Skip to contentSkip to conversation

mtd: brcmnand: clean up flash cache for parameter pages

PATCH Applied Brian Norris 2 participants 3 messages Nov 17, 2015 – Nov 18, 2015

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

PATCH 1/1 mtd: brcmnand: clean up flash cache for parameter pages

The read_byte() handling for accessing the flash cache has some awkward swapping being done in the read_byte() function. Let's just make this a byte array, and do the swapping with the word-level macros during the initial buffer copy.

This is just a refactoring patch, with no (intended) functional change.

---
I wanted to get this out there as a proper patch, since there was some confusion on the previous thread in which this appeared.

This does *not* fix the bug reported by Cyrille in the thread "mtd: brcmnand:
Fix NAND_CMD_PARAM byte order". There is a candidate patch which fixes this:

http://patchwork.ozlabs.org/patch/535330/

But I'm awaiting a better explanation from Broadcom folks before I respin/accept anything like that.

 drivers/mtd/nand/brcmnand/brcmnand.c | 13 +++++++++----
 1 file changed, 9 insertions(+), 4 deletions(-)
drivers/mtd/nand/brcmnand/brcmnand.c +9 -4
index 2a437c7ed175..0f43bc95ece4 100644
@@ -134,7 +134,7 @@ struct brcmnand_controller {
 	dma_addr_t		dma_pa;
 
 	/* in-memory cache of the FLASH_CACHE, used only for some commands */
-	u32			flash_cache[FC_WORDS];
+	u8			flash_cache[FC_BYTES];
 
 	/* Controller revision details */
 	const u16		*reg_offsets;
@@ -1188,6 +1188,8 @@ static void brcmnand_cmdfunc(struct mtd_info *mtd, unsigned command,
 
 	if (native_cmd == CMD_PARAMETER_READ ||
 			native_cmd == CMD_PARAMETER_CHANGE_COL) {
+		/* Copy flash cache word-wise */
+		u32 *flash_cache = (u32 *)ctrl->flash_cache;
 		int i;
 
 		brcmnand_soc_data_bus_prepare(ctrl->soc);
@@ -1197,7 +1199,11 @@ static void brcmnand_cmdfunc(struct mtd_info *mtd, unsigned command,
 		 * SECTOR_SIZE_1K may invalidate it
 		 */
 		for (i = 0; i < FC_WORDS; i++)
-			ctrl->flash_cache[i] = brcmnand_read_fc(ctrl, i);
+			/*
+			 * Flash cache is big endian for parameter pages, at
+			 * least on STB SoCs
+			 */
+			flash_cache[i] = be32_to_cpu(brcmnand_read_fc(ctrl, i));
 
 		brcmnand_soc_data_bus_unprepare(ctrl->soc);
 
@@ -1250,8 +1256,7 @@ static uint8_t brcmnand_read_byte(struct mtd_info *mtd)
 		if (host->last_byte > 0 && offs == 0)
 			chip->cmdfunc(mtd, NAND_CMD_RNDOUT, addr, -1);
 
-		ret = ctrl->flash_cache[offs >> 2] >>
-					(24 - ((offs & 0x03) << 3));
+		ret = ctrl->flash_cache[offs];
 		break;
 	case NAND_CMD_GET_FEATURES:
 		if (host->last_byte >= ONFI_SUBFEATURE_PARAM_LEN) {
Signature · 1 line
2.6.0.rc2.230.g3dd15c0
#1 lore raw
Quoted 17 lines · Brian Norris

The read_byte() handling for accessing the flash cache has some awkward swapping being done in the read_byte() function. Let's just make this a byte array, and do the swapping with the word-level macros during the initial buffer copy.

This is just a refactoring patch, with no (intended) functional change.

---
I wanted to get this out there as a proper patch, since there was some confusion on the previous thread in which this appeared.

Quoted 7 lines

This does *not* fix the bug reported by Cyrille in the thread "mtd: brcmnand:
Fix NAND_CMD_PARAM byte order". There is a candidate patch which fixes this:

   http://patchwork.ozlabs.org/patch/535330/

But I'm awaiting a better explanation from Broadcom folks before I respin/accept anything like that.

Yeah still on our to-do list. Will find time to talk to our ASIC engineer.

Quoted 41 lines
  drivers/mtd/nand/brcmnand/brcmnand.c | 13 +++++++++----
  1 file changed, 9 insertions(+), 4 deletions(-)
drivers/mtd/nand/brcmnand/brcmnand.c +8 -2
index 2a437c7ed175..0f43bc95ece4 100644
@@ -134,7 +134,7 @@ struct brcmnand_controller {
  	dma_addr_t		dma_pa;
 
  	/* in-memory cache of the FLASH_CACHE, used only for some commands */
-	u32			flash_cache[FC_WORDS];
+	u8			flash_cache[FC_BYTES];
 
  	/* Controller revision details */
  	const u16		*reg_offsets;
@@ -1188,6 +1188,8 @@ static void brcmnand_cmdfunc(struct mtd_info *mtd, unsigned command,
 
  	if (native_cmd == CMD_PARAMETER_READ ||
  			native_cmd == CMD_PARAMETER_CHANGE_COL) {
+		/* Copy flash cache word-wise */
+		u32 *flash_cache = (u32 *)ctrl->flash_cache;
  		int i;
 
  		brcmnand_soc_data_bus_prepare(ctrl->soc);
@@ -1197,7 +1199,11 @@ static void brcmnand_cmdfunc(struct mtd_info *mtd, unsigned command,
  		 * SECTOR_SIZE_1K may invalidate it
  		 */
  		for (i = 0; i < FC_WORDS; i++)
-			ctrl->flash_cache[i] = brcmnand_read_fc(ctrl, i);
+			/*
+			 * Flash cache is big endian for parameter pages, at
+			 * least on STB SoCs
+			 */
+			flash_cache[i] = be32_to_cpu(brcmnand_read_fc(ctrl, i));
 
  		brcmnand_soc_data_bus_unprepare(ctrl->soc);
 

drivers/mtd/nand/brcmnand/brcmnand.c static uint8_t brcmnand_read_byte(struct mtd_info *mtd) · line 1262

Quoted 9 lines
drivers/mtd/nand/brcmnand/brcmnand.c +1 -2
@@ -1250,8 +1256,7 @@ static uint8_t brcmnand_read_byte(struct mtd_info *mtd)
  		if (host->last_byte > 0 && offs == 0)
  			chip->cmdfunc(mtd, NAND_CMD_RNDOUT, addr, -1);
 
-		ret = ctrl->flash_cache[offs >> 2] >>
-					(24 - ((offs & 0x03) << 3));
+		ret = ctrl->flash_cache[offs];
  		break;
  	case NAND_CMD_GET_FEATURES:
  		if (host->last_byte >= ONFI_SUBFEATURE_PARAM_LEN) {

"Tested" on Cygnus, and as expected, nothing changes. NAND still works and ONFI still fails.

Thanks,

Ray

author Applied
#2 lore raw
In reply to Brian Norris
Quoted 5 lines · Brian Norris

The read_byte() handling for accessing the flash cache has some awkward swapping being done in the read_byte() function. Let's just make this a byte array, and do the swapping with the word-level macros during the initial buffer copy.

Quoted 8 lines

This is just a refactoring patch, with no (intended) functional change.

Pushed to l2-mtd.git