Re: [PATCH 5/6] hw/pci-host/sabre: Report IOMMU address range as unimplemented

[Mark Cave-Ayland]

On 12/10/2020 18:09, Philippe Mathieu-Daudé wrote:

> Signed-off-by: Philippe Mathieu-Daudé <f4bug@amsat.org>
> ---
>   hw/pci-host/sabre.c | 2 ++
>   1 file changed, 2 insertions(+)
>
> diff --git a/hw/pci-host/sabre.c b/hw/pci-host/sabre.c
> index 4412e23131c..67699ac9058 100644
> --- a/hw/pci-host/sabre.c
> +++ b/hw/pci-host/sabre.c
> @@ -122,6 +122,7 @@ static void sabre_config_write(void *opaque, hwaddr addr,
>   
>       switch (addr) {
>       case  0x30 ...  0x4f: /* DMA error registers */
> +    case 0x200 ... 0x21f: /* IOMMU registers */
>           qemu_log_mask(LOG_UNIMP,
>                         "%s: Register 0x%02" HWADDR_PRIX " not implemented\n",
>                         __func__, addr);
> @@ -201,6 +202,7 @@ static uint64_t sabre_config_read(void *opaque,
>   
>       switch (addr) {
>       case  0x30 ...  0x4f: /* DMA error registers */
> +    case 0x200 ... 0x21f: /* IOMMU registers */
>           qemu_log_mask(LOG_UNIMP,
>                         "%s: Register 0x%02" HWADDR_PRIX " not implemented\n",
>                         __func__, addr);

In theory this should never happen since a reference to the IOMMU should always be 
set using an object property link (i.e. it is a developer error rather than an 
unimplemented error) and its memory region overlaps this space within the PCI host 
bridge.

Rather than add these logging statemants and/or failing if the property is not set, I 
think now it may be possible to simply embed the IOMMU device within sabre itself 
using the updated QOM APIs. I can take a look to see if this approach will work later 
in the week.


ATB,

Mark.