Re: [PATCH v2] piix: fix regression during unplug in Xen HVM domUs

[Olaf Hering]

I need advice on how to debug this.

One thing that stands out is uhci_irq().
It reads a u16 from the USBSTS register. 

On the qemu side, this read is served from bmdma_read. Since the read
size is 2, the result is ~0, and uhci_irq() turns the controller off.
In other words, memory_region_ops_read from addr=0xc102 is served from 
"piix-bmdma"

If the pci_set_word calls in piix_ide_reset are skipped, the read is
served from uhci_port_write. This is the expected behavior.
In other words, memory_region_ops_read from addr=0xc102 is served from "uhci".

So far I was unable to decipher how the pci_set_word calls can
possibly affect the outcome and the owner of memory_region_ops_read.


Thanks,
Olaf

Wed, 10 May 2023 00:58:27 +0200 Olaf Hering <olaf@aepfle.de>:

> Resuming this old thread about an unfixed bug, which was introduced in 
> qemu-4.2:
> 
> qemu ends up in piix_ide_reset from pci_unplug_disks.
> This was not the case prior 4.2, the removed call to
> qemu_register_reset(piix3_reset, d) in
> ee358e919e385fdc79d59d0d47b4a81e349cd5c9 did apparently nothing.
> 
> In my debugging (with v8.0.0) it turned out the three pci_set_word
> causes the domU to hang. In fact, it is just the last one:
> 
>    pci_set_byte(pci_conf + 0x20, 0x01);  /* BMIBA: 20-23h */
> 
> It changes the value from 0xc121 to 0x1.
> 
> The question is: what does this do in practice?
> 
> Starting with recent qemu (like 7.2), the domU sometimes proceeds with
> these messages:
> 
>     [    1.631161] uhci_hcd 0000:00:01.2: host system error, PCI problems?
>     [    1.634965] uhci_hcd 0000:00:01.2: host controller process error, 
> something bad happened!
>     [    1.634965] uhci_hcd 0000:00:01.2: host controller halted, very bad!
>     [    1.634965] uhci_hcd 0000:00:01.2: HC died; cleaning up
>     Loading basic drivers...[    2.398048] Disabling IRQ #23

pgpDssZWsvZvf.pgp
Description: Digitale Signatur von OpenPGP