htpt366 PCI latency value is really high

htpt366 PCI latency value is really high

[Mike Mattie]

[-- Attachment #1: Type: text/plain, Size: 1832 bytes --]

Hello,

while hunting down some latency problems I found something quite odd.
The latency reported by lspci -v for the HTP203N card is enormous.

00:09.0 RAID bus controller: Triones Technologies, Inc. HPT302/302N
(rev 02) Subsystem: Triones Technologies, Inc. Unknown device 0001
        Flags: bus master, 66MHz, medium devsel, latency 120, IRQ 17
        I/O ports at ec00 [size=8]
        I/O ports at e800 [size=4]
        I/O ports at e400 [size=8]
        I/O ports at e000 [size=4]
        I/O ports at dc00 [size=256]
        Expansion ROM at dffe0000 [disabled by cmd] [size=128K]
        Capabilities: [60] Power Management version 2


I am assuming that the "latency" field here is the PCI latency timer
which means this card is a bus hog. 

From some reading on this issue linux methodically sets a sane value for 
all the PCI cards it sets up, which looks normal on the rest of the system, 
which is set to the value: 32

setting the value 32 with:

setpci -v -s "00:09.0" latency_timer=32

00:09.0 RAID bus controller: Triones Technologies, Inc. HPT302/302N (rev 02)
        Subsystem: Triones Technologies, Inc. Unknown device 0001
        Flags: bus master, 66MHz, medium devsel, latency 48, IRQ 17
        I/O ports at ec00 [size=8]
        I/O ports at e800 [size=4]
        I/O ports at e400 [size=8]
        I/O ports at e000 [size=4]
        I/O ports at dc00 [size=256]
        Expansion ROM at dffe0000 [disabled by cmd] [size=128K]
        Capabilities: [60] Power Management version 2

Results in 48, which is not what I asked, but hopefully this is
linux doing the right thing.

I know this chipset is pretty brain-damaged, but is this
high latency value a work-around for broken hardware, or
just a oversight ?

Cheers,
Mike Mattie - codermattie@gmail.com

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: htpt366 PCI latency value is really high

[Sergei Shtylyov]

Hello.

Mike Mattie wrote:

> while hunting down some latency problems I found something quite odd.
> The latency reported by lspci -v for the HTP203N card is enormous.

> 00:09.0 RAID bus controller: Triones Technologies, Inc. HPT302/302N
> (rev 02) Subsystem: Triones Technologies, Inc. Unknown device 0001
>         Flags: bus master, 66MHz, medium devsel, latency 120, IRQ 17
>         I/O ports at ec00 [size=8]
>         I/O ports at e800 [size=4]
>         I/O ports at e400 [size=8]
>         I/O ports at e000 [size=4]
>         I/O ports at dc00 [size=256]
>         Expansion ROM at dffe0000 [disabled by cmd] [size=128K]
>         Capabilities: [60] Power Management version 2

> I am assuming that the "latency" field here is the PCI latency timer
> which means this card is a bus hog. 

> From some reading on this issue linux methodically sets a sane value for 
> all the PCI cards it sets up, which looks normal on the rest of the system, 
> which is set to the value: 32

   Hm, I'm only seeing clamping to the smallest of 64 and pcibios_max_latency (255) in arch/i386/pci/i386.c if the latency value is too low... Which arch are you using?

> setting the value 32 with:

> setpci -v -s "00:09.0" latency_timer=32

> 00:09.0 RAID bus controller: Triones Technologies, Inc. HPT302/302N (rev 02)
>         Subsystem: Triones Technologies, Inc. Unknown device 0001
>         Flags: bus master, 66MHz, medium devsel, latency 48, IRQ 17
>         I/O ports at ec00 [size=8]
>         I/O ports at e800 [size=4]
>         I/O ports at e400 [size=8]
>         I/O ports at e000 [size=4]
>         I/O ports at dc00 [size=256]
>         Expansion ROM at dffe0000 [disabled by cmd] [size=128K]
>         Capabilities: [60] Power Management version 2

> Results in 48, which is not what I asked, but hopefully this is
> linux doing the right thing.

   Not sure -- seems likely that it's the chip's own enforced minimum instead...

> I know this chipset is pretty brain-damaged, but is this
> high latency value a work-around for broken hardware, or

   More like it.  Although HighPoint's own drivers force 64.

> just a oversight ?

   Not likely since the value is too "special"...

> Cheers,
> Mike Mattie - codermattie@gmail.com

WBR, Sergei

Re: htpt366 PCI latency value is really high

[Mike Mattie]

[-- Attachment #1: Type: text/plain, Size: 3483 bytes --]

On Thu, 26 Apr 2007 18:30:33 +0400
Sergei Shtylyov <sshtylyov@ru.mvista.com> wrote:

> Hello.
> 
> Mike Mattie wrote:
> 
> > while hunting down some latency problems I found something quite
> > odd. The latency reported by lspci -v for the HTP203N card is
> > enormous.
> 
> > 00:09.0 RAID bus controller: Triones Technologies, Inc. HPT302/302N
> > (rev 02) Subsystem: Triones Technologies, Inc. Unknown device 0001
> >         Flags: bus master, 66MHz, medium devsel, latency 120, IRQ 17
> >         I/O ports at ec00 [size=8]
> >         I/O ports at e800 [size=4]
> >         I/O ports at e400 [size=8]
> >         I/O ports at e000 [size=4]
> >         I/O ports at dc00 [size=256]
> >         Expansion ROM at dffe0000 [disabled by cmd] [size=128K]
> >         Capabilities: [60] Power Management version 2
> 
> > I am assuming that the "latency" field here is the PCI latency timer
> > which means this card is a bus hog. 
> 
> > From some reading on this issue linux methodically sets a sane
> > value for all the PCI cards it sets up, which looks normal on the
> > rest of the system, which is set to the value: 32
> 
>    Hm, I'm only seeing clamping to the smallest of 64 and
> pcibios_max_latency (255) in arch/i386/pci/i386.c if the latency
> value is too low... Which arch are you using?
> 
> > setting the value 32 with:
> 
> > setpci -v -s "00:09.0" latency_timer=32
> 
> > 00:09.0 RAID bus controller: Triones Technologies, Inc. HPT302/302N
> > (rev 02) Subsystem: Triones Technologies, Inc. Unknown device 0001
> >         Flags: bus master, 66MHz, medium devsel, latency 48, IRQ 17
> >         I/O ports at ec00 [size=8]
> >         I/O ports at e800 [size=4]
> >         I/O ports at e400 [size=8]
> >         I/O ports at e000 [size=4]
> >         I/O ports at dc00 [size=256]
> >         Expansion ROM at dffe0000 [disabled by cmd] [size=128K]
> >         Capabilities: [60] Power Management version 2
> 
> > Results in 48, which is not what I asked, but hopefully this is
> > linux doing the right thing.

output is decimal, input is hex - self-rtfm.
 
>    Not sure -- seems likely that it's the chip's own enforced minimum
> instead...
> 
> > I know this chipset is pretty brain-damaged, but is this
> > high latency value a work-around for broken hardware, or
> 
>    More like it.  Although HighPoint's own drivers force 64.
> 
> > just a oversight ?
> 
>    Not likely since the value is too "special"...

I have ran with a setting of 40 for a couple of days without any trouble.
There are many reasons a card would default to a higher level, but in
the end it is basically tuning the card for a server application.

A warning that some cards default to high PCI latency values; settings that 
can interfere with latency sensitive devices such as sounds cards - this 
could be a help to others.  

00:09.0 RAID bus controller: Triones Technologies, Inc. HPT302/302N
(rev 02) Subsystem: Triones Technologies, Inc. Unknown device 0001
        Flags: bus master, 66MHz, medium devsel, latency 40, IRQ 19
        I/O ports at ec00 [size=8]
        I/O ports at e800 [size=4]
        I/O ports at e400 [size=8]
        I/O ports at e000 [size=4]
        I/O ports at dc00 [size=256]
        Expansion ROM at dffe0000 [disabled by cmd] [size=128K]
        Capabilities: [60] Power Management version 2


> > Cheers,
> > Mike Mattie - codermattie@gmail.com
> 
> WBR, Sergei

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: htpt366 PCI latency value is really high

[Alan Cox]

O> I have ran with a setting of 40 for a couple of days without any trouble.
> There are many reasons a card would default to a higher level, but in
> the end it is basically tuning the card for a server application.
> 
> A warning that some cards default to high PCI latency values; settings that 
> can interfere with latency sensitive devices such as sounds cards - this 
> could be a help to others.  

A warning that there are quite a few cards that subtly corrupt your data
if you don't have the latency set to values to work around chip errata is
also worth making.

You fiddle with PCI latency values at your peril, and preferably with the
data sheet and errata docs to hand.

Alan

Re: htpt366 PCI latency value is really high

[Sergei Shtylyov]

Hello.

Alan Cox wrote:

>>There are many reasons a card would default to a higher level, but in
>>the end it is basically tuning the card for a server application.

>>A warning that some cards default to high PCI latency values; settings that 
>>can interfere with latency sensitive devices such as sounds cards - this 
>>could be a help to others.  

> A warning that there are quite a few cards that subtly corrupt your data
> if you don't have the latency set to values to work around chip errata is
> also worth making.

> You fiddle with PCI latency values at your peril, and preferably with the
> data sheet and errata docs to hand.

   And I don't have the latter for the HighPoint chips. Not sure if anybody ever had it. :-)

> Alan

MBR, Sergei