Re: [Xen-devel] [PATCH v3 4/7] vpci: fix updating the command register

["Jan Beulich" <JBeulich@xxxxxxxx>]

>>> On 30.10.18 at 16:41, <roger.pau@xxxxxxxxxx> wrote:
> When switching the memory decoding bit in the command register the
> rest of the changes where dropped, leading to only the memory decoding
> bit being updated.
> 
> Fix this by unconditionally writing the guest-requested command except
> for the memory decoding bit, which will be updated once the p2m
> changes are performed.

Are you convinced there are no devices (or drivers) with errata
requiring the write to happen in a single step? Remember that
the register value immediately becomes visible to other (v)CPUs.

> --- a/xen/drivers/vpci/header.c
> +++ b/xen/drivers/vpci/header.c
> @@ -333,8 +333,10 @@ static void cmd_write(const struct pci_dev *pdev, 
> unsigned int reg,
>           * hoping the guest will realize and try again.
>           */
>          modify_bars(pdev, cmd & PCI_COMMAND_MEMORY, false);
> -    else
> -        pci_conf_write16(pdev->seg, pdev->bus, slot, func, reg, cmd);
> +
> +    /* Write the new command without updating the memory decoding bit. */
> +    cmd = (cmd & ~PCI_COMMAND_MEMORY) | (current_cmd & PCI_COMMAND_MEMORY);
> +    pci_conf_write16(pdev->seg, pdev->bus, slot, func, reg, cmd);

Furthermore, if the mapping didn't get deferred, aren't you
discarding the new decode bit value here, as written by
modify_bars() -> apply_map() -> modify_decoding()?

Jan



_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/mailman/listinfo/xen-devel