is built into the function name.

drivers/staging/pi433/rf69.c int rf69_set_dagc(struct spi_device *spi, enum dagc dagc) · line 829

Quoted 7 lines

int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size) {

drivers/staging/pi433/rf69.c +0 -2
-#ifdef DEBUG_FIFO_ACCESS
 	int i;
-#endif
 	struct spi_transfer transfer;
 	u8 local_buffer[FIFO_SIZE + 1];

You did not introduce this but we are potentially printing out uninitialized data if spi_sync_transfer() fails. Please initialize this with:

	u8 local_buffer[FIFO_SIZE + 1] = {};

Do that in a separate patch, though.

drivers/staging/pi433/rf69.c int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size) · line 852

Quoted 13 lines
drivers/staging/pi433/rf69.c +1 -2
 	int retval;
@@ -851,10 +844,9 @@ int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size)
 
 	retval = spi_sync_transfer(spi, &transfer, 1);
 
-#ifdef DEBUG_FIFO_ACCESS
+	/* print content read from fifo for debugging purposes */
 	for (i = 0; i < size; i++)
 		dev_dbg(&spi->dev, "%d - 0x%x\n", i, local_buffer[i + 1]);
-#endif
 
 	memcpy(buffer, &local_buffer[1], size);
 

regards,
dan carpenter

#2 lore raw
Quoted 30 lines · Dan Carpenter
Quoted 15 lines · Paulo Miguel Almeida

Debugging content present in the FIFO register is tricky as when we read the FIFO register that changes the content of fifo struct which reduces number of possible ways of debugging it. Rf69 uC has the possibility of triggering certain IRQs depending on how many items are in the FIFO queue, so being able to know what's in there is an important way to troubleshoot certain problems.

This patch removes the requirement of having to compile pi433 driver with DEBUG_FIFO_ACCESS set and let that be driven by printk verbositity level and/or dynamic debug config instead.

---
Meta-comments:

Quoted 10 lines

#1
In my mind, I didn't like the idea of having to change the code and then echo "module pi433 +p" > <debugfs>/dynamic_debug/control to only then be able to read stuff being sent/retrieved from fifo. It felt somewhat redundant at a certain level. On the other hand, I understand that removing the conditional compilation will force a for-loop to iterate for no real reason most of the time (max 66 iterations)... so I made a trade-off and in case anyone disagrees with that, just let me know and I will be happy to change to a different approach.

This is fine. It's useful information to you. It's makes the code nicer by removing ifdefs. It's not going to show up in benchmarking.

Quoted 11 lines
Quoted 5 lines

#2
In the past, it's been pointed out to me during code review that I tend to add code comments which could be omitted. In this case, the for-loop seemed a bit odd without explaining why it's in there. Let me know if you think I should keep/remove it.

Remove. Everyone knows what dev_dbg() does and the "read from fifo" vs "written from[sic] fifo" is built into the function name.

Quoted 2 lines

int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size) {

[]

Quoted 9 lines

drivers/staging/pi433/rf69.c int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size) · line 857

Quoted 9 lines
drivers/staging/pi433/rf69.c +1 -2
@@ -851,10 +844,9 @@ int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size)
 
 	retval = spi_sync_transfer(spi, &transfer, 1);
 
-#ifdef DEBUG_FIFO_ACCESS
+	/* print content read from fifo for debugging purposes */
 	for (i = 0; i < size; i++)
 		dev_dbg(&spi->dev, "%d - 0x%x\n", i, local_buffer[i + 1]);
-#endif

If you use

print_hex_dump_debug

perhaps the DEBUG_FIFO_ACCESS could be removed.

#3 lore raw
Quoted 8 lines · Joe Perches
Quoted 5 lines

drivers/staging/pi433/rf69.c int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size) · line 857

Quoted 5 lines
drivers/staging/pi433/rf69.c +1 -2
-#ifdef DEBUG_FIFO_ACCESS
+	/* print content read from fifo for debugging purposes */
 	for (i = 0; i < size; i++)
 		dev_dbg(&spi->dev, "%d - 0x%x\n", i, local_buffer[i + 1]);
-#endif

If you use

Quoted 4 lines

print_hex_dump_debug

perhaps the DEBUG_FIFO_ACCESS could be removed.

Hi Joe, thanks for taking the time to review my patch.

print_hex_dump_debug is pretty convenient and straight to the point which I do like. The only thing that I "didn't like" is the fact that when configuring dynamic debugging, print_hex_dump_debug only cares about 'p' flag and ignores all of the other flags that can be helpful for tracking things down.

So essentially I lose track of which function and device instance the message belongs to (users can have more than 1 at the same time).

But, because of your suggestion, I came across hex_dump_to_buffer() which bridged the gap. thanks a lot :)

Would you be comfortable with the following implementation? (Dan, feel free to chip in)

	char linebuf[FIFO_SIZE * 3] = {0};
	hex_dump_to_buffer(local_buffer + 1, size, 16, 1, linebuf,
			   ARRAY_SIZE(linebuf), false);
	dev_dbg(&spi->dev, "%s\n", linebuf);

thanks,

Paulo Almeida

#4 lore raw
In reply to Dan Carpenter
Quoted 9 lines · Dan Carpenter
Quoted 5 lines

#2
In the past, it's been pointed out to me during code review that I tend to add code comments which could be omitted. In this case, the for-loop seemed a bit odd without explaining why it's in there. Let me know if you think I should keep/remove it.

Remove. Everyone knows what dev_dbg() does and the "read from fifo" vs "written from[sic] fifo" is built into the function name.

fair enough

Quoted 12 lines

drivers/staging/pi433/rf69.c int rf69_set_dagc(struct spi_device *spi, enum dagc dagc) · line 829

Quoted 7 lines

int rf69_read_fifo(struct spi_device *spi, u8 *buffer, unsigned int size) {

drivers/staging/pi433/rf69.c +0 -2
-#ifdef DEBUG_FIFO_ACCESS
 	int i;
-#endif
 	struct spi_transfer transfer;
 	u8 local_buffer[FIFO_SIZE + 1];

You did not introduce this but we are potentially printing out uninitialized data if spi_sync_transfer() fails. Please initialize this with:

Quoted 4 lines
	u8 local_buffer[FIFO_SIZE + 1] = {};

Do that in a separate patch, though.

good point, will do. Thanks a lot :)

thanks,

Paulo Almeida