From: Paul Brook <paul@codesourcery.com> To: qemu-devel@nongnu.org Cc: "Chris Webb" <chris@arachsys.com>, "Avi Kivity" <avi@redhat.com>, kvm@vger.kernel.org, "Jernej Simončič" <jernej's-kvm@eternallybored.org> Subject: Re: [Qemu-devel] High CPU use of -usbdevice tablet (was Re: KVM usability) Date: Sun, 4 Apr 2010 15:25:17 +0100 [thread overview] Message-ID: <201004041525.18211.paul@codesourcery.com> (raw) In-Reply-To: <20100404123116.GA19866@arachsys.com> > > Looks like the tablet is set to 100 Hz polling rate. We may be able > > to get away with 30 Hz or even less (ep_bInterval, in ms, in > > hw/usb-wacom.c). > > Changing the USB tablet polling interval from 10ms to 100ms in both > hw/usb-wacom.c and hw/usb-hid.c made no difference except the an increase > in bInterval shown in lsusb -v in the guest and the hint of jerky mouse > movement I expected from setting this value so high. A similar change to > the polling interval for the keyboard and mouse also made no difference to > their performance impact. The USB HID devices implement the SET_IDLE command, so the polling interval will have no real effect on performance. My guess is that the overhead you're seeing is entirely from the USB host adapter having to wake up and check the transport descriptor lists. This will only result in the guest being woken if a device actually responds (as mentioned above it should not). >Taking the FRAME_TIMER_FREQ down to 100 in hw/usb-uhci.c does seem to reduce >the CPU load quite a bit, but at the expense of making the USB tablet (and >presumably all other USB devices) very laggy. The guest USB driver explicitly decides which devices to poll each frame. Slowing down the frame rate will effectively change the polling period by the same factor. e.g. the HID device requests a polling rate of 10ms, you slowed down frame rate by 10x, so you're efectively only polling every 100ms. If you want a quick and nasty hack then you can probably make the device wake up less often, and process multiple frames every wakeup. However this is probably going to do bad things (at best extremely poor performance) when using actual USB devices. Fixing this properly is hard because the transport descriptor lists are stores in system RAM, and polled by the host adapter. The first step is to read the whole table of descriptors, and calculate when the next event is due. However the guest will not explicitly notify the HBA when these tables are modified, so you also need some sort of MMU trap to trigger recalculation. This only gets you down to the base polling interval requested by the device. Increasing this interval causes significant user visible latency, so increasing it is not an option. The guest is also likely to distribute polling events evenly, further reducing the effective sleep interval. To fix this you need additional APIs so that a device can report when an endpoint will become unblocked, rather than just waiting to be polled and NAKing the request. Paul
WARNING: multiple messages have this Message-ID (diff)
From: Paul Brook <paul@codesourcery.com> To: qemu-devel@nongnu.org Cc: "Chris Webb" <chris@arachsys.com>, Jernej@gnu.org, "Avi Kivity" <avi@redhat.com>, kvm@vger.kernel.org, Simončič <jernej's-kvm@eternallybored.org> Subject: Re: [Qemu-devel] High CPU use of -usbdevice tablet (was Re: KVM usability) Date: Sun, 4 Apr 2010 15:25:17 +0100 [thread overview] Message-ID: <201004041525.18211.paul@codesourcery.com> (raw) In-Reply-To: <20100404123116.GA19866@arachsys.com> > > Looks like the tablet is set to 100 Hz polling rate. We may be able > > to get away with 30 Hz or even less (ep_bInterval, in ms, in > > hw/usb-wacom.c). > > Changing the USB tablet polling interval from 10ms to 100ms in both > hw/usb-wacom.c and hw/usb-hid.c made no difference except the an increase > in bInterval shown in lsusb -v in the guest and the hint of jerky mouse > movement I expected from setting this value so high. A similar change to > the polling interval for the keyboard and mouse also made no difference to > their performance impact. The USB HID devices implement the SET_IDLE command, so the polling interval will have no real effect on performance. My guess is that the overhead you're seeing is entirely from the USB host adapter having to wake up and check the transport descriptor lists. This will only result in the guest being woken if a device actually responds (as mentioned above it should not). >Taking the FRAME_TIMER_FREQ down to 100 in hw/usb-uhci.c does seem to reduce >the CPU load quite a bit, but at the expense of making the USB tablet (and >presumably all other USB devices) very laggy. The guest USB driver explicitly decides which devices to poll each frame. Slowing down the frame rate will effectively change the polling period by the same factor. e.g. the HID device requests a polling rate of 10ms, you slowed down frame rate by 10x, so you're efectively only polling every 100ms. If you want a quick and nasty hack then you can probably make the device wake up less often, and process multiple frames every wakeup. However this is probably going to do bad things (at best extremely poor performance) when using actual USB devices. Fixing this properly is hard because the transport descriptor lists are stores in system RAM, and polled by the host adapter. The first step is to read the whole table of descriptors, and calculate when the next event is due. However the guest will not explicitly notify the HBA when these tables are modified, so you also need some sort of MMU trap to trigger recalculation. This only gets you down to the base polling interval requested by the device. Increasing this interval causes significant user visible latency, so increasing it is not an option. The guest is also likely to distribute polling events evenly, further reducing the effective sleep interval. To fix this you need additional APIs so that a device can report when an endpoint will become unblocked, rather than just waiting to be polled and NAKing the request. Paul
next prev parent reply other threads:[~2010-04-04 14:25 UTC|newest] Thread overview: 109+ messages / expand[flat|nested] mbox.gz Atom feed top [not found] <1267068445.1726.25.camel@localhost> [not found] ` <1267089644.12790.74.camel@laptop> 2010-02-26 2:49 ` Enhance perf to support KVM Zhang, Yanmin 2010-02-26 9:01 ` Ingo Molnar 2010-02-26 9:53 ` Avi Kivity 2010-02-26 10:35 ` Ingo Molnar 2010-02-26 10:47 ` Avi Kivity 2010-02-26 11:17 ` Ingo Molnar 2010-02-26 11:44 ` Avi Kivity 2010-02-26 12:46 ` Ingo Molnar 2010-02-26 12:54 ` Avi Kivity 2010-02-26 13:16 ` Ingo Molnar 2010-02-26 13:57 ` Jes Sorensen 2010-02-26 14:04 ` Avi Kivity 2010-02-26 14:23 ` Ingo Molnar 2010-02-26 15:06 ` Avi Kivity 2010-03-02 16:46 ` Paolo Bonzini 2010-03-02 17:12 ` Arnaldo Carvalho de Melo 2010-03-02 17:20 ` Paolo Bonzini 2010-03-02 17:24 ` Ingo Molnar 2010-03-02 17:17 ` Ingo Molnar 2010-03-07 14:17 ` Avi Kivity 2010-02-26 18:33 ` Avi Kivity 2010-02-27 10:56 ` KVM usability Ingo Molnar 2010-02-27 13:30 ` Jan Kiszka 2010-02-27 13:30 ` [Qemu-devel] " Jan Kiszka 2010-02-27 14:48 ` Ian Kirk 2010-02-27 15:32 ` Zachary Amsden 2010-02-27 17:25 ` Ingo Molnar 2010-03-01 15:33 ` Anthony Liguori 2010-03-01 16:48 ` Zachary Amsden 2010-03-01 17:41 ` Arnaldo Carvalho de Melo 2010-03-01 18:29 ` Zachary Amsden 2010-03-01 20:56 ` Ingo Molnar 2010-03-01 21:45 ` Anthony Liguori 2010-03-01 22:06 ` Zachary Amsden 2010-03-02 0:33 ` Ingo Molnar 2010-03-02 0:30 ` Ingo Molnar 2010-03-02 2:34 ` Anthony Liguori 2010-03-02 8:39 ` Chris Webb 2010-03-07 18:42 ` Avi Kivity 2010-03-02 10:30 ` Ingo Molnar 2010-03-07 9:35 ` Avi Kivity 2010-03-07 9:56 ` Pekka Enberg 2010-03-07 10:11 ` Avi Kivity 2010-03-07 18:42 ` Ingo Molnar 2010-03-07 15:14 ` Luca Barbieri 2010-03-07 18:16 ` Avi Kivity 2010-03-07 18:01 ` Arnaldo Carvalho de Melo 2010-03-07 18:15 ` Avi Kivity 2010-03-07 18:53 ` Arnaldo Carvalho de Melo 2010-03-07 19:05 ` Avi Kivity 2010-03-07 18:25 ` Avi Kivity 2010-03-01 9:25 ` Ingo Molnar 2010-03-01 15:36 ` Anthony Liguori 2010-03-01 15:14 ` Anthony Liguori 2010-03-01 15:42 ` Daniel P. Berrange 2010-03-02 1:12 ` Dustin Kirkland 2010-03-02 10:11 ` Peter Zijlstra 2010-03-02 13:37 ` Nikolai K. Bochev 2010-03-02 14:22 ` Gerd Hoffmann 2010-03-02 14:29 ` Ingo Molnar 2010-03-07 9:22 ` Avi Kivity 2010-03-02 14:37 ` Daniel P. Berrange 2010-03-02 14:52 ` Gerd Hoffmann 2010-03-02 14:56 ` Daniel P. Berrange 2010-03-02 15:13 ` Gerd Hoffmann 2010-03-04 20:00 ` Lucas Meneghel Rodrigues 2010-03-04 20:13 ` Zachary Amsden 2010-03-04 20:34 ` Anthony Liguori 2010-03-04 22:23 ` H. Peter Anvin 2010-03-05 7:44 ` Markus Armbruster 2010-03-07 11:25 ` Avi Kivity 2010-03-01 21:12 ` Dustin Kirkland 2010-03-01 21:59 ` Anthony Liguori 2010-03-02 2:34 ` Alexander Graf 2010-03-02 2:36 ` Anthony Liguori 2010-03-09 13:32 ` Avi Kivity 2010-03-09 14:32 ` Dustin Kirkland 2010-03-09 14:38 ` Alexander Graf 2010-03-09 14:50 ` Anthony Liguori 2010-03-09 14:52 ` Avi Kivity 2010-03-09 14:57 ` Anthony Liguori 2010-03-09 17:11 ` Avi Kivity 2010-03-09 17:27 ` Anthony Liguori 2010-03-09 17:30 ` Avi Kivity 2010-03-09 14:49 ` Anthony Liguori 2010-03-09 14:54 ` Avi Kivity 2010-03-02 3:02 ` Dustin Kirkland 2010-03-02 8:21 ` Chris Webb 2010-03-02 14:54 ` Dustin Kirkland [not found] ` <428008581.20100302103400@eternallybored.org> 2010-03-07 11:35 ` Avi Kivity 2010-04-04 12:31 ` High CPU use of -usbdevice tablet (was Re: KVM usability) Chris Webb 2010-04-04 12:31 ` [Qemu-devel] " Chris Webb 2010-04-04 14:25 ` Paul Brook [this message] 2010-04-04 14:25 ` Paul Brook 2010-04-04 16:58 ` Avi Kivity 2010-04-04 16:58 ` Avi Kivity 2010-04-04 21:03 ` Paul Brook 2010-04-04 21:03 ` Paul Brook 2010-04-04 21:53 ` Paul Brook 2010-04-04 21:53 ` Paul Brook 2010-04-05 8:22 ` Avi Kivity 2010-04-05 8:22 ` Avi Kivity 2010-03-03 2:57 ` KVM usability Ross Boylan 2010-03-03 8:55 ` Daniel P. Berrange 2010-03-03 15:42 ` Ross Boylan 2010-03-07 14:29 ` Avi Kivity 2010-02-26 11:48 ` Enhance perf to support KVM Peter Zijlstra 2010-02-26 11:53 ` Avi Kivity 2010-02-26 20:17 ` Anthony Liguori
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=201004041525.18211.paul@codesourcery.com \ --to=paul@codesourcery.com \ --cc=avi@redhat.com \ --cc=chris@arachsys.com \ --cc=jernej's-kvm@eternallybored.org \ --cc=kvm@vger.kernel.org \ --cc=qemu-devel@nongnu.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.