This patch takes care of a problem with function write_buff in drivers/cfi_flash.c driver when the CFG_FLASH_USE_BUFFER_WRITE setting is defined.

---------------------

Description of problem: When writing to a flash buffer, only a multiple of the byte size of the flash chip port width is written. For example, if attempting to write 59 bytes when the port width is 16 bits, only 58 bytes get written to the buffer. The remaining byte is left to be handled as an individual byte write. However, the write_buff in cfi_flash.c did not take into account the rounding down of the number of bytes written with the buffered write. As a result, the handling of unaligned tail bytes would see 0 in cnt and think the full amount of data had been written.

---------------------

The fix: The variable i is rounded down to the nearest multiple of the portwidth before incrementing/decrementing the source/destination/size variables.

---------------------

Log Message: * Patch by Michael Bendzick, 15 Jul 2004: Fixes to drivers/cfi_flash.c: - Handling of writes with sizes that are not multiples of the byte size of the flash chip port width when CFG_FLASH_USE_BUFFER_WRITE is defined. - Fix minor comment typo.

---------------------

The patch:

Index: CHANGELOG =================================================================== RCS file: /cvsroot/u-boot/u-boot/CHANGELOG,v retrieving revision 1.342 diff -p -u -r1.342 CHANGELOG --- CHANGELOG 12 Jul 2004 22:34:51 -0000 1.342 +++ CHANGELOG 15 Jul 2004 15:26:24 -0000 @@ -2,6 +2,13 @@ Changes since U-Boot 1.1.1: ======================================================================

+* Patch by Michael Bendzick, 15 Jul 2004: + Fixes to drivers/cfi_flash.c: + - Handling of writes with sizes that are not multiples of the byte + size of the flash chip port width when CFG_FLASH_USE_BUFFER_WRITE + is defined. + - Fix minor comment typo. + * Patch by Michael Bendzick, 12 Jul 2004: fix output formatting in drivers/cfi_flash.c

Index: drivers/cfi_flash.c =================================================================== RCS file: /cvsroot/u-boot/u-boot/drivers/cfi_flash.c,v retrieving revision 1.10 diff -p -u -r1.10 cfi_flash.c --- drivers/cfi_flash.c 12 Jul 2004 22:34:52 -0000 1.10 +++ drivers/cfi_flash.c 15 Jul 2004 15:26:37 -0000 @@ -517,6 +517,7 @@ int write_buff (flash_info_t * info, uch i = buffered_size > cnt ? cnt : buffered_size; if ((rc = flash_write_cfibuffer (info, wp, src, i)) != ERR_OK) return rc; + i -= (i % info->portwidth); wp += i; src += i; cnt -= i; @@ -1231,5 +1232,5 @@ static int flash_write_cfibuffer (flash_ flash_write_cmd (info, sector, 0, FLASH_CMD_CLEAR_STATUS); return retcode; } -#endif /* CFG_USE_FLASH_BUFFER_WRITE */ +#endif /* CFG_FLASH_USE_BUFFER_WRITE */ #endif /* CFG_FLASH_CFI */

---------------------

Example output demonstrating the problem & solution:

(Tests performed on TI OMAP1510 Innovator, aka board/omap1510inn, with U-Boot build modified to use the cfi_flash.c driver. The file 59-bytes is 59 bytes grabbed from /dev/urandom.)

[Old Driver] OMAP1510 Innovator # erase 1:2 Erase Flash Sectors 2-2 in Bank # 1 . done OMAP1510 Innovator # tftp 10000000 59-bytes TFTP from server 192.168.1.177; our IP address is 192.168.1.249 Filename '59-bytes'. Load address: 0x10000000 Loading: # done Bytes transferred = 59 (3b hex) OMAP1510 Innovator # cp.b 10000000 40000 $(filesize) Copy to Flash... done OMAP1510 Innovator # cmp.b 10000000 40000 $(filesize) byte at 0x1000003a (0x17) != byte at 0x0004003a (0xff) Total of 58 bytes were the same

[New Driver] 1510 Innov. JMP10a # erase 1:2 Erase Flash Sectors 2-2 in Bank # 1 . done OMAP1510 Innovator # tftp 10000000 59-bytes TFTP from server 192.168.1.177; our IP address is 192.168.1.249 Filename '59-bytes'. Load address: 0x10000000 Loading: # done Bytes transferred = 59 (3b hex) OMAP1510 Innovator # cp.b 10000000 40000 $(filesize) Copy to Flash... done OMAP1510 Innovator # cmpb. 10000000 40000 $(filesize) Unknown command 'cmpb.' - try 'help' OMAP1510 Innovator # cmp.b 10000000 40000 $(filesize) Total of 59 bytes were the same

---------------------

Testing: The following boards have CFG_FLASH_CFI_DRIVER and CFG_FLASH_USE_BUFFER_WRITE defined, making them the only boards to be affected by my patch. "OK" below indicates that it compiled properly. I don't have actual hardware to test though.

assabet OK csb272 OK csb472 OK eXalion Not even unpatched from CVS compiled for me gcplus OK ppmc8260 OK sbc8560 ELDK didn't appear to compile for PPC 85xx, so this board remains untested. If someone can point me to a pre-made 85xx toolchain, I would gladly test it out.

-Michael Bendzick Systems and Software Engineering Logic Product Development michael.b@logicpd.com 612-436-5122 www.logicpd.com