Re: Re: The MSI Driver Guide HOWTO

[Bjorn Helgaas <helgaas@kernel.org>]

[I added back linux-pci so the discussion doesn't get broken up]

On Sat, Mar 11, 2023 at 08:48:35AM +0100, gael.seibert@gmx.fr wrote:
> > Envoyé: vendredi 10 mars 2023 À 23:25
> > De: "Bjorn Helgaas" <bjorn.helgaas@gmail.com>
> > À: gael.seibert@gmx.fr
> > Cc: "Bjorn Helgaas" <helgaas@kernel.org>, linux-pci@vger.kernel.org
> > Objet: Re: The MSI Driver Guide HOWTO
> > 
> > On Fri, Mar 10, 2023 at 3:32 PM <gael.seibert@gmx.fr> wrote:
> > > On 10/03/2023 19:41:02, Bjorn Helgaas wrote:
> > > > On Fri, Mar 10, 2023 at 11:23:14AM +0100, gael.seibert@gmx.fr wrote:
> > > > > On 09/03/2023 23:55:03, Bjorn Helgaas wrote:
> > > > > > On Thu, Mar 09, 2023 at 10:57:51AM +0100, rec wrote:
> > > > > > > On 09/03/2023 00:03:04, Bjorn Helgaas wrote:
> > > > > > > > On Tue, Mar 07, 2023 at 12:22:44PM +0100, rec wrote:
> > > > > > > > > Like asked in :
> > > > https://www.kernel.org/doc/html/latest/PCI/msi-howto.html#disabling-msis-globally
> > > > > > >
> > > > > > > > Thanks for the report!  I assume this means your system has
> > > > problems
> > > > > > > > with MSIs, and booting with "pci=nomsi" makes it work better?
> > > > > > >
> > > > > > > You are welcome,
> > > > > > > The system doesn't boot completely without the "pci=nomsi"
> > > > option.
> > > > > >
> > > > > > What exactly do you mean by "it doesn't boot completely"?  I
> > > > compared
> > > > > > the two dmesg logs, and I see that the "with MSI" log also has the
> > > > > > "single" parameter, so it will only boot to single-user mode.
> > > > >
> > > > > It does it mean than either the boot stop or the system halt,
> > > > power-off
> > > > > before it can be possible to connect tty console or display manager.
> > > >
> > > > Wow.  I'm not sure what would cause a sudden halt or power-off like
> > > > that.  Is there any indication on the console when this happens?  Can
> > > > you try adding the following to your kernel boot parameters to see if
> > > > you can catch anything via a photo or video (you may have to adjust
> > > > the boot_delay to make things readable):
> > > >
> > > >   nosmp ignore_loglevel lpj=lpj=7000000 boot_delay=100
> > >
> > > It will be possible that is a fan problem with a cpu temperature.
> > > (Probably)
> > > I attach a video to the boot.
> > 
> > Thanks for this.  I should have asked at the very beginning whether
> > there are any older kernels that work correctly without "pci=nomsi".
> > If there is such an older kernel, we can try to figure out what change
> > broke it.  Otherwise, I'm running out of ideas.
> 
> I could remove this olde kernel 4. I dont know why i'm keeping this.
> The matter, i was not be able access to a Debian display manager
> with the kernel 5. Since kernel 6, from bookworm repertory, GNOME
> get fine.

I'm not sure what you mean.  I guess you mean that v4.x Linux kernels
didn't need "pci=nomsi"?

> > > > I'm curious about the Ricoh thing because I don't see an obvious MSI
> > > > connection.  Can you collect the output of "sudo lspci -vv"?  The
> > > > lspci output in your initial email wasn't collected as root, so it
> > > > doesn't include information about Capabilities (including MSI).
> > >
> > > Output of #lspci --vv attached
> > 
> > Thanks!  I was hoping something from lspci would connect with
> > ricoh_mmc_fixup_rl5c476(), where we get the "proprietary Ricoh MMC
> > controller" message, e.g., if that function looked at the MSI
> > Capability or something.  But 00:0d has four functions and none of
> > them has an MSI Capability.  And 00:0d.0 has nothing we know about at
> > the offsets the function uses:
> > 
> >   00:0d.0 FireWire (IEEE 1394): Ricoh Co Ltd R5C832
> >     Capabilities: [dc] Power Management version 2
> 
> Well, i Guess this 00:0d.0 FireWire is a link from bios CMOS memory
> adresse. Nothing Can be do without more informations from Ricoh Co
> Ltd R5C832 ?

I asked about this because when you boot with "pci=nomsi", your dmesg
log contains this:

  pci 0000:00:0d.0: proprietary Ricoh MMC controller disabled (via FireWire function)

This comes from ricoh_mmc_fixup_rl5c476(), and you *only* see it when
booting with "pci=nomsi".  So I was trying to figure out why
"pci=nomsi" would affect that function.  But I don't see anything that
relates them.

I'm trying to figure out what happens when you don't use "pci=nomsi".
Earlier you said the system doesn't boot completely.  But you *did*
manage to collect the journalctl log.  Do you mean you are able to log
in on the text console, but the graphical display manager doesn't
start?

Bjorn