Skip to contentSkip to conversation

staging:mt29f_spinand: MT29F2G failing as only 16-bit arguments and variables used for addressing.

PATCH In review Jheng-Jhong Wu 4 participants 10 messages Aug 1, 2018 – Aug 6, 2018

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 staging:mt29f_spinand: MT29F2G failing as only 16-bit arguments and variables used for addressing.

For NAND flash chips with more than 1Gbit (e.g. MT29F2G) more than 16 bits are necessary to address the correct page. The driver sets the address for more than 16 bits, but it uses 16-bit arguments and variables (these are page_id, block_id, row) to do address operations. Obviously, these arguments and variables cannot deal with more than 16-bit address.

---
 drivers/staging/mt29f_spinand/mt29f_spinand.c | 22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)
drivers/staging/mt29f_spinand/mt29f_spinand.c +11 -11
index 4484784..a0f4cbcb 100644
@@ -308,10 +308,10 @@ static int spinand_write_enable(struct spi_device *spi_nand)
 	return spinand_cmd(spi_nand, &cmd);
 }
 
-static int spinand_read_page_to_cache(struct spi_device *spi_nand, u16 page_id)
+static int spinand_read_page_to_cache(struct spi_device *spi_nand, u32 page_id)
 {
 	struct spinand_cmd cmd = {0};
-	u16 row;
+	u32 row;
 
 	row = page_id;
 	cmd.cmd = CMD_READ;
@@ -331,7 +331,7 @@ static int spinand_read_page_to_cache(struct spi_device *spi_nand, u16 page_id)
  *   locations.
  *   No tRd delay.
  */
-static int spinand_read_from_cache(struct spi_device *spi_nand, u16 page_id,
+static int spinand_read_from_cache(struct spi_device *spi_nand, u32 page_id,
 				   u16 byte_id, u16 len, u8 *rbuf)
 {
 	struct spinand_cmd cmd = {0};
@@ -362,7 +362,7 @@ static int spinand_read_from_cache(struct spi_device *spi_nand, u16 page_id,
  *   The read includes two commands to the Nand - 0x13 and 0x03 commands
  *   Poll to read status to wait for tRD time.
  */
-static int spinand_read_page(struct spi_device *spi_nand, u16 page_id,
+static int spinand_read_page(struct spi_device *spi_nand, u32 page_id,
 			     u16 offset, u16 len, u8 *rbuf)
 {
 	int ret;
@@ -430,7 +430,7 @@ static int spinand_read_page(struct spi_device *spi_nand, u16 page_id,
  *   Since it is writing the data to cache, there is no tPROG time.
  */
 static int spinand_program_data_to_cache(struct spi_device *spi_nand,
-					 u16 page_id, u16 byte_id,
+					 u32 page_id, u16 byte_id,
 					 u16 len, u8 *wbuf)
 {
 	struct spinand_cmd cmd = {0};
@@ -457,10 +457,10 @@ static int spinand_program_data_to_cache(struct spi_device *spi_nand,
  *   the Nand array.
  *   Need to wait for tPROG time to finish the transaction.
  */
-static int spinand_program_execute(struct spi_device *spi_nand, u16 page_id)
+static int spinand_program_execute(struct spi_device *spi_nand, u32 page_id)
 {
 	struct spinand_cmd cmd = {0};
-	u16 row;
+	u32 row;
 
 	row = page_id;
 	cmd.cmd = CMD_PROG_PAGE_EXC;
@@ -486,7 +486,7 @@ static int spinand_program_execute(struct spi_device *spi_nand, u16 page_id)
  *   Poll to wait for the tPROG time to finish the transaction.
  */
 static int spinand_program_page(struct spi_device *spi_nand,
-				u16 page_id, u16 offset, u16 len, u8 *buf)
+				u32 page_id, u16 offset, u16 len, u8 *buf)
 {
 	int retval;
 	u8 status = 0;
@@ -573,10 +573,10 @@ static int spinand_program_page(struct spi_device *spi_nand,
  *   one block--64 pages
  *   Need to wait for tERS.
  */
-static int spinand_erase_block_erase(struct spi_device *spi_nand, u16 block_id)
+static int spinand_erase_block_erase(struct spi_device *spi_nand, u32 block_id)
 {
 	struct spinand_cmd cmd = {0};
-	u16 row;
+	u32 row;
 
 	row = block_id;
 	cmd.cmd = CMD_ERASE_BLK;
@@ -599,7 +599,7 @@ static int spinand_erase_block_erase(struct spi_device *spi_nand, u16 block_id)
  *   and then send the 0xd8 erase command
  *   Poll to wait for the tERS time to complete the tranaction.
  */
-static int spinand_erase_block(struct spi_device *spi_nand, u16 block_id)
+static int spinand_erase_block(struct spi_device *spi_nand, u32 block_id)
 {
 	int retval;
 	u8 status = 0;
Signature · 1 line
2.7.4
#1 lore raw

Hi Jheng-Jhong,

Quoted 5 lines · On Wed, 1 Aug 2018 11:24:19 +0800 Jheng-Jhong Wu <[email protected]>…

For NAND flash chips with more than 1Gbit (e.g. MT29F2G) more than 16 bits are necessary to address the correct page. The driver sets the address for more than 16 bits, but it uses 16-bit arguments and variables (these are page_id, block_id, row) to do address operations. Obviously, these arguments and variables cannot deal with more than 16-bit address.

I plan to remove this driver soon (after 4.19-rc1 is out). It's now been replaced by the SPI framework [1]. Can you check if your NAND SPI NAND is supported, if it's not add support for it, and if you find a bug report/fix it.

Thanks,

Boris

[1]https://git.kernel.org/pub/scm/linux/kernel/git/next/linux-next.git/tree/drivers/mtd/nand/spi?h=next-20180731

Quoted 94 lines
---
 drivers/staging/mt29f_spinand/mt29f_spinand.c | 22 +++++++++++-----------
 1 file changed, 11 insertions(+), 11 deletions(-)
drivers/staging/mt29f_spinand/mt29f_spinand.c +11 -11
index 4484784..a0f4cbcb 100644
@@ -308,10 +308,10 @@ static int spinand_write_enable(struct spi_device *spi_nand)
 	return spinand_cmd(spi_nand, &cmd);
 }
 
-static int spinand_read_page_to_cache(struct spi_device *spi_nand, u16 page_id)
+static int spinand_read_page_to_cache(struct spi_device *spi_nand, u32 page_id)
 {
 	struct spinand_cmd cmd = {0};
-	u16 row;
+	u32 row;
 
 	row = page_id;
 	cmd.cmd = CMD_READ;
@@ -331,7 +331,7 @@ static int spinand_read_page_to_cache(struct spi_device *spi_nand, u16 page_id)
  *   locations.
  *   No tRd delay.
  */
-static int spinand_read_from_cache(struct spi_device *spi_nand, u16 page_id,
+static int spinand_read_from_cache(struct spi_device *spi_nand, u32 page_id,
 				   u16 byte_id, u16 len, u8 *rbuf)
 {
 	struct spinand_cmd cmd = {0};
@@ -362,7 +362,7 @@ static int spinand_read_from_cache(struct spi_device *spi_nand, u16 page_id,
  *   The read includes two commands to the Nand - 0x13 and 0x03 commands
  *   Poll to read status to wait for tRD time.
  */
-static int spinand_read_page(struct spi_device *spi_nand, u16 page_id,
+static int spinand_read_page(struct spi_device *spi_nand, u32 page_id,
 			     u16 offset, u16 len, u8 *rbuf)
 {
 	int ret;
@@ -430,7 +430,7 @@ static int spinand_read_page(struct spi_device *spi_nand, u16 page_id,
  *   Since it is writing the data to cache, there is no tPROG time.
  */
 static int spinand_program_data_to_cache(struct spi_device *spi_nand,
-					 u16 page_id, u16 byte_id,
+					 u32 page_id, u16 byte_id,
 					 u16 len, u8 *wbuf)
 {
 	struct spinand_cmd cmd = {0};
@@ -457,10 +457,10 @@ static int spinand_program_data_to_cache(struct spi_device *spi_nand,
  *   the Nand array.
  *   Need to wait for tPROG time to finish the transaction.
  */
-static int spinand_program_execute(struct spi_device *spi_nand, u16 page_id)
+static int spinand_program_execute(struct spi_device *spi_nand, u32 page_id)
 {
 	struct spinand_cmd cmd = {0};
-	u16 row;
+	u32 row;
 
 	row = page_id;
 	cmd.cmd = CMD_PROG_PAGE_EXC;
@@ -486,7 +486,7 @@ static int spinand_program_execute(struct spi_device *spi_nand, u16 page_id)
  *   Poll to wait for the tPROG time to finish the transaction.
  */
 static int spinand_program_page(struct spi_device *spi_nand,
-				u16 page_id, u16 offset, u16 len, u8 *buf)
+				u32 page_id, u16 offset, u16 len, u8 *buf)
 {
 	int retval;
 	u8 status = 0;
@@ -573,10 +573,10 @@ static int spinand_program_page(struct spi_device *spi_nand,
  *   one block--64 pages
  *   Need to wait for tERS.
  */
-static int spinand_erase_block_erase(struct spi_device *spi_nand, u16 block_id)
+static int spinand_erase_block_erase(struct spi_device *spi_nand, u32 block_id)
 {
 	struct spinand_cmd cmd = {0};
-	u16 row;
+	u32 row;
 
 	row = block_id;
 	cmd.cmd = CMD_ERASE_BLK;
@@ -599,7 +599,7 @@ static int spinand_erase_block_erase(struct spi_device *spi_nand, u16 block_id)
  *   and then send the 0xd8 erase command
  *   Poll to wait for the tERS time to complete the tranaction.
  */
-static int spinand_erase_block(struct spi_device *spi_nand, u16 block_id)
+static int spinand_erase_block(struct spi_device *spi_nand, u32 block_id)
 {
 	int retval;
 	u8 status = 0;
#2 lore raw
In reply to Jheng-Jhong Wu
Quoted 7 lines · Jheng-Jhong Wu

For NAND flash chips with more than 1Gbit (e.g. MT29F2G) more than 16 bits are necessary to address the correct page. The driver sets the address for more than 16 bits, but it uses 16-bit arguments and variables (these are page_id, block_id, row) to do address operations. Obviously, these arguments and variables cannot deal with more than 16-bit address.

This seems reasonable... It would be needed to make commit 6efb21d6d0e7 ("staging:mt29f_spinand: MT29F2G failing as only 16 bits used for addressing.") work. It also fixes a static checker warning.

My only concern is that the mtd/nand code seems to use -1 as a magical page_id. For example:

  2069  /**
  2070   * nand_exit_status_op - Exit a STATUS operation
  2071   * @chip: The NAND chip
  2072   *
  2073   * This function sends a READ0 command to cancel the effect of the STATUS
  2074   * command to avoid reading only the status until a new read command is sent.
  2075   *
  2076   * This function does not select/unselect the CS line.
  2077   *
  2078   * Returns 0 on success, a negative error code otherwise.
  2079   */
  2080  int nand_exit_status_op(struct nand_chip *chip)
  2081  {
  2082          struct mtd_info *mtd = nand_to_mtd(chip);
  2083  
  2084          if (chip->exec_op) {
  2085                  struct nand_op_instr instrs[] = {
  2086                          NAND_OP_CMD(NAND_CMD_READ0, 0),
  2087                  };
  2088                  struct nand_operation op = NAND_OPERATION(instrs);
  2089  
  2090                  return nand_exec_op(chip, &op);
  2091          }
  2092  
  2093          chip->cmdfunc(mtd, NAND_CMD_READ0, -1, -1);
                                   ^^^^^^^^^^^^^^      ^^
  2094  
  2095          return 0;
  2096  }
  2097  EXPORT_SYMBOL_GPL(nand_exit_status_op);

I'm not sure if this affect spinand_read_page() etc.

regards,
dan carpenter

#3 lore raw

Hi Dan,

Dan Carpenter <[email protected]> wrote on Wed, 1 Aug 2018 15:05:51 +0300:

Quoted 7 lines · Jheng-Jhong Wu
Quoted 6 lines

For NAND flash chips with more than 1Gbit (e.g. MT29F2G) more than 16 bits are necessary to address the correct page. The driver sets the address for more than 16 bits, but it uses 16-bit arguments and variables (these are page_id, block_id, row) to do address operations. Obviously, these arguments and variables cannot deal with more than 16-bit address.

Quoted 8 lines
Quoted 1 line

This seems reasonable... It would be needed to make commit 6efb21d6d0e7 ("staging:mt29f_spinand: MT29F2G failing as only 16 bits used for addressing.") work. It also fixes a static checker warning.

My only concern is that the mtd/nand code seems to use -1 as a magical page_id. For example:

I guess you missed Boris' comment: this driver is very likely to be removed. A SPI-NAND framework has been added. It does not use the raw NAND framework anymore, which was wrong anyway.

Thanks,
Miquèl

#4 lore raw
Quoted 18 lines · Miquel Raynal

Hi Dan,

Dan Carpenter <[email protected]> wrote on Wed, 1 Aug 2018 15:05:51 +0300:

Quoted 13 lines · Jheng-Jhong Wu
Quoted 7 lines

For NAND flash chips with more than 1Gbit (e.g. MT29F2G) more than 16 bits are necessary to address the correct page. The driver sets the address for more than 16 bits, but it uses 16-bit arguments and variables (these are page_id, block_id, row) to do address operations. Obviously, these arguments and variables cannot deal with more than 16-bit address.

This seems reasonable... It would be needed to make commit 6efb21d6d0e7 ("staging:mt29f_spinand: MT29F2G failing as only 16 bits used for addressing.") work. It also fixes a static checker warning.

Quoted 7 lines
Quoted 2 lines

My only concern is that the mtd/nand code seems to use -1 as a magical page_id. For example:

I guess you missed Boris' comment: this driver is very likely to be removed. A SPI-NAND framework has been added. It does not use the raw NAND framework anymore, which was wrong anyway.

We should probably revert commit 6efb21d6d0e7 ("staging:mt29f_spinand:
MT29F2G failing as only 16 bits used for addressing.") in that case if all it does is add static checker warnings...

regards,
dan carpenter

#5 lore raw

So should I change this into a revert patch, or you will revert commit 6efb21d6d0e7 ("staging:mt29f_spinand: MT29F2G failing as only 16 bits used for addressing.") by yourself?
Thanks.

Best Regards,
Jheng-Jhong Wu (Victor Wu)

#6 lore raw
Quoted 5 lines · Jheng-Jhong Wu

So should I change this into a revert patch, or you will revert commit 6efb21d6d0e7 ("staging:mt29f_spinand: MT29F2G failing as only 16 bits used for addressing.") by yourself?
Thanks.

Are you asking me? I'm not going to revert it.

regards,
dan carpenter

#7 lore raw
In reply to Dan Carpenter

Hi Dan,

Quoted 7 lines · On Wed, 1 Aug 2018 15:05:51 +0300 Dan Carpenter <[email protected]>…
Quoted 6 lines · Jheng-Jhong Wu

For NAND flash chips with more than 1Gbit (e.g. MT29F2G) more than 16 bits are necessary to address the correct page. The driver sets the address for more than 16 bits, but it uses 16-bit arguments and variables (these are page_id, block_id, row) to do address operations. Obviously, these arguments and variables cannot deal with more than 16-bit address.

Quoted 8 lines
Quoted 1 line

This seems reasonable... It would be needed to make commit 6efb21d6d0e7 ("staging:mt29f_spinand: MT29F2G failing as only 16 bits used for addressing.") work. It also fixes a static checker warning.

My only concern is that the mtd/nand code seems to use -1 as a magical page_id. For example:

Yes, -1 means "don't issue the row/page address cycles", though I don't think page can be -1 for NAND_CMD_READ{1,0} commands.

Anyway, if you want this patch merged to fix a static checker warning, I'm fine with that. In any case, I still plan to send a patch removing this driver for v4.20, so, anyone using this driver should start testing the new SPI NAND driver (drivers/mtd/nand/spi) and tweak/fix the new implementation if needed.

Regards,

Boris

#8 lore raw
Quoted 18 lines · Boris Brezillon

Hi Dan,

Quoted 13 lines · On Wed, 1 Aug 2018 15:05:51 +0300 Dan Carpenter <[email protected]>…
Quoted 7 lines · Jheng-Jhong Wu

For NAND flash chips with more than 1Gbit (e.g. MT29F2G) more than 16 bits are necessary to address the correct page. The driver sets the address for more than 16 bits, but it uses 16-bit arguments and variables (these are page_id, block_id, row) to do address operations. Obviously, these arguments and variables cannot deal with more than 16-bit address.

This seems reasonable... It would be needed to make commit 6efb21d6d0e7 ("staging:mt29f_spinand: MT29F2G failing as only 16 bits used for addressing.") work. It also fixes a static checker warning.

Quoted 6 lines
Quoted 2 lines

My only concern is that the mtd/nand code seems to use -1 as a magical page_id. For example:

Yes, -1 means "don't issue the row/page address cycles", though I don't think page can be -1 for NAND_CMD_READ{1,0} commands.

It sure looks like it can be in nand_exit_status_op()...

Quoted 5 lines

Anyway, if you want this patch merged to fix a static checker warning, I'm fine with that. In any case, I still plan to send a patch removing this driver for v4.20, so, anyone using this driver should start testing the new SPI NAND driver (drivers/mtd/nand/spi) and tweak/fix the new implementation if needed.

I don't think we should make the code more complicated than necessary just to make static checkers happy. When you say "this driver", you mean the staging driver? In that case, there is no need to revert commit 6efb21d6d0e7 ("staging:mt29f_spinand: MT29F2G failing as only 16 bits used for addressing.").

regards,
dan carpenter

#9 lore raw
Quoted 19 lines · On Mon, 6 Aug 2018 15:01:37 +0300 Dan Carpenter <[email protected]>…
Quoted 18 lines · Boris Brezillon

Hi Dan,

Quoted 13 lines · On Wed, 1 Aug 2018 15:05:51 +0300 Dan Carpenter <[email protected]>…
Quoted 7 lines · Jheng-Jhong Wu

For NAND flash chips with more than 1Gbit (e.g. MT29F2G) more than 16 bits are necessary to address the correct page. The driver sets the address for more than 16 bits, but it uses 16-bit arguments and variables (these are page_id, block_id, row) to do address operations. Obviously, these arguments and variables cannot deal with more than 16-bit address.

This seems reasonable... It would be needed to make commit 6efb21d6d0e7 ("staging:mt29f_spinand: MT29F2G failing as only 16 bits used for addressing.") work. It also fixes a static checker warning.

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

My only concern is that the mtd/nand code seems to use -1 as a magical page_id. For example:

Yes, -1 means "don't issue the row/page address cycles", though I don't think page can be -1 for NAND_CMD_READ{1,0} commands.

It sure looks like it can be in nand_exit_status_op()...

True, but nand_exit_status_op() won't be called here. It's only used when the NAND controller driver is implementing ->exec_op() and needs to do status polling, and from the micron NAND code that deals with raw NAND chips with on-die ECC (the mt29f driver is supposed to deal with SPI NANDs).

Quoted 10 lines
Quoted 6 lines

Anyway, if you want this patch merged to fix a static checker warning, I'm fine with that. In any case, I still plan to send a patch removing this driver for v4.20, so, anyone using this driver should start testing the new SPI NAND driver (drivers/mtd/nand/spi) and tweak/fix the new implementation if needed.

I don't think we should make the code more complicated than necessary just to make static checkers happy. When you say "this driver", you mean the staging driver?

Yes.

Quoted 3 lines

In that case, there is no need to revert
commit 6efb21d6d0e7 ("staging:mt29f_spinand: MT29F2G failing as only 16 bits used for addressing.").

Okay.