Sorry about the long delay. On Thu, Jun 30, 2011 at 05:53:32PM +0000, Derry Bryson wrote: > I have included info from kern.log below showing turning the bay on and off before and > after I patched the kernel. kern.log tends to be too cluttered with extra timestamps. Can you please use 'dmesg -c' after each phase of testing? printk timestamps included there should be enough. > I first applied the patch from your previous email and changed the > second timing value to 1000 and that makes it work. It also works > if you leave the timing values alone and up the retries (i.e > ATA_EH_PMP_LINK_TRIES) from 3 to 5. It seems to me the drives are > taking a long (relatively) time to spin up and either way all we are > doing is giving it more time to spin up. That's debouncing timing. It doesn't have much to do with spinning up. Spinning up can take over ten seconds. PHY is usually ready well under a sec once power is applied. The reason why libata stops waiting during reset is because PHY flickers - it comes up and then goes out again. libata EH already debounces to work around these glitches but it seems this 'flickering' is larger scale than libata parameters expect. One suspicion I have is that the PSU in the enclosure isn't stable enough to maintain PHY state while multiples drives are powering up. If this is the case, PHYs may keep flickering well over a second which libata doesn't expect them to. Beefing up the PSU (or using a second PSU to power some of the harddrives) and see whether anything changes would be a good way to test it. > Is there some way to know it is spinning up and wait for that rather > than just trying to reset the controller over and over? Also I > notice from the kernel log that the 'hotplugged' flag is only set > the first time it does the hard reset and is then cleared. If this > didn't get cleared it may work that way as well. All of this only > fixes the problem until an slower drive comes out. The hotplugged timing is supposed to kick in only once after a hotplug event as some PHYs tend to take longer time to lock on after hotplug event. > From the SMART info on the drives for the WD 3TB that fails the spin > up value was 188 vs. a Seagate 160GB that was 87. I believe these > values are in milliseconds so we can see that the drive that fails > takes much longer to spin up. That's more likely centisecs instead of millisecs. 8.7sec would be about normal for regular drives. 18.8 isn't too far off for large ones with more platters. The more important thing probably is that the WD 3TB drive is likely to draw much more power than the smaller one taxing the PSU in the enclosure which could have been designed to have just enough power for more regular drives. > Jun 30 10:02:26 HR-NETSWAP kernel: [ 99.510350] ata1.00: XXXX hardreset hotplugged = false > Jun 30 10:02:26 HR-NETSWAP kernel: [ 99.510356] ata1.00: XXX0 hardreset debounce 5 100 2000 > Jun 30 10:02:26 HR-NETSWAP kernel: [ 99.510361] ata1.00: XXX1 hardreset debounce 5 100 2000 So if you bump the second timing value to 1000, it works without retrying? Can you please post kernel log w/ that change? Thanks. -- tejun -- To unsubscribe from this list: send the line "unsubscribe linux-ide" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
