On Thursday, February 10, 2011 07:24:54 pm Aaron Williams wrote:
On Thursday, February 10, 2011 07:08:01 am Andrew Dyer wrote:
On Thu, Feb 10, 2011 at 00:28, Aaron Williams
Aaron.Williams@caviumnetworks.com wrote:
I would suggest to make sure any caching/prefetching stuff is off, hardware is doing one flash bus access per CPU read/write.
In cmdset_amd_read_jedec_ids() after this line:
manuId = flash_read_uchar (info, FLASH_OFFSET_MANUFACTURER_ID);
add something like
printf("test manf id word = %04x\n", *(volatile uint16_t *)0x1f400000); printf("test device id word = %04x\n", *(volatile uint16_t *)0x1f400002); printf("test device id word = %04x\n", *(volatile uint16_t *)0x1f40001c); printf("test device id word = %04x\n", *(volatile uint16_t *)0x1f40001e);
and see what you get.
fwc addr 1f400000 cmd f0 f0f0 16bit x 8 bit flash_write16: address: 1f400000, value: 0xf0f0 fwc addr 1f401554 cmd aa aaaa 16bit x 8 bit flash_write16: address: 1f401554, value: 0xaaaa fwc addr 1f400aaa cmd 55 5555 16bit x 8 bit flash_write16: address: 1f400aaa, value: 0x5555 fwc addr 1f401554 cmd 90 9090 16bit x 8 bit flash_write16: address: 1f401554, value: 0x9090 flash_read8: address: 1f400001, value: 0x0 test manf id word = 1000 test device id word = 013f test device id word = da6c test device id word = 2926
looks like garbage :-( What's in the flash at those addresses? Maybe something is happening to mess up the unlock sequence and you're reading the memory data instead of the device codes.
It's odd that earlier in the sequence when the CFI query data is read the byte data is mirrored across both bytes of the response, here the two bytes are different.
I received an email back from Spansion about this problem. They suggested that instead of the following:
fwc addr 1f401554 cmd aa aaaa 16bit x 8 bit flash_write16: Wrote 0xaaaa to address 1f401554 funlock writing 0xaa to address 0x555 fwc addr 1f400aaa cmd 55 5555 16bit x 8 bit flash_write16: Wrote 0x5555 to address 1f400aaa fwc addr 1f401554 cmd 90 9090 16bit x 8 bit flash_write16: Wrote 0x9090 to address 1f401554
to instead do the following: write 0xAA to 0x1F400AAA write 0x55 to 0x1F400555 write 0x90 to 0x1F400AAA
read 0x1F400001 returns 0x01 read 0x1F400003 returns 0x7e
This looks correct.
I found two things I need to do to work around the problem. First, in flash_write_cmd I use info->chipwidth instead of info->portwidth for the write commands, and second, in __flash_detect_cfi I don't modify the address in compatibility mode. This works. If either of those steps are left out then reading the manufacturer ID fails.
-Aaron
It looks like I spoke too soon. It fixes the manufacturer problem but everything else is now broken.
-Aaron