Re: [PATCH] ppc: fix boot with sam460ex

["Michael S. Tsirkin" <mst@redhat.com>]

On Fri, May 27, 2022 at 12:46:57PM +0200, BALATON Zoltan wrote:
> Hello,
> 
> Some changes to commit message (patch is OK).

Want to write the commit message for me then?


> On Thu, 26 May 2022, Michael S. Tsirkin wrote:
> > Recent changes to pcie_host corrected size of its internal region to
> > match what it expects - only the low 28 bits are ever decoded. Previous
> > code just ignored bit 29 (if size was 1 << 29) in the address which does
> > not make much sense.  We are now asserting on size > 1 << 28 instead,
> > but it so happened that ppc actually allows guest to configure as large
> > a size as it wants to, and current firmware set it to 1 << 29.
> > 
> > With just qemu-system-ppc -M sam460ex this triggers an assert which
> > seems to happen when the guest (board firmware?) writes a value to
> > CFGMSK reg:
> > 
> > (gdb) bt
> 
> Backtrace is missing but you could just drop this line and replace : with .
> at end of previous line as we probably don't need the full backtrace, the
> commit message is already too long in my opinion.
> 
> > This is done in the board firmware here:
> > 
> > https://git.qemu.org/?p=u-boot-sam460ex.git;a=blob;f=arch/powerpc/cpu/ppc4xx/4xx_pcie.c;h=13348be93dccc74c13ea043d6635a7f8ece4b5f0;hb=HEAD
> > 
> > when trying to map config space.
> > 
> > Note that what firmware does matches
> > https://www.hardware.com.br/comunidade/switch-cisco/1128380/
> 
> That's not it. It's a different hardware and firmware, just quoted it as an
> example that this value seems to be common to that SoC even on different
> hardware/OS/firmware (probably comes from reference hardware/devel board?).
> The sam460ex is here
> 
> https://www.acube-systems.biz/index.php?page=hardware&pid=5
> 
> the U-Boot in above repo is matching the firmware from the acube page but I
> had to fix some bugs in it to make it compile and work.
> 
> Otherwise this should be OK.
> 
> Regards,
> BALATON Zoltan
> 
> > So it's not clear what the proper fix should be.
> > 
> > However, allowing guest to trigger an assert in qemu is not good practice anyway.
> > 
> > For now let's just force the mask to 256MB on guest write, this way
> > anything outside the expected address range is ignored.
> > 
> > Fixes: commit 1f1a7b2269 ("include/hw/pci/pcie_host: Correct PCIE_MMCFG_SIZE_MAX")
> > Reviewed-by: BALATON Zoltan <balaton@eik.bme.hu>
> > Tested-by: BALATON Zoltan <balaton@eik.bme.hu>
> > Signed-off-by: Michael S. Tsirkin <mst@redhat.com>
> > ---
> > 
> > Affected system is orphan so I guess I will merge the patch unless
> > someone objects.
> > 
> > hw/ppc/ppc440_uc.c | 8 ++++++++
> > 1 file changed, 8 insertions(+)
> > 
> > diff --git a/hw/ppc/ppc440_uc.c b/hw/ppc/ppc440_uc.c
> > index 993e3ba955..a1ecf6dd1c 100644
> > --- a/hw/ppc/ppc440_uc.c
> > +++ b/hw/ppc/ppc440_uc.c
> > @@ -1180,6 +1180,14 @@ static void dcr_write_pcie(void *opaque, int dcrn, uint32_t val)
> >     case PEGPL_CFGMSK:
> >         s->cfg_mask = val;
> >         size = ~(val & 0xfffffffe) + 1;
> > +        /*
> > +         * Firmware sets this register to E0000001. Why we are not sure,
> > +         * but the current guess is anything above PCIE_MMCFG_SIZE_MAX is
> > +         * ignored.
> > +         */
> > +        if (size > PCIE_MMCFG_SIZE_MAX) {
> > +            size = PCIE_MMCFG_SIZE_MAX;
> > +        }
> >         pcie_host_mmcfg_update(PCIE_HOST_BRIDGE(s), val & 1, s->cfg_base, size);
> >         break;
> >     case PEGPL_MSGBAH:
> >