Hello Frederic, On Fri, May 10, 2013 at 12:54 PM, Frederik Schmid <frederik.schmid@xxxxxxxxx> wrote: > Well, my conclusion is that this setup, IDS-camera + musb, is horribly > sensitive to interrupt latency. > > If the musb-interrupt is blocked for ~100us the pipe is stalled. Most of the > interrupts on my target were routed to CPU0 while CPU1 had few. > > When I re-route the musb interrupts to CPU1 throughput increases to ~120Mbps > but I've only reduced the probability of the interrupt being blocked. If you want to increase throughput over musb on omap4, usually next steps help: - route musb IRQ to another CPU if possible - check if musb uses DMA mode (not PIO) - check if the data (urb) passed to musb is aligned, if not - apply next patch: https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git/commit/?id=8408fd1d83e39bf856d31a36b70bcc53527702fd - play with double packet buffering - increase OPP - this will reduce latency But I can say that on OMAP4 throughput over musb will be lower than over ehci controller because ehci one does many things in HW whereas musb have to do some of them in SW. My scores for musb were ~145 Mbits/s for bulk transfers, but they probably may be increased after additional tuning. > > The only hope this application has is if MODE1 in the controller could > alleviate latency dependence but MODE1 seems quite broken in the hardware? > > musb_host.c: > > /* Disadvantage of using mode 1: > * It's basically usable only for mass storage class; essentially all > * other protocols also terminate transfers on short packets. > * ... > */ > > Best regards, > Frederik > > On Friday 10 May 2013 07.26.35 Frederik Schmid wrote: >> Hi Greg, >> >> The kernel is based on 3.4.11 from omapzoom.org (ti-ubuntu-3.4-1487) with >> Variscite BSP patches from: >> >> http://www.variwiki.com/index.php?title=VAR-SOM-OM44_-_Ubuntu_Precise >> >> I was expecting 200-250Mbps which is entirly possible on the same chip using >> its ehci-controller. musb seems to sustain this speed also, for most of the >> time, until a "glitch" like the one in the screenshot stalls the pipe. >> >> We don't want to use the ehci-controller because it's already bogged down >> with an ethernet controller. >> >> Regards, >> Frederik >> >> On Thursday 09 May 2013 08.12.47 Greg KH wrote: >> > On Thu, May 09, 2013 at 10:44:05AM +0200, Frederik Schmid wrote: >> > > Hi, >> > > >> > > I'm developing a camera application on a TI OMAP4460. I have a cmos >> > > usb-camera from IDS (UI-1551-LE-C-HQ) connected to the OTG-port (musb as >> > > host) on the OMAP. >> > > >> > > I get very poor frame rates with this setup reliably. The throughput >> > > rate >> > > is ~50-70Mbps. (1600x1200,8bpp,3-5fps) >> > > >> > > I uploaded a screenshot of a packet trace here: >> > > >> > > http://i.imgur.com/26XL23T.png >> > > >> > > The host seems to keep up with the camera most of the time but >> > > occasionally >> > > some kind of "glitch" causes the host to lag behind and the camera >> > > signals >> > > STALL because of buffer overrun. >> > >> > What kernel version are you using? And what data rate do you expect to >> > be getting with this hardware configuration? >> > >> > thanks, >> > >> > greg k-h >> >> -- >> 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 > -- > 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 -- Best regards, Ruslan Bilvol -- 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
