PCI daughter card with USB on a 405GP

[llandre]

>On Tue, Feb 24, 2004 at 10:38:41PM +0800, song sam wrote:
> >
> > Hi,
> >
> > Wolfgang Denk wrote:
> > > We tested a Adaptec AUA-200LP USB 2.0 card in some PPC sytems. It
> > > worked fine when the devices were already connected at power-on or
> > > at least when booting Linux, while hot-plug would reliably crash the
> > > kernel. As far as I understand such behaviour is typical of some of
> > > the alignment/cache bugs in the Linux kernel's USB code.
> >
> > I also met this USB problem on LITE_DW board when
> > using USB keyboard and mouse.Has it solved or not?
>
>I use the following hack locally to work around some of the 2.4 USB
>stack DMA problems.
>
>-Matt
>
>===== arch/ppc/mm/cachemap.c 1.12 vs edited =====
>--- 1.12/arch/ppc/mm/cachemap.c Tue Dec 10 18:09:59 2002
>+++ edited/arch/ppc/mm/cachemap.c       Tue Dec 12 09:43:06 2002
>@@ -152,7 +152,26 @@
>         case PCI_DMA_NONE:
>                 BUG();
>         case PCI_DMA_FROMDEVICE:        /* invalidate only */
>+#if (defined(CONFIG_4xx) && (defined(CONFIG_USB)) ||
>defined(CONFIG_USB_MODULE))
>+               /*
>+                * USB code does DMA off the stack, so when a driver
>+                * uses the stock pci DMA routines, they end up calling
>+                * invalidate_dcache_range().
>+                * The cache invalidate code in the 405 port invalidates
>+                * an entire cache line, regardless of what addresses are
>+                * passed in - it 'rounds out' to the surrounding cache line.
>+                * While other ports will notice if part of the cache line
>+                * is NOT part of the range being invalidated and 'do
>+                * the right thing', the 405 code will invalidate the
>+                * entire line, potentially trashing the stack, causing a
>+                * panic.
>+                * Until all the USB drivers are fixed, we flush the data
>+                * buffer, a much safer operation.
>+                */
>+               flush_dcache_range(start, end);
>+#else
>                 invalidate_dcache_range(start, end);
>+#endif
>                 break;
>         case PCI_DMA_TODEVICE:          /* writeback only */
>                 clean_dcache_range(start, end);


Hi Matt,

we experienced similar problems and, after applying your patch, everything
works fine (even hot-plug)!
Well done!

Our configuration is:
- PPChameleonEVB board (405EP-based)
- Linux 2.4.20 from Denx's kernel tree
- CiUSB USB controller by Conceptronic, based on 82C861 (it is
OHCI-compatible and not UHCI as documentation claims)

We performed some tests with the following devices:
- Logitech 4-port hub
- Logitech Scroll USB mouse
- Manhattan USB keyboard


For Wolfgang: are you going to include this patch in your kernel tree?


Thanks again for your patch and best regards,

llandre


** Sent via the linuxppc-embedded mail list. See http://lists.linuxppc.org/