Status of Promise drivers?

Status of Promise drivers?

[Daniel Brahneborg]

Hi,

I'm having serious problems with my Silicon Image SATA card,
and am considering replacing with a (more expensive) Promise
SATA TX2 Plus card.

First I have some questions, to avoid getting into the same
mess as I'm in now.

 1. What is the status of the Promise driver?  Stable for
    everybody? (The SI3112 has problems for some people when
    using the IDE driver, and for some with the SCSI driver.)
    A binary module is ok, a binary bzImage definately is not.

 2. Has anybody used the driver on an Abit KV7 motherboard?

 3. Has anybody used it with *2* network cards?  Bonus
    points for the builtin Via Rhine and an 8139 card.
    (The SCSI driver for the SI3112 kills the machine when a
    second card is activated.)

 4. Has anybody used it together with two other disks on the
    VIA SATA interface on the KV7?

 5. Has anybody used it for a Raid5 setup? (With SI3112 Raid1
    works better than Raid5, since Raid5 gives corrupted files
    all over the place and Raid1 only corrupts them once in
    a while.)

 6. Has anybody used it in a non-root setup, having hda on
    the PATA interface?  (The SI3112 causes lots of problems
    with I/O port and IRQ conflicts, requiring hours of
    tweaking of the boot parameters.)

 7. Do you have DMA enabled?  (SI3112 without DMA works fine,
    but with terrible performance.)

 8. Do you get decent performance? (The 5MB/sec I get without
    DMA on the SI3112 card is _not_ decent.)

The VIA SATA drivers work perfectly with great performance,
so obviously Linux is able to talk SATA.  Unfortunately both
the SI drivers still require some work.

I'm more than happy to help debug and fix the SI driver, but
I don't know enough about the kernel to do it myself.

Lots of thanks.

/Basic

Re: Status of Promise drivers?

[Jeff Garzik]

Daniel Brahneborg wrote:
> Hi,
> 
> I'm having serious problems with my Silicon Image SATA card,
> and am considering replacing with a (more expensive) Promise
> SATA TX2 Plus card.
> 
> First I have some questions, to avoid getting into the same
> mess as I'm in now.
> 
>  1. What is the status of the Promise driver?  Stable for
>     everybody? (The SI3112 has problems for some people when
>     using the IDE driver, and for some with the SCSI driver.)
>     A binary module is ok, a binary bzImage definately is not.

It's beta, and stable.


>  2. Has anybody used the driver on an Abit KV7 motherboard?
> 
>  3. Has anybody used it with *2* network cards?  Bonus
>     points for the builtin Via Rhine and an 8139 card.
>     (The SCSI driver for the SI3112 kills the machine when a
>     second card is activated.)
> 
>  4. Has anybody used it together with two other disks on the
>     VIA SATA interface on the KV7?
> 
>  5. Has anybody used it for a Raid5 setup? (With SI3112 Raid1
>     works better than Raid5, since Raid5 gives corrupted files
>     all over the place and Raid1 only corrupts them once in
>     a while.)

Any comparison with Silicon Image is false.  The Silicon Image driver 
has been repeatedly stated (by me, the author) to be BROKEN.  It is 
thusly marked with CONFIG_BROKEN.

Those who ignore the author stating the Silicon Image driver is _alpha_ 
and unfinished do so at their own risk.


>  6. Has anybody used it in a non-root setup, having hda on
>     the PATA interface?  (The SI3112 causes lots of problems
>     with I/O port and IRQ conflicts, requiring hours of
>     tweaking of the boot parameters.)

PATA interfaces on Promise SATA are not yet supported.


>  7. Do you have DMA enabled?  (SI3112 without DMA works fine,
>     but with terrible performance.)

DMA cannot be disabled :)


>  8. Do you get decent performance? (The 5MB/sec I get without
>     DMA on the SI3112 card is _not_ decent.)

Performance is far beyond this.

	Jeff

Re: Status of Promise drivers?

[Daniel Brahneborg]

Thanks for your quick reply, Jeff.

On Tue, Dec 30, 2003 at 02:06:41PM -0500, Jeff Garzik wrote:
> Daniel Brahneborg wrote:
> > I'm having serious problems with my Silicon Image SATA card,
> > and am considering replacing with a (more expensive) Promise
> > SATA TX2 Plus card.
> > 
> > First I have some questions, to avoid getting into the same
> > mess as I'm in now.
> > 
> >  1. What is the status of the Promise driver?  Stable for
> >     everybody? (The SI3112 has problems for some people when
> >     using the IDE driver, and for some with the SCSI driver.)
> >     A binary module is ok, a binary bzImage definately is not.
> 
> It's beta, and stable.

Great!

> Any comparison with Silicon Image is false.  The Silicon Image driver 
> has been repeatedly stated (by me, the author) to be BROKEN.  It is 
> thusly marked with CONFIG_BROKEN.
> 
> Those who ignore the author stating the Silicon Image driver is _alpha_ 
> and unfinished do so at their own risk.

Is this true for both the IDE driver and the SCSI driver?

> >  6. Has anybody used it in a non-root setup, having hda on
> >     the PATA interface?  (The SI3112 causes lots of problems
> >     with I/O port and IRQ conflicts, requiring hours of
> >     tweaking of the boot parameters.)
> 
> PATA interfaces on Promise SATA are not yet supported.

Oh, I wasn't even aware that the Promise had any PATA slots. :)
It's the builting PATA interface on the motherboard that I want
to keep.

> >  7. Do you have DMA enabled?  (SI3112 without DMA works fine,
> >     but with terrible performance.)
> 
> DMA cannot be disabled :)

Cool, that's the way it should be.

Is the same driver used for the TX4 card? Checking my logs even the
VIA driver seems to have problems running under Raid. Both Raid1
and Raid5 gives me corrupted files, unless I turn off DMA.

/Basic

Re: Status of Promise drivers?

[Jeff Garzik]

Daniel Brahneborg wrote:
> On Tue, Dec 30, 2003 at 02:06:41PM -0500, Jeff Garzik wrote:
>>Any comparison with Silicon Image is false.  The Silicon Image driver 
>>has been repeatedly stated (by me, the author) to be BROKEN.  It is 
>>thusly marked with CONFIG_BROKEN.
>>
>>Those who ignore the author stating the Silicon Image driver is _alpha_ 
>>and unfinished do so at their own risk.
> 
> 
> Is this true for both the IDE driver and the SCSI driver?

Only the "scsi" driver, sata_sil.  People do have some problems with the 
IDE driver siimage, but it is on the whole much more stable.


>>> 6. Has anybody used it in a non-root setup, having hda on
>>>    the PATA interface?  (The SI3112 causes lots of problems
>>>    with I/O port and IRQ conflicts, requiring hours of
>>>    tweaking of the boot parameters.)
>>
>>PATA interfaces on Promise SATA are not yet supported.
> 
> 
> Oh, I wasn't even aware that the Promise had any PATA slots. :)
> It's the builting PATA interface on the motherboard that I want
> to keep.

That's a different driver, then :)


> Is the same driver used for the TX4 card? Checking my logs even the
> VIA driver seems to have problems running under Raid. Both Raid1
> and Raid5 gives me corrupted files, unless I turn off DMA.

sata_promise is used for all known Promise SATA cards.

Which VIA driver are you referring to?  And PATA or SATA?  You need to 
be more specific :)   The drivers/ide via82cxxx driver seems to work for 
PATA.  The sata_via ("scsi") driver and the generic IDE driver work for 
VIA SATA, but people have reported problems on x86-64.

	Jeff

Re: Status of Promise drivers?

[Daniel Brahneborg]

On Tue, Dec 30, 2003 at 02:57:34PM -0500, Jeff Garzik wrote:
> Daniel Brahneborg wrote:
> > On Tue, Dec 30, 2003 at 02:06:41PM -0500, Jeff Garzik wrote:
> >>Those who ignore the author stating the Silicon Image driver is _alpha_ 
> >>and unfinished do so at their own risk.
> > 
> > Is this true for both the IDE driver and the SCSI driver?
> 
> Only the "scsi" driver, sata_sil.  People do have some problems with the 
> IDE driver siimage, but it is on the whole much more stable.

I'm one of the people with problems...

> > Is the same driver used for the TX4 card? Checking my logs even the
> > VIA driver seems to have problems running under Raid. Both Raid1
> > and Raid5 gives me corrupted files, unless I turn off DMA.
> 
> sata_promise is used for all known Promise SATA cards.

Great!

> Which VIA driver are you referring to?  And PATA or SATA?  You need to 
> be more specific :)   The drivers/ide via82cxxx driver seems to work for 
> PATA.  The sata_via ("scsi") driver and the generic IDE driver work for 
> VIA SATA, but people have reported problems on x86-64.

SATA, with the via82cxxx driver.  It works fine for simple partitions,
which I guess is what most people uses.

In short: First I went 100% scsi (except for / on hda) with sata_sil
and sata_via. This made my computer halt when I activated my second
network card (8139 + via-rhine or another 8139). Then I went 100% ide
with the via82cxxx and siimage drivers, which made the network cards
work.  Disks on the VIA interface worked fine for simple file systems
but failed when used for Raid. Disks on the Silicon Image interface
failed both.  It even got badblocks from the Ext2 suite to report tons
of failures.  Smartmontools has no complaints, so the disks as such
are ok.

/Basic

Re: Status of Promise drivers?

[Hendrik Visage]

On Tue, Dec 30, 2003 at 02:57:34PM -0500, Jeff Garzik wrote:
> 
> Which VIA driver are you referring to?  And PATA or SATA?  You need to 
> be more specific :)   The drivers/ide via82cxxx driver seems to work for 
> PATA.  The sata_via ("scsi") driver and the generic IDE driver work for 
> VIA SATA, but people have reported problems on x86-64.

Hi Jeff, what type of problems have been seen on x86-64 and sata_via?

Haven't had much trouble personaly yet...

Re: Status of Promise drivers?

[Jeff Garzik]

Daniel Brahneborg wrote:
> In short: First I went 100% scsi (except for / on hda) with sata_sil
> and sata_via. This made my computer halt when I activated my second
> network card (8139 + via-rhine or another 8139). Then I went 100% ide
> with the via82cxxx and siimage drivers, which made the network cards
> work.  Disks on the VIA interface worked fine for simple file systems
> but failed when used for Raid. Disks on the Silicon Image interface
> failed both.  It even got badblocks from the Ext2 suite to report tons
> of failures.  Smartmontools has no complaints, so the disks as such
> are ok.

I definitely recommend IDE siimage driver for Silicon Image SATA (though 
I will announce when my driver is fixed).

For VIA SATA, neither IDE nor libata program the chipset 100%, so I'm 
not surprised you're seeing failures.  I'm working with VIA to get more 
details as we speak.  They've been helpful so far.  I would recommend 
sata_via over the generic IDE driver, but that's sorta a toss-up at present.

	Jeff

Re: Status of Promise drivers?

[Jeff Garzik]

Hendrik Visage wrote:
> On Tue, Dec 30, 2003 at 02:57:34PM -0500, Jeff Garzik wrote:
> 
>>Which VIA driver are you referring to?  And PATA or SATA?  You need to 
>>be more specific :)   The drivers/ide via82cxxx driver seems to work for 
>>PATA.  The sata_via ("scsi") driver and the generic IDE driver work for 
>>VIA SATA, but people have reported problems on x86-64.
> 
> 
> Hi Jeff, what type of problems have been seen on x86-64 and sata_via?
> 
> Haven't had much trouble personaly yet...


For some people it flat-out doesn't work at all :)  This might be a 
problem with the x86-64 iommu though...

Typically with sata_via, it either works 100%, or it doesn't work at 
all.  (it works for most people, but as we see with Daniel there are 
still some problems)

	Jeff

Re: Status of Promise drivers?

[Daniel Brahneborg]

On Tue, Dec 30, 2003 at 03:25:15PM -0500, Jeff Garzik wrote:
> Daniel Brahneborg wrote:
> > In short: First I went 100% scsi (except for / on hda) with sata_sil
> > and sata_via. This made my computer halt when I activated my second
> > network card (8139 + via-rhine or another 8139). Then I went 100% ide
> > with the via82cxxx and siimage drivers, which made the network cards
> > work.  Disks on the VIA interface worked fine for simple file systems
> > but failed when used for Raid. Disks on the Silicon Image interface
> > failed both.  It even got badblocks from the Ext2 suite to report tons
> > of failures.  Smartmontools has no complaints, so the disks as such
> > are ok.
> 
> I definitely recommend IDE siimage driver for Silicon Image SATA (though 
> I will announce when my driver is fixed).

Not even the IDE driver handles raid correctly though, and the
badblocks thing was with that IDE driver.  Is there any debug information
I can provide you with, in order to help you fix this?

> For VIA SATA, neither IDE nor libata program the chipset 100%, so I'm 
> not surprised you're seeing failures.  I'm working with VIA to get more 
> details as we speak.  They've been helpful so far.  I would recommend 
> sata_via over the generic IDE driver, but that's sorta a toss-up at present.

Good, I like companies that are helpful. What would you estimate
the time frame for working raid support to?  10 minutes? A week?
A month?  Kernel 2.8?

/Basic

Re: Status of HighPoint / Promise drivers?

[Daniel Brahneborg]

On Tue, Dec 30, 2003 at 08:00:12PM +0100, Daniel Brahneborg wrote:
> I'm having serious problems with my Silicon Image SATA card,
> and am considering replacing with a (more expensive) Promise
> SATA TX2 Plus card.

While I'm investigating: How about the HighPoint drivers, more
specifically the one for HPT374 (the HighPoint 1540 card)?
It's significantly cheaper than the Promise cards.

/Basic

Re: Status of HighPoint / Promise drivers?

[Jeff Garzik]

Daniel Brahneborg wrote:
> On Tue, Dec 30, 2003 at 08:00:12PM +0100, Daniel Brahneborg wrote:
> 
>>I'm having serious problems with my Silicon Image SATA card,
>>and am considering replacing with a (more expensive) Promise
>>SATA TX2 Plus card.
> 
> 
> While I'm investigating: How about the HighPoint drivers, more
> specifically the one for HPT374 (the HighPoint 1540 card)?
> It's significantly cheaper than the Promise cards.


There is a GPL'd driver for HighPoint SATA floating around somewhere... 
  not really well support though, is the basic answer.

	Jeff

Re: Status of Promise drivers?

[Jeff Garzik]

Daniel Brahneborg wrote:
> On Tue, Dec 30, 2003 at 03:25:15PM -0500, Jeff Garzik wrote:
> 
>>Daniel Brahneborg wrote:
>>
>>>In short: First I went 100% scsi (except for / on hda) with sata_sil
>>>and sata_via. This made my computer halt when I activated my second
>>>network card (8139 + via-rhine or another 8139). Then I went 100% ide
>>>with the via82cxxx and siimage drivers, which made the network cards
>>>work.  Disks on the VIA interface worked fine for simple file systems
>>>but failed when used for Raid. Disks on the Silicon Image interface
>>>failed both.  It even got badblocks from the Ext2 suite to report tons
>>>of failures.  Smartmontools has no complaints, so the disks as such
>>>are ok.
>>
>>I definitely recommend IDE siimage driver for Silicon Image SATA (though 
>>I will announce when my driver is fixed).
> 
> 
> Not even the IDE driver handles raid correctly though, and the
> badblocks thing was with that IDE driver.  Is there any debug information
> I can provide you with, in order to help you fix this?

Not really, I mainly need free time :)


>>For VIA SATA, neither IDE nor libata program the chipset 100%, so I'm 
>>not surprised you're seeing failures.  I'm working with VIA to get more 
>>details as we speak.  They've been helpful so far.  I would recommend 
>>sata_via over the generic IDE driver, but that's sorta a toss-up at present.
> 
> 
> Good, I like companies that are helpful. What would you estimate
> the time frame for working raid support to?  10 minutes? A week?
> A month?  Kernel 2.8?

I would say "under a day"

	Jeff

Re: Status of HighPoint / Promise drivers?

[jlewis]

On Tue, 30 Dec 2003, Jeff Garzik wrote:

> > While I'm investigating: How about the HighPoint drivers, more
> > specifically the one for HPT374 (the HighPoint 1540 card)?
> > It's significantly cheaper than the Promise cards.
> 
> There is a GPL'd driver for HighPoint SATA floating around somewhere... 
>   not really well support though, is the basic answer.

Based on my experience with a HPT Rocket100 (not a RAID card), and the
2.4.x kernel driver and the drivers supplied by HPT, I'd be very afraid.  
I bought the Rocket100 thinking it was supported.  It's not.  Red Hat 8.0,
9, and Fedora Core 1.0 installs either don't recognize it or lock up when
probing the drives connected to it.  HPT's driver disks for 8.0 and 9 get
as far as starting to install RPMS, then die with disk i/o errors.  Same
hardware with a Promise UltraTX2-100 swapped for the Rocket100 works just
fine.
 
----------------------------------------------------------------------
 Jon Lewis *jlewis@lewis.org*|  I route
 Senior Network Engineer     |  therefore you are
 Atlantic Net                |  
_________ http://www.lewis.org/~jlewis/pgp for PGP public key_________

Re: Status of Promise drivers?

[Daniel Brahneborg]

On Tue, Dec 30, 2003 at 03:25:15PM -0500, Jeff Garzik wrote:
> For VIA SATA, neither IDE nor libata program the chipset 100%, so I'm 
> not surprised you're seeing failures.  I'm working with VIA to get more 
> details as we speak.  They've been helpful so far.  I would recommend 
> sata_via over the generic IDE driver, but that's sorta a toss-up at present.

I've made some tests with 2.6.1 and 2.6.1-mm2.

Good news: RAID1 works on two SATA disks with the IDE VIA
driver (via82cxxx).  At least after a while (see below).

Bug 1: Doing a "modprobe via82cxxx" fails, since it tries to
aquire ide0 and ide1 even though they are used by other drivers.
Statically linked, it takes ide2 and ide3, just as it should.
My ide0 and ide1 are PATA connectors that I'd like to keep
even though the SATA connectors are activated.  The siimage
driver behaves correctly here, but only with the -mm2 version.

Bug 2: Using the RAID1 array while it is being synced causes
data corruption.  When the sync is complete, everything seems
to work fine.

/Basic

Re: Status of Promise drivers?

[Daniel Brahneborg]

(Continued from an earlier discussion on the linux-raid list.)

I made the mistake of getting a Highpoint Rocket 1540 instead
of a Promise SATA card, and I simply can't get it to work
with the hpt366 driver in the kernel.  The driver from Highpoint
works better, but only works for the RedHat 2.4.20-8 kernel,
so I won't bother anybody with the problems with that one.

A major factor seems to be the KT600 chipset on the Abit KV7
motherboard. Is this a known issue?

On Tue, Dec 30, 2003 at 02:06:41PM -0500, Jeff Garzik wrote:
> Daniel Brahneborg wrote:
> > I'm having serious problems with my Silicon Image SATA card,
> > and am considering replacing with a (more expensive) Promise
> > SATA TX2 Plus card.
> > 
> > First I have some questions, to avoid getting into the same
> > mess as I'm in now.
> > 
> >  1. What is the status of the Promise driver?  Stable for
> >     everybody? (The SI3112 has problems for some people when
> >     using the IDE driver, and for some with the SCSI driver.)
> >     A binary module is ok, a binary bzImage definately is not.
> 
> It's beta, and stable.

Since I'm paranoid: What motherboards has it been tested and
verified with?

> >  3. Has anybody used it with *2* network cards?  Bonus
> >     points for the builtin Via Rhine and an 8139 card.
> >     (The SCSI driver for the SI3112 kills the machine when a
> >     second card is activated.)

Interestingly enough, the Highpoint card has the same effect
when activating the first network card (it doesn't matter which
one), but only with the 2.4 kernel. The 2.6 kernel works fine.
Again, could this be a motherboard issue?

/Basic

Re: Status of Promise drivers?

[Gunnlaugur Thor Briem]

Hi,

Daniel Brahneborg wrote:

>>It's beta, and stable.
> 
> 
> Since I'm paranoid: What motherboards has it been tested and
> verified with?

Blessed are the paranoid: take a look at

http://bugzilla.kernel.org/show_bug.cgi?id=2011

(which hadn't been reported when Jeff Garzik called the driver stable)

Don't know about any other motherboards, but in any case your kernel is 
likely to hang when trying to use a PDC 20376 (on a motherboard, as in 
my case, or on a separate board), for the time being.

	- Gulli

Re: Status of Promise drivers?

[Gunnlaugur Thor Briem]

Gunnlaugur Thor Briem wrote:
 > http://bugzilla.kernel.org/show_bug.cgi?id=2011
[...]
 > Don't know about any other motherboards, but in any case your kernel is
 > likely to hang when trying to use a PDC 20376 (on a motherboard, as in
 > my case, or on a separate board), for the time being.

... and now he corrects himself. I just tried applying Ross Dickson's 
patches for APIC quirks on nForce2 chipsets:

http://lkml.org/lkml/2003/12/21/7

and that seems to have taken care of my problem completely. I have not 
tested for very long, but I got through the bonnie++ test suite; never even 
came close before.

So the Promise controller and driver appear to be innocent of my woes, and I 
do not know any reason to think that your kernel will hang because of them.

Sorry 'bout that. (And thanks Ross!)

     - Gulli