Re: [Qemu-devel] [PATCH] qemu-ga: make get-fsinfo work over pci bridges

[Laszlo Ersek]

Hi Marc-André,

On 04/20/18 17:25, Marc-André Lureau wrote:
> Iterate over the PCI bridges to lookup the PCI device associated with
> the block device.
> 
> Fixes:
> https://bugzilla.redhat.com/show_bug.cgi?id=1567041
> 
> Signed-off-by: Marc-André Lureau <address@hidden>
> ---
>  qga/commands-posix.c | 21 ++++++++++++++++++---
>  1 file changed, 18 insertions(+), 3 deletions(-)
> 
> diff --git a/qga/commands-posix.c b/qga/commands-posix.c
> index 0dc219dbcf..5d4e5f627f 100644
> --- a/qga/commands-posix.c
> +++ b/qga/commands-posix.c
> @@ -879,9 +879,24 @@ static void build_guest_fsinfo_for_real_device(char 
> const *syspath,
>          return;
>      }
>  
> -    driver = get_pci_driver(syspath, (p + 12 + pcilen) - syspath, errp);
> -    if (!driver) {
> -        goto cleanup;
> +    p += 12 + pcilen;
> +    while (true) {
> +        driver = get_pci_driver(syspath, p - syspath, errp);
> +        if (!driver) {
> +            goto cleanup;
> +        }
> +
> +        if (g_str_equal(driver, "pcieport")) {
> +            if (sscanf(p, "/%x:%x:%x.%x%n",
> +                       pci, pci + 1, pci + 2, pci + 3, &pcilen) < 4) {
> +                g_debug("only pci device is supported: sysfs path \"%s\"",
> +                        syspath);
> +                return;
> +            }
> +            p += pcilen;
> +            continue;
> +        }
> +        break;
>      }
>  
>      p = strstr(syspath, "/target");
> 

this patch looked like suitable chewing gum for my brain before I turn
off my work laptop (Saturday 01:36AM, me at work, what else is new), so
here goes:

- Please consider adding a few example pathnames to the commit message
or even the code (as comments). Up to you.

- More importantly, I think this will only work if you use PCIe root
ports and maybe PCIe switches. The original environment where at least I
reproduced it was the "old-style" Q35 libvirt hierarchy: root complex ->
DMI-PCI bridge -> PCI-PCI bridge -> virtio-scsi controller. And, in that
case, the DMI-PCI bridge had no associated driver with it at all in the
guest -- meaning that get_pci_driver() in your patch would return NULL
just the same.

So, I think you should continue scanning as long as:
- driver is either missing or not on the list of desired drivers, AND
- the next pathname component looks PCI-like, and can be parsed.

If the first sub-condition fails, then you found a driver that we can
report about. If the second sub-condition fails, then we give up.

My understanding anyway.

Thanks
Laszlo