Laszlo and Gerd The timer mentioned in my commit log is pure s/w timer, which is used to poll if those async interrupt transfers are done or not.
In other words, it does not impact host controller and device behavior, just check the completion of those transactions with higher frequency. So the cpu consumption in virtualization platform for Uhci/Ehci/Xhci h/w simulation should be same, just there is one more 1ms s/w timer. Thanks Feng -----Original Message----- From: Gerd Hoffmann [mailto:kra...@redhat.com] Sent: Thursday, June 04, 2015 23:36 To: Laszlo Ersek Cc: Tian, Feng; edk2-devel@lists.sourceforge.net; Zeng, Star Subject: Re: [edk2] [patch 0/4] EDKII Usb stack async transfer performance tunning On Do, 2015-06-04 at 15:24 +0200, Laszlo Ersek wrote: > On 06/04/15 08:16, Tian Feng wrote: > > Update the value of those async polling interval macros from 50ms to 1ms. > > Out of pure curiosity, how does this manifest in CPU consumption? > > USB 1 and 2 are CPU hungry when virtualized. > > USB 3 / XHCI is more amicable and (as far as I know) the answer for > the "my virtual USB devices consume too much CPU even when idle" > problem. I wonder if increasing the XHCI polling rate 50-fold will > impact VMs visibly. (Admittedly, a VM spends a fraction of its > lifetime executing OVMF, hence "pure curiosity" on my part.) Probably not enough to make the user notice, but I'm pretty sure this is measurable with any host adapter, assuming there is a device hooked up which needs polling (kbd or mouse for example). cheers, Gerd ------------------------------------------------------------------------------ _______________________________________________ edk2-devel mailing list edk2-devel@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/edk2-devel
