Seagate NCQ and Sata_Sil issues on 2.6.12
From: Roger Heflin <hidden>
Date: 2005-06-21 21:24:36
Hello,
I have been trying to get a Sata SIL 3114 controller to work with
seagate 200GB disks (NCQ) on 2.6.12, with it has the boot disk,
I get massive pauses while a dd is running off of the boot disk,
the pauses are bad enough that typing the login name, results in
the password prompt never coming up (or at least not coming up
in 30+ seconds) without the dd being stopped, stopping the dd
results in the password prompt coming up. The behaviour is also
seen when you run any command in another window that is not expected
to be in the disk cache, the command hangs until the dd is stopped,
the performance is bad enough that I would expect the performance
to be unacceptable/non-existant under high disk loads.
Removing the disk from the blacklist appears to correct the issue,
this is with the disk not booting, I did not test the disk
booting/installing with it removed from the blacklist.
The behavior does not seem to happen with the fedora core 3 kernel
in non-smp mode, thought the fedora core 3 kernel in smp mode has
other serious issues with our setup that results in us not being
able to test fc3 smp mode.
Here is the output from vmstat 1 for a WD and a Seagate drive with
the dd running:
The 19xxx runs are the seagate disk, the 58xxx runs are the Western disks.
The Western digital disk is roughly 3x as fast as the seagate
with 1/4 the interrupts, and with 1/10 of the context switches.
This agrees with what others have seen around the seagates.
Seagate stock 2.6.12 kernel:
0 1 0 6008384 72060 1160860 0 0 19840 0 5985 10000 0 9
75 16
0 1 0 5988048 92028 1160912 0 0 19968 0 5989 10006 0 9
75 16
0 1 0 5967588 111996 1160964 0 0 19968 0 5984 9994 0 9
75 16
0 1 0 5947252 131836 1160884 0 0 19840 0 5989 10014 0 9
75 16
0 1 0 5926792 151808 1160672 0 0 19968 28 5986 10003 0 9
75 16
0 1 0 5906332 171776 1160724 0 0 19968 0 5982 9999 0 9
75 16
0 1 0 5885996 191616 1160904 0 0 19840 0 5985 9995 0 9
75 16
0 1 0 5865412 211584 1160696 0 0 19968 0 5986 10009 0 9
75 16
0 1 0 5844952 231552 1160748 0 0 19968 0 5985 9992 0 9
75 16
0 1 0 5824740 251392 1160668 0 0 19840 0 5985 10000 0 9
75 16
0 1 0 5804280 271360 1160720 0 0 19968 0 5988 10005 0 9
75 16
0 1 0 5783820 291328 1160772 0 0 19968 0 5982 9993 0 9
75 16
1 1 0 5763484 311168 1160692 0 0 19840 0 5982 9990 0 9
75 16
0 1 0 5743148 331136 1161004 0 0 19968 0 5978 9986 0 9
75 16
1 1 0 5722812 350976 1160924 0 0 19840 0 5984 9987 0 9
75 16
0 1 0 5702352 370944 1160716 0 0 19968 0 5979 9994 0 9
75 16
0 1 0 5682016 390784 1160896 0 0 19840 0 5980 9993 0 9
75 16
0 1 0 5661556 410752 1160688 0 0 19968 0 5982 9999 0 9
75 16
0 1 0 5641220 430592 1160868 0 0 19840 0 5981 9988 0 9
75 16
1 1 0 5620760 450560 1160660 0 0 19968 0 5979 9990 0 9
75 16
0 0 0 6037780 43168 1161152 0 0 6272 0 2637 3187 0 7
88 5
0 0 0 6037780 43168 1161152 0 0 0 0 1032 138 0 0
100 0
Western Digital 2.6.12 kernel:
0 1 0 6036532 44444 1161176 0 0 1276 0 1026 94 0 0
99 1
1 0 0 5976516 103196 1161184 0 0 58752 0 1465 957 1 6
75 19
0 1 0 5915756 162460 1161200 0 0 59264 0 1467 955 0 6
75 18
1 0 0 5854996 221860 1161340 0 0 59436 32 1473 979 1 5
75 19
0 1 0 5794484 280868 1161092 0 0 58964 0 1465 951 0 6
75 19
0 1 0 5733476 340388 1161112 0 0 59648 0 1471 979 1 6
75 18
0 1 0 5672592 399908 1161132 0 0 59392 0 1469 964 0 6
75 18
0 1 0 5611584 459428 1161152 0 0 59520 0 1469 970 0 6
75 18
0 1 0 5550576 518948 1161172 0 0 59520 0 1469 960 0 6
75 18
0 1 0 5489692 578468 1161192 0 0 59520 0 1469 972 1 6
75 18
1 0 0 5428808 637860 1161080 0 0 59392 0 1469 963 0 6
75 18
1 0 0 5367676 697380 1161100 0 0 59520 0 1469 975 0 6
75 18
0 1 0 5306544 757028 1160992 0 0 59648 0 1470 967 1 6
75 19
0 1 0 5245784 816420 1161140 0 0 59392 0 1468 984 0 6
75 19
Here is output with the Seagate being removed from the blacklist
in the 2.6.12 kernel:
1 0 0 34560 7065132 20388 0 0 54912 0 1220 511 1 6 74
19
0 1 0 36420 7063468 20232 0 0 54912 0 1219 517 0 7 74
19
0 1 0 34064 7065792 20508 0 0 54784 0 1218 512 0 6 74
19
0 1 0 35676 7064236 20244 0 0 55040 0 1222 512 1 7 74
19
0 1 0 32948 7066796 20024 0 0 54784 0 1219 506 0 6 74
19
1 0 0 34560 7065156 20364 0 0 54912 0 1219 511 1 6 74
19
2 1 0 35304 7064316 20424 0 0 54912 0 1218 511 0 6 74
19
0 1 0 33816 7066084 19956 0 0 54784 0 1221 506 0 7 74
19
0 1 0 35304 7064552 20188 0 0 55040 0 1221 520 1 6 74
19
1 0 0 33568 7066212 20348 0 0 54016 0 1215 499 0 6 74
19
I did not get any crashes in the limited testing the we did, this seems to
indicate the the theory Jeff had about the blacklist growing incorrectly
may be correct.
The high/screaming interrupt/context switch issue seems to be related to the
blacklist issue since removing the disk from the blacklist seemed to make
that go away.
Roger Heflin
Atipa Technologies