loading EHCI_HCD slows down IDE disk performance by 50%

loading EHCI_HCD slows down IDE disk performance by 50%

[Alessandro Suardi]

Quite a while ago I filed

https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=188630

 as it looked like moving from FC3 to FC5 caused a 40%
 drop in IDE disk performance.

After finally finding (a LOT of) time to dig into the issue,
 turns out it's a kernel problem that currently can be
 described as a 50% performance hit for IDE disks, and
 that only shows up when loading the USB2.0 ehci_hcd
 module (I originally didn't have an external USB2.0 disk
 and bought it at the same time I moved from FC3 to FC5).


The problem traces back at least to 2.6.16-rc5-git8, which
 was the kernel du jour at the time I originally reported
 the issue on lkml:

http://www.ussg.iu.edu/hypermail/linux/kernel/0604.1/0046.html

 and is still present in 2.6.19-rc3-git4 (running both FC3
 and FC5) and the current FC6 kernel.

Unfortunately, the uhci_hcd module which also can
 speak to my external USB2.0 disk is too slow to be used
 in place of ehci_hcd... like this:

[root@donkey ~]# uname -a
Linux donkey 2.6.18-1.2798.fc6 #1 SMP Mon Oct 16 14:37:32 EDT 2006
i686 athlon i386 GNU/Linux
[root@donkey ~]# hdparm -t /dev/hda

/dev/hda:
 Timing buffered disk reads:   66 MB in  3.06 seconds =  21.56 MB/sec
[root@donkey ~]# hdparm -t /dev/sda

/dev/sda:
 Timing buffered disk reads:   68 MB in  3.07 seconds =  22.18 MB/sec

<okay, replace ehci_hcd with uhci_hcd...>

[root@donkey ~]# modprobe -r ehci_hcd; modprobe uhci_hcd

<and yes, we have full speed back from the IDE disk...>

[root@donkey ~]# hdparm -t /dev/hda

/dev/hda:
 Timing buffered disk reads:  120 MB in  3.04 seconds =  39.51 MB/sec

<...but the USB2.0 external disk gets a 20x hit performance-wise>

[root@donkey ~]# hdparm -t /dev/sda

/dev/sda:
 Timing buffered disk reads:    4 MB in  4.18 seconds = 980.44 kB/sec


Machine is a K7-800 with 512MB RAM, two internal 160GB
 IDE disks, an external 250GB USB2.0 disk, IDE DVD burner.

>From my dmesg under the FC6 kernel:

VP_IDE: chipset revision 16
VP_IDE: not 100% native mode: will probe irqs later
VP_IDE: VIA vt82c686a (rev 22) IDE UDMA66 controller on pci0000:00:07.1
    ide0: BM-DMA at 0xffa0-0xffa7, BIOS settings: hda:DMA, hdb:DMA
    ide1: BM-DMA at 0xffa8-0xffaf, BIOS settings: hdc:DMA, hdd:pio
Probing IDE interface ide0...
hda: SAMSUNG SP1604N, ATA DISK drive
hdb: Maxtor 6Y160P0, ATA DISK drive
ide0 at 0x1f0-0x1f7,0x3f6 on irq 14
Probing IDE interface ide1...
hdc: TSSTcorpCD/DVDW TS-H552B, ATAPI CD/DVD-ROM drive
ide1 at 0x170-0x177,0x376 on irq 15
hda: max request size: 512KiB
hda: 312581808 sectors (160041 MB) w/2048KiB Cache, CHS=19457/255/63, UDMA(66)
hda: cache flushes supported
 hda: hda1 hda2 hda3 hda4 < hda5 hda6 >
hdb: max request size: 512KiB
hdb: 320173056 sectors (163928 MB) w/7936KiB Cache, CHS=19929/255/63, UDMA(66)
hdb: cache flushes supported

...and...

usb 5-4: new high speed USB device using ehci_hcd and address 2
usb 5-4: configuration #1 chosen from 1 choice
SCSI subsystem initialized
Initializing USB Mass Storage driver...
scsi0 : SCSI emulation for USB Mass Storage devices
usbcore: registered new driver usb-storage
USB Mass Storage support registered.
usb-storage: device found at 2
usb-storage: waiting for device to settle before scanning
  Vendor: ST325082  Model: 3A                Rev:  0 0
  Type:   Direct-Access                      ANSI SCSI revision: 00
usb-storage: device scan complete


Any hints/tips about what to try with this issue will be
 of course very welcome.

Thanks in advance, ciao,

--alessandro

"...when I get it, I _get_ it"

     (Lara Eidemiller)

Re: loading EHCI_HCD slows down IDE disk performance by 50%

[Lee Revell]

On Sun, 2006-10-29 at 19:13 +0100, Alessandro Suardi wrote:
> Any hints/tips about what to try with this issue will be
>  of course very welcome. 

git bisect

Re: loading EHCI_HCD slows down IDE disk performance by 50%

[Alessandro Suardi]

On 10/29/06, Lee Revell <rlrevell@joe-job.com> wrote:
> On Sun, 2006-10-29 at 19:13 +0100, Alessandro Suardi wrote:
> > Any hints/tips about what to try with this issue will be
> >  of course very welcome.
>
> git bisect

As I clarified to Lee in private email, this method doesn't
 seem to apply in this situation, as the problem appeared
 when I selected CONFIG_EHCI_HCD in my build due to the
 purchase of the USB2.0 disk, back in the 2.6.16-rc cycle;
 so if the combination of high performance IDE + EHCI_HCD
 ever existed, it must be earlier than 2.6.16-rc5-git8.

Lacking any input I'll try to find out more in the next week
 by starting out at 2.6.10 and eventually earlier.

Thanks, ciao,

--alessandro

"...when I get it, I _get_ it"

     (Lara Eidemiller)

Re: loading EHCI_HCD slows down IDE disk performance by 50%

[Lee Revell]

On Sun, 2006-10-29 at 20:52 +0100, Alessandro Suardi wrote:
> On 10/29/06, Lee Revell <rlrevell@joe-job.com> wrote:
> > On Sun, 2006-10-29 at 19:13 +0100, Alessandro Suardi wrote:
> > > Any hints/tips about what to try with this issue will be
> > >  of course very welcome.
> >
> > git bisect
> 
> As I clarified to Lee in private email, this method doesn't
>  seem to apply in this situation, as the problem appeared
>  when I selected CONFIG_EHCI_HCD in my build due to the
>  purchase of the USB2.0 disk, back in the 2.6.16-rc cycle;
>  so if the combination of high performance IDE + EHCI_HCD
>  ever existed, it must be earlier than 2.6.16-rc5-git8.
> 
> Lacking any input I'll try to find out more in the next week
>  by starting out at 2.6.10 and eventually earlier.

Maybe you could close the Bugzilla ticket you posted, as most of the
comments relate to HAL and polling for media changes which you seem to
be saying have nothing to do with this bug?

Lee

Re: loading EHCI_HCD slows down IDE disk performance by 50%

[Alessandro Suardi]

On 10/30/06, Lee Revell <rlrevell@joe-job.com> wrote:
> On Sun, 2006-10-29 at 20:52 +0100, Alessandro Suardi wrote:
> > On 10/29/06, Lee Revell <rlrevell@joe-job.com> wrote:
> > > On Sun, 2006-10-29 at 19:13 +0100, Alessandro Suardi wrote:
> > > > Any hints/tips about what to try with this issue will be
> > > >  of course very welcome.
> > >
> > > git bisect
> >
> > As I clarified to Lee in private email, this method doesn't
> >  seem to apply in this situation, as the problem appeared
> >  when I selected CONFIG_EHCI_HCD in my build due to the
> >  purchase of the USB2.0 disk, back in the 2.6.16-rc cycle;
> >  so if the combination of high performance IDE + EHCI_HCD
> >  ever existed, it must be earlier than 2.6.16-rc5-git8.
> >
> > Lacking any input I'll try to find out more in the next week
> >  by starting out at 2.6.10 and eventually earlier.
>
> Maybe you could close the Bugzilla ticket you posted, as most of the
> comments relate to HAL and polling for media changes which you seem to
> be saying have nothing to do with this bug?

The bugzilla entry was hijacked by people who were incorrectly
 thinking their problem had to do with my report, I myself have
 always stated HAL/polling had nothing to do with the issue. Not
 my fault, there...

I may of course re-file the bug, but it'd be nice if I were asked
 by someone looking after the Fedora bugzilla; as a matter of
 fact I did ask for opinions there after my last findings.

--alessandro

"...when I get it, I _get_ it"

     (Lara Eidemiller)