We've been having troubles with the reliability of the I2C interface
on the Au1550. Basically 1% of the I2C transactions would timeout.
Sometimes the failures would cluster in long runs which was causing
pain.
Last night I got annoyed with the problem enough that I applied a
portion of a Au1200 I2C patch posted by Jordan Crouse on 2005.12.02.
This patch was not applied to the the linux-mips tree (as of
2.6.16.16) but it seems to have fixed our timeout problems. I ran a
I2C test for 14 hours doing constant I2C transactions from user space
and did not see an error.
Maybe Jordan could try again with a fresh patch because it really does
seem to help...
This is the subset of the patch I used:
--- drivers/i2c/busses/i2c-au1550.c (revision 2271)
+++ drivers/i2c/busses/i2c-au1550.c (working copy)
@@ -118,13 +118,19 @@
/* Reset the FIFOs, clear events.
*/
- sp->psc_smbpcr = PSC_SMBPCR_DC;
+ stat = sp->psc_smbstat;
sp->psc_smbevnt = PSC_SMBEVNT_ALLCLR;
au_sync();
- do {
- stat = sp->psc_smbpcr;
+
+ if (!(stat & PSC_SMBSTAT_TE) || !(stat & PSC_SMBSTAT_RE)) {
+ sp->psc_smbpcr = PSC_SMBPCR_DC;
au_sync();
- } while ((stat & PSC_SMBPCR_DC) != 0);
+ do {
+ stat = sp->psc_smbpcr;
+ au_sync();
+ } while ((stat & PSC_SMBPCR_DC) != 0);
+ udelay(50);
+ }
/* Write out the i2c chip address and specify operation
*/
