Hi, > -----Original Message----- > From: Felipe Balbi [mailto:felipe.balbi@xxxxxxxxx] > Sent: Friday, May 14, 2010 4:37 PM > To: Gupta, Ajay Kumar > Cc: me@xxxxxxxxxxxxxxx; linux-usb@xxxxxxxxxxxxxxx; linux- > omap@xxxxxxxxxxxxxxx > Subject: Re: [PATCH 4/5] musb: use system DMA for unaligned buffers on RTL > >= 1.8 > > On Thu, May 13, 2010 at 06:13:00AM +0200, ext Gupta, Ajay Kumar wrote: > >How about this 'memcpy', this would affect both cpu loading > >and performance. That's why using system dma would be a better > >approach. > > but that's not portable. > Even if you come up with nice wrappers for > system dma usage (which is already ugly, more exported functions) you > might end up on a board where system dma has same alignment constraints > as mentor dma. It can't be an OMAP3 based board but yes, future platform could have such limitations but they should not restrict other platforms to use system dma which can handle unaligned buffers. > Either fall back to pio or use a bounce buffer. That'll > work always. Agreed but performance we should allow system dma approach also. > > Besides, the only offending gadget driver we have is g_ether and that's > already 'fixed' by the alignment patch to g_ether. No, it's not fixed. It had two issues, 1. Cache coherency as we observed iperf failing and after using Cache_flush_all() things started working. 2. High cpu loading of 100% and bit rate of only 30Mbits/sec due To memmove and cache_flush.. -Ajay > Any other gadget > driver causing that, should be fixed too... > > -- > balbi > > DefectiveByDesign.org -- To unsubscribe from this list: send the line "unsubscribe linux-usb" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
