Re: [Qemu-devel] [PATCH] usb-ohci: Add address masking.

[takasi-y]

Ah, sorry, this question might be because of my bad choice of word....
# Which word was better, may be "actually" ?

> Why are you writing "usb-ohci is *really* broken now." ??
> Is it supposed to be working or not ?
It doesn't work when the memory region is not aligned to page. Otherwise works.

What I wanted to say by "really broken" was not that it was "broken beyond 
fix", 
but was
- Logically, the mmio change has a possibility to make usb-ohci not work.
- You can produce this condition on current code.
 (this part became the word "really")

r2d was the case(, though it was off-tree at that time).
hw/ppc* s are the same, because they initialize NICs first then usb-ohci.
 qemu -nic model,rtl8139 -usb ...
can produce the case.

I wrote it "potentially breaks" because
- r2d + linux was obviously in trouble, but it was off-tree code at that time.
- I didn't have evaluation environment ready for ppc. I checked them with
  only qemu-ppc without software on it.
- Whether it results in a problem or not depends on the software run on it,
  because PCI configuration is software task.

> Is it a good thing to completely break USB ohci in the development trunk?
No. This issue can be fixed by my patch (if masking is acceptable).
/yoshii