[U-Boot] The purpose of enabling CONFIG_PCI_BOOTDELAY

[U-Boot] The purpose of enabling CONFIG_PCI_BOOTDELAY

[Mike.Johnson at apc.com]

All,

What is the purpose of CONFIG_PCI_BOOTDELAY? I am using uboot version
1.1.4. I know it's old, but in \drivers\pci.c there is a section of code
right before pci_init.c is called
that can delay the call to pci_init.c.

The code is enabled by defining CONFIG_PCI_BOOTDELAY. But why would it be
necessary to delay initializing the PCI hardware?

Specifically in my case, all the resets on my board have occurred well
before (500 msec) this portion of the code would execute, so it would seem
safe to say that any peripherals like PCI controllers would be satisfied
reset-wise.

In my case though, I need to enable CONFIG_PCI_BOOTDELAY to eliminate a WDT
problem with a PCI access.

If I do not use CONFIG_PCI_BOOTDELAY, my hardware can exhibit a PCI hang,
leading to the above mentioned WDT.

Any information or background that this group can provide would be
appreciated.

Thank-you,

Mike

Michael Johnson
American Power Conversion
132 Fairgrounds Road
West Kingston, RI  02892
tele: (401) 789 - 5735 x2982

[U-Boot] The purpose of enabling CONFIG_PCI_BOOTDELAY

[Wolfgang Denk]

Dear Mike,

In message <OF532F17EB.D2119CC3-ON852576F0.0072502F-852576F0.00725A73@apcc.com> you wrote:
> 
> What is the purpose of CONFIG_PCI_BOOTDELAY? I am using uboot version

Adding a delay so the PCI controllers can come up and stabilize.

> Specifically in my case, all the resets on my board have occurred well
> before (500 msec) this portion of the code would execute, so it would seem
> safe to say that any peripherals like PCI controllers would be satisfied
> reset-wise.

500ms is not safe. We have seen cases (for example on 440SP / 440SPe)
where 5...10 seconds (!) were needed.


Best regards,

Wolfgang Denk

-- 
DENX Software Engineering GmbH,     MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich, Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-10 Fax: (+49)-8142-66989-80 Email: wd at denx.de
Microsoft Compatibility:
     your old Windows 3.11 application crash exactly as the new ones.

[U-Boot] The purpose of enabling CONFIG_PCI_BOOTDELAY

[Andrew Dyer]

On Wed, Mar 24, 2010 at 4:10 PM, Wolfgang Denk <wd@denx.de> wrote:
> Dear Mike,
>
> In message <OF532F17EB.D2119CC3-ON852576F0.0072502F-852576F0.00725A73@apcc.com> you wrote:
>>
>> What is the purpose of CONFIG_PCI_BOOTDELAY? I am using uboot version
>
> Adding a delay so the PCI controllers can come up and stabilize.
>
>> Specifically in my case, all the resets on my board have occurred well
>> before (500 msec) this portion of the code would execute, so it would seem
>> safe to say that any peripherals like PCI controllers would be satisfied
>> reset-wise.
>
> 500ms is not safe. We have seen cases (for example on 440SP / 440SPe)
> where 5...10 seconds (!) were needed.

I've seen this as well, so I agree that 500ms seems reasonable for a
general value.  Per the PCI 2.3 spec, the time is supposed to be min.
100mS.

this pdf is where they added it to PCI 2.2 as an ECN:
http://www.pcisig.com/specifications/conventional/conventional_pci/reset_ecn1_011300.pdf

[U-Boot] The purpose of enabling CONFIG_PCI_BOOTDELAY

[Stefan Roese]

On Wednesday 24 March 2010 22:34:59 Andrew Dyer wrote:
> >> What is the purpose of CONFIG_PCI_BOOTDELAY? I am using uboot version
> > 
> > Adding a delay so the PCI controllers can come up and stabilize.
> > 
> >> Specifically in my case, all the resets on my board have occurred well
> >> before (500 msec) this portion of the code would execute, so it would
> >> seem safe to say that any peripherals like PCI controllers would be
> >> satisfied reset-wise.
> > 
> > 500ms is not safe. We have seen cases (for example on 440SP / 440SPe)
> > where 5...10 seconds (!) were needed.
> 
> I've seen this as well, so I agree that 500ms seems reasonable for a
> general value.  Per the PCI 2.3 spec, the time is supposed to be min.
> 100mS.

Per PCI 2.2, the PCI target must respond to config accesses within 2^25 clock 
cycles after #RST is de-asserted (PCI System Architecture, fourth edition, 
page 321). With 33MHz this is approx. 1 second.

Cheers,
Stefan

--
DENX Software Engineering GmbH,      MD: Wolfgang Denk & Detlev Zundel
HRB 165235 Munich,  Office: Kirchenstr.5, D-82194 Groebenzell, Germany
Phone: (+49)-8142-66989-0 Fax: (+49)-8142-66989-80 Email: office at denx.de

[U-Boot] The purpose of enabling CONFIG_PCI_BOOTDELAY

[Matthias Fuchs]

Hi,

we've seen platforms where the delay brings you out of problems.
Of course this is caused by weird PCI hardware in most cases :-)

I must also admit that the implementation of CONFIG_PCI_BOOTDELAY
is not usable in all situations. We have some boards (e.g. PMC440)
that can be system cpu or PCI adapter. It's not a good idea to
add a pci delay in adapter mode because your device will probably not 
get configured. So we implemented our own pcidelay mechanism 
(see board/esd/pmc440/pmc440.c), because we didn't want to
adjust pcidelay when switching from system cpu to adapter mode and 
vice vera.

CONFIG_PCI_BOOTDELAY is of course not intended for fixing WDT issues :-)

It's a good idea to turn on this option when you are running U-Boot
on system CPUs that connect to "unknown PCI hardware", e.g you have some
PCI/PCIe slots on your board. We never needed it for well known 
onboard PCI devices.

Matthias

On Wednesday 24 March 2010 21:49, Mike.Johnson at apc.com wrote:
> 
> All,
> 
> What is the purpose of CONFIG_PCI_BOOTDELAY? I am using uboot version
> 1.1.4. I know it's old, but in \drivers\pci.c there is a section of code
> right before pci_init.c is called
> that can delay the call to pci_init.c.
> 
> The code is enabled by defining CONFIG_PCI_BOOTDELAY. But why would it be
> necessary to delay initializing the PCI hardware?
> 
> Specifically in my case, all the resets on my board have occurred well
> before (500 msec) this portion of the code would execute, so it would seem
> safe to say that any peripherals like PCI controllers would be satisfied
> reset-wise.
> 
> In my case though, I need to enable CONFIG_PCI_BOOTDELAY to eliminate a WDT
> problem with a PCI access.
> 
> If I do not use CONFIG_PCI_BOOTDELAY, my hardware can exhibit a PCI hang,
> leading to the above mentioned WDT.
> 
> Any information or background that this group can provide would be
> appreciated.
> 
> Thank-you,
>