[PASEMI] Nemo board doesn't recognize any ATA disks with the pci-v5.16 updates

[Marc Zyngier]

HI all,

On Wed, 10 Nov 2021 18:41:06 +0000,
Bjorn Helgaas <helgaas at kernel.org> wrote:
> 
> On Wed, Nov 10, 2021 at 07:07:24PM +0100, Christian Zigotzky wrote:
> > On 09 November 2021 at 03:45 pm, Christian Zigotzky wrote:
> > > Hello,
> > >
> > > The Nemo board [1] doesn't recognize any ATA disks with the pci-v5.16
> > updates [2].
> > >
> > > Error messages:
> > >
> > > ata4.00: gc timeout cmd 0xec
> > > ata4.00: failed to IDENTIFY (I/O error, error_mask=0x4)
> > > ata1.00: gc timeout cmd 0xec
> > > ata1.00: failed to IDENTIFY (I/O error, error_mask=0x4)
> > > ata3.00: gc timeout cmd 0xec
> > > ata3.00: failed to IDENTIFY (I/O error, error_mask=0x4)
> > >
> > > I was able to revert the new pci-v5.16 updates [2]. After a new
> > compiling, the kernel recognize all ATA disks correctly.
> > >
> > > Could you please check the pci-v5.16 updates [2]?
> > >
> > > Please find attached the kernel config.
> > >
> > > Thanks,
> > > Christian
> > >
> > > [1] https://en.wikipedia.org/wiki/AmigaOne_X1000
> > > [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=0c5c62ddf88c34bc83b66e4ac9beb2bb0e1887d4
> > 
> > Hi All,
> > 
> > Many thanks for your nice responses.
> > 
> > I bisected today [1]. 0412841812265734c306ba5ef8088bcb64d5d3bd (of/irq:
> > Allow matching of an interrupt-map local to an interrupt controller) [2] is
> > the first bad commit.
> > 
> > I was able to revert the first bad commit [1]. After a new compiling, the
> > kernel detects all ATA disks without any problems.
> > 
> > I created a patch for an easy reverting the bad commit [1]. With this patch
> > we can do further our kernel tests.
> > 
> > Could you please check the first bad commit [2]?
> > 
> > Thanks,
> > Christian
> > 
> > [1] https://forum.hyperion-entertainment.com/viewtopic.php?p=54398#p54398
> > [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=0412841812265734c306ba5ef8088bcb64d5d3bd
> > 
> > [+ Marc Zyngier, Alyssa Rosenzweig, Lorenzo Pieralisi, and Rob Herring
> > because of the first bad commit]
> 
> Thank you very much for the bisection and for also testing the revert!
> 
> It's easy enough to revert 041284181226 ("of/irq: Allow matching of an
> interrupt-map local to an interrupt controller"), and it seems like
> that's what we need to do.  I have it tentatively queued up.
> 
> That commit was part of the new support for the Apple M1 PCIe
> interface, and I don't know what effect a revert will have on that
> support.  Marc, Alyssa?

It is going to badly break the M1 support, as we won't be able to take
interrupts to detect that the PCIe link is up.

Before we apply a full blown revert and decide that this isn't
workable (and revert the whole M1 PCIe series, because they are
otherwise somewhat pointless), I'd like to understand *what* breaks
exactly.

Christian, could you point me to the full DT that this machine uses?
This would help understanding what goes wrong, and cook something for
you to test.

Thanks,

	M.

-- 
Without deviation from the norm, progress is not possible.