Skip to contentSkip to conversation

Sluggishness in 2.6.7 caused by IDE stack

DISCUSSION Scott A Crosby 2 participants 3 messages Sep 25, 2004 – Oct 2, 2004

lore
Share & tools
lore
#0 lore raw

I was burning a DVD-R on a new brand of media and I was experience fairly high system sluggishness. I've burned other DVD's on a different brand of media on this drive at a slower speed without this effect.

The system is an Athlon XP 2500+ running 2.6.7 with a Liteon 832s dvd burner.

Much of the CPU time was spent in system mode. I setup a quick oprofile, which blamed the function task_no_data_intr, but an opannotate reports confusing results, possibly from interrupts? dmesg reported nothing interesting.

Scott

*** vmstat output ***

procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu----

 1  1 410496   8520  14224 1253716    0    0  2180    16  409   633 10 43  0 48
 7  1 410496   6920  14236 1255380    0    0  1668    20  589   832  9 18  0 73
 3  0 410496   8264  14232 1253988    0    0  2436    80  334   438 18 82  0  0
 0  1 410496   8384  14244 1253860    0    0  1932     0  464   706 13 35  0 52
 5  1 410496   8056  14264 1253872  328    0  2280     0  717   965 14 20  0 65
 8  1 410496   7352  14268 1254640    0    0  2692     0  351   485 17 83  0  0
 2  2 410496   6968  14264 1255036   32    0  2336     0  332   522 17 83  0  0
 5  0 410496   8568  14276 1253384    0    0  2192    12  464   794 19 33  0 48

*** oprofile output ***

samples  %        app name                 symbol name
1217485  94.2994  vmlinux                  task_no_data_intr
7714      0.5975  firefox-bin              (no symbols)
5975      0.4628  libc-2.3.2.so            (no symbols)
5199      0.4027  vmlinux                  pci_bus_match_dynids
3798      0.2942  XFree86                  (no symbols)

*** opreport -d output ***

c022f520 1218267 89.3898 task_no_data_intr
c022f530 2 1.6e-04

  c022f535 30        0.0025
  c022f53a 2186      0.1794
  c022f54a 90        0.0074

c022f561 2 1.6e-04
c022f56d 1196860 98.2428
c022f570 19037 1.5626
c022f571 1 8.2e-05
c022f584 2 1.6e-04
c022f589 57 0.0047

*** Relevant assembly code in task_no_data_intr ***

c022f520 <task_no_data_intr>: /* task_no_data_intr total: 1218267 89.3898 */

               :c022f520:       sub    $0x1c,%esp
               :c022f523:       mov    %edi,0x14(%esp,1)
               :c022f527:       mov    %ebx,0xc(%esp,1)
               :c022f52b:       mov    %esi,0x10(%esp,1)
               :c022f52f:       mov    0x20(%esp,1),%edi
               :c022f533:       mov    %ebp,0x18(%esp,1)
               :c022f537:       mov    0x70(%edi),%esi
  2186  0.1604 :c022f53a:       mov    0x8(%esi),%eax
               :c022f53d:       mov    0x20(%eax),%eax
               :c022f540:       mov    0x68(%eax),%ebp
               :c022f543:       sti    
               :c022f544:       mov    0x70(%edi),%eax
               :c022f547:       mov    0x34(%eax),%eax
    90  0.0066 :c022f54a:       mov    %eax,(%esp,1)
               :c022f54d:       call   *0x4b8(%esi)
               :c022f553:       movzbl %al,%ebx
               :c022f556:       mov    %ebx,%eax
               :c022f558:       and    $0xc9,%eax
               :c022f55d:       cmp    $0x40,%eax
               :c022f560:       je     c022f58b <task_no_data_intr+0x6b>
               :c022f562:       mov    0x1c(%edi),%eax
               :c022f565:       mov    %edi,(%esp,1)
               :c022f568:       movl   $0xc02dfecf,0x4(%esp,1)
 19037  1.3968 :c022f570:       mov    %ebx,0x8(%esp,1)
               :c022f574:       call   *0x20(%eax)
               :c022f577:       mov    0xc(%esp,1),%ebx
               :c022f57b:       mov    0x10(%esp,1),%esi
               :c022f57f:       mov    0x14(%esp,1),%edi
               :c022f583:       mov    0x18(%esp,1),%ebp
               :c022f587:       add    $0x1c,%esp
               :c022f58a:       ret    
               :c022f58b:       test   %ebp,%ebp
               :c022f58d:       je     c022f5b1 <task_no_data_intr+0x91>
               :c022f58f:       mov    0x70(%edi),%eax
               :c022f592:       mov    0x1c(%eax),%eax
               :c022f595:       mov    %eax,(%esp,1)
               :c022f598:       call   *0x4b8(%esi)
               :c022f59e:       mov    %edi,(%esp,1)
               :c022f5a1:       mov    %ebx,0x4(%esp,1)
               :c022f5a5:       movzbl %al,%eax
               :c022f5a8:       mov    %eax,0x8(%esp,1)
               :c022f5ac:       call   c0229b90 <ide_end_drive_cmd>
               :c022f5b1:       xor    %eax,%eax
               :c022f5b3:       jmp    c022f577 <task_no_data_intr+0x57>
               :c022f5b5:       lea    0x0(%esi,1),%esi
               :c022f5b9:       lea    0x0(%edi,1),%edi
               :

** Source code for function ***
** drivers/ide/ide-taskfile.c:278

ide_startstop_t task_no_data_intr (ide_drive_t *drive)

{
        ide_task_t *args        = HWGROUP(drive)->rq->special;
        ide_hwif_t *hwif        = HWIF(drive);
        u8 stat;

        local_irq_enable();
        if (!OK_STAT(stat = hwif->INB(IDE_STATUS_REG),READY_STAT,BAD_STAT)) {
                DTF("%s: command opcode 0x%02x\n", drive->name,
                        args->tfRegister[IDE_COMMAND_OFFSET]);
                return DRIVER(drive)->error(drive, "task_no_data_intr", stat);
                /* calls ide_end_drive_cmd */
        }
        if (args)
                ide_end_drive_cmd(drive, stat, hwif->INB(IDE_ERROR_REG));

        return ide_stopped;
}
#1 lore raw
Quoted 12 lines · Scott A Crosby

Much of the CPU time was spent in system mode. I setup a quick oprofile, which blamed the function task_no_data_intr, but an opannotate reports confusing results, possibly from interrupts? dmesg reported nothing interesting.

Scott

*** vmstat output ***

procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu----

Quoted 8 lines
 1  1 410496   8520  14224 1253716    0    0  2180    16  409   633 10 43  0 48
 7  1 410496   6920  14236 1255380    0    0  1668    20  589   832  9 18  0 73
 3  0 410496   8264  14232 1253988    0    0  2436    80  334   438 18 82  0  0
 0  1 410496   8384  14244 1253860    0    0  1932     0  464   706 13 35  0 52
 5  1 410496   8056  14264 1253872  328    0  2280     0  717   965 14 20  0 65
 8  1 410496   7352  14268 1254640    0    0  2692     0  351   485 17 83  0  0
 2  2 410496   6968  14264 1255036   32    0  2336     0  332   522 17 83  0  0
 5  0 410496   8568  14276 1253384    0    0  2192    12  464   794 19 33  0 48

I think you have the same problem as me. The interrupt rate drops under 1000 during use of the DVD and that is strange as the HZ is 1000 and that should be the lowest possible value unless I misunderstood something.

#2 lore raw
Quoted 22 lines · Kenneth Johansson
Quoted 12 lines · Scott A Crosby

Much of the CPU time was spent in system mode. I setup a quick oprofile, which blamed the function task_no_data_intr, but an opannotate reports confusing results, possibly from interrupts? dmesg reported nothing interesting.

Scott

*** vmstat output ***

procs -----------memory---------- ---swap-- -----io---- --system-- ----cpu----

Quoted 8 lines
 1  1 410496   8520  14224 1253716    0    0  2180    16  409   633 10 43  0 48
 7  1 410496   6920  14236 1255380    0    0  1668    20  589   832  9 18  0 73
 3  0 410496   8264  14232 1253988    0    0  2436    80  334   438 18 82  0  0
 0  1 410496   8384  14244 1253860    0    0  1932     0  464   706 13 35  0 52
 5  1 410496   8056  14264 1253872  328    0  2280     0  717   965 14 20  0 65
 8  1 410496   7352  14268 1254640    0    0  2692     0  351   485 17 83  0  0
 2  2 410496   6968  14264 1255036   32    0  2336     0  332   522 17 83  0  0
 5  0 410496   8568  14276 1253384    0    0  2192    12  464   794 19 33  0 48
Quoted 4 lines

I think you have the same problem as me. The interrupt rate drops under 1000 during use of the DVD and that is strange as the HZ is 1000 and that should be the lowest possible value unless I misunderstood something.

I did some more testing and it turns out that I was not using the via IDE driver but some generic code. I had the IDE driver as a module and it was loaded but it did only drive the two unused ports the generic driver had control over the DVD drive. I removed CONFIG_IDE_GENERIC and compiled the via driver into the kernel and everything work OK now.

I wonder why the generic code blocks interrupts but now I'm not that interested in this problem anymore :)