RE: [rtw89] PCI Passthrough Issue: rtl8852be on ARM VM (Ubuntu host/OpenWrt guest)

[Ping-Ke Shih <pkshih@realtek.com>]

lonthn <lonthn@163.com> wrote:
> At 2025-06-12 15:38:09, lonthn@163.com wrote:
> >Thanks for your quick response.
> >
> >> This looks like interoperability problem of 36-bit DMA.
> >> If you have below commit in your guest OS:
> >> 1fd4b3fe52ef ("wifi: rtw89: pci: support 36-bit PCI DMA address")
> >>
> >> Please also have below commit to rollback 32-bit DMA.
> >> aa70ff0945fe ("wifi: rtw89: pci: early chips only enable 36-bit DMA on specific PCI hosts")
> >>
> >> With the latest kernel, you can just comment code as below to use 32-bit DMA.
> >
> >I tried it and found that it still doesn't work

The error means that firmware isn't ready, and download firmware is the
first part driver asks WiFi device does DMA. Before the DMA, driver has
done many memory mapping IO, so IO is okay, but DMA isn't. 

I think you can try to force 36-bit DMA to see if it can work. (I suppose
the experiments before is 32-bit DMA.)

> >
> >> I have not ever tried passthrough, so I'm not sure if it can work.
> >> Can I know how different the passthrough is?
> >
> >You can refer to this: https://docs.kernel.org/driver-api/vfio.html

I don't quite understand what I should modify the driver. Do you have
hints for me?

I feel PCI passthrough should be transparent to driver/device. I mean
driver/device shouldn't need to know that it operates in host OS or
guest OS (passthrough). 

> >
> >> Does rtl8852be work on host OS?
> >
> >I'm trying, but when I load the 8852be driver, strangely, I get nothing.
> >$> dmesg |grep 8852be
> >no message!
> >$> lsmod |grep rtw
> >rtw_8852be             16384  0
> >rtw_8852b             352256  1 rtw_8852be
> >rtw89pci               61440  1 rtw_8852be
> >rtw89core             569344  2 rtw89pci,rtw_8852b
> >It looks like my device is not detected
> 
> I've confirmed that the host system is working

What were you do to fix that? Did you turn off iommu/VT-d or something?