On Wed, 11 Jul 2007, Jeff Garzik wrote: > This patch is diff'd against virgin 2.6.22 kernel, and does not require > any other patches (and indeed no other sata_mv patches should be applied > alongside this one). > > Thanks to dean's feedback (and similar feedback from Dave), I think I > killed two key sata_mv bugs, either of which could cause the WARNING > spew being reported. In sum, sata_mv was not doing a good job of > managing the software-managed request and response queue pointers. > > Testing feedback with 2.6.22 + this patch would be appreciated. It > works on my 6041 (Gen-II) and 5081 (Gen-I). Haven't yet tried a Gen-IIE > chip (6042/7042-compatible). oh interesting... with this one i still get lots of errors during boot... but i can conditionally hit 70MB/s+ i/o rates on one drive at a time. see attached log for boot errors. i went through a series like this: 1. dd if=/dev/sda of=/dev/null & 2. dd if=/dev/sdb of=/dev/null & 3. then i noticed errors in the log and i killed both those tasks 4. dd if=/dev/sdc of=/dev/null & 5. "iostat -kx /dev/sd? 5" observe 70MB/s read rate on sdc 5. kill dd from step 4 6. dd if=/dev/sdd of=/dev/null & 7. "iostat -kx /dev/sd? 5" observe 70MB/s read rate on sdd 8. kill dd from step 6 9. dd if=/dev/sdb of=/dev/null & 10. "iostat -kx /dev/sd? 5" observe 70MB/s read rate on sdb 11. kill dd from step 6 12. dd if=/dev/sdc of=/dev/null & 13. dd if=/dev/sdd of=/dev/null & 14. massive errors in the logs it looks like doing reads on two of the drives on the 6081 causes mass errors in the logs... also if i use "iflag=direct" on the dd i see only ~12MB/s i/o, but that could be entirely unrelated. -dean
Attachment:
log.31.20070711_0126.bz2
Description: Binary data
