On Thu, Jul 29, 2010 at 10:12:09PM -0700, Sarah Sharp wrote: > This patch set increases the performance of the xHCI driver, and > decreases the CPU usage by the driver. > > The driver was tested by using dd to read the raw disk of an SSD behind > a USB 3.0 hard drive. This test was selected to eliminate any slowness > on the hardware side and reveal any software bottle necks (because SSD > read speed is much faster than write speed). I used the kernel perf > tool to look at the top kernel symbols during the dd to find the bottle > necks. perf annotate is very nice for revealing the exact code in the > function that is taking the most time. > > Along with the MSI patch from Dong Nguyen, these increase performance > from 172 MB/s to 204 MB/s, and decrease CPU usage on a 64-bit quad core > Intel system from 28% CPU usage to 19%. Heh, that's a performance increase bigger than most people ever saw USB 2.0 running at, very nice job. > Most of the improvements come from eliminating unnecessary register > reads and enabling 64-bit DMA to avoid bounce buffers. A little bit > of improvement comes from making functions static so the compiler can > optimize them. > > There is still more low-hanging performance work to do, but I wanted to > submit this set before the 2.6.36 merge window. > > Greg, this patchset should apply on top of Andiry's isochronous patches. > Please let me know if you run into any conflicts. Nope, looks good, and all now queued up. Thanks again for this work, it's wonderful and much appreciated. 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
