On Thu, Oct 1, 2015 at 5:50 PM, Wolfgang Denk wd@denx.de wrote:
On ARM (a LE architecture), clrsetbits_le16() maps down into:
clrsetbits_le16 -> out_le16 / in_le16 -> out_arch, w,le16 / in_arch, w,le16 -> __raw_writew(cpu_to_le16()) / le16__to_cpu(__raw_readw()) -> __raw_writew() / __raw_readw()while
writew() -> __raw_writew(cpu_to_le16(v),__mem_pci(c)) __raw_writew()Both map into __raw_writew() [which then boild down into __arch_putw() which is just a volatile unsigned short write access.
So both clrsetbits_le16() and writew() are little endian accessors. In which way would one write other data to the device than the other?
Yes, you are right.
The issue seems to be related to the effect of writing doing 'writew(WCR_WDE, &wdog->wcr)', which writes 0x40 to the register WCR versus 'clrsetbits_le16(&wdog->wcr, 0, WCR_WDE)' which only enables the WDE bit.
The kernel driver also does the full write to the WCR register(like we used to have prior to 623d96e89aca6) https://git.kernel.org/cgit/linux/kernel/git/stable/linux-stable.git/tree/dr...
This has also the effect of clearing the SRS bit.
By the way we need to implement erratum ERR004346 (WDOG: WDOG SRS bit requires to be written twice) in U-boot, but this is an unrelated issue.
So the revert seems to be appropriate. The commit log should be adjusted though.
Regards,
Fabio Estevam