Patchwork PATCH: QEMU support for UHCI suspend / remote wake up

login
register
mail settings
Submitter Marcelo Tosatti
Date Nov. 25, 2010, 3:34 p.m.
Message ID <20101125153455.GA4097@amt.cnet>
Download mbox | patch
Permalink /patch/73072/
State New
Headers show

Comments

Marcelo Tosatti - Nov. 25, 2010, 3:34 p.m.
This patch enables USB UHCI global suspend/resume feature. The OS will
stop the HC once all ports are suspended. If there is activity on the
port(s), an interrupt signalling remote wakeup will be triggered.

To enable autosuspend for the USB tablet on Linux guests:

echo auto > /sys/devices/pci0000:00/0000:00:01.2/usb1/1-1/power/level

It reduces CPU consumption of an idle FC12 guest from 2.7% to 0.3%.

Signed-off-by: Marcelo Tosatti <mtosatti@redhat.com>
Gerd Hoffmann - Nov. 25, 2010, 4:15 p.m.
Hi,

> +        dev->info->remote_wakeup_cb = uhci_event;

> diff --git a/hw/usb.h b/hw/usb.h
> index 00d2802..16de1c9 100644
> --- a/hw/usb.h
> +++ b/hw/usb.h
> @@ -189,6 +189,11 @@ struct USBDeviceInfo {
>        */
>       int (*handle_data)(USBDevice *dev, USBPacket *p);
>
> +    /*
> +     * Process remote wakeup request.
> +     */
> +    void (*remote_wakeup_cb)(USBDevice *dev);
> +

No way.  DeviceInfo holds informations about the device 
*implementation*.  Multiple instances will share that, placing runtime 
data there is just plain wrong.

Also this is a callback from the usb device to the usb host adapter, not 
the other way around.  I'd suggest to create a USBBusOps for host 
adapter callbacks and and stick it into USBBus.

cheers,
   Gerd
Marcelo Tosatti - Nov. 25, 2010, 5:04 p.m.
See patch 2 for details.

v2: 
- Add remote wakeup callback in USBBus, as suggested by Gerd.
Paul Brook - Nov. 26, 2010, 12:38 a.m.
> This patch enables USB UHCI global suspend/resume feature. The OS will
> stop the HC once all ports are suspended. If there is activity on the
> port(s), an interrupt signalling remote wakeup will be triggered.

I'm pretty sure this is wrong.  Suspend/resume works based on physical 
topology, i.e. the resume notification should go to the the port/hub to which 
the device is connected, not directly to the host controller.

Paul
Marcelo Tosatti - Nov. 26, 2010, 2:15 a.m.
On Fri, Nov 26, 2010 at 12:38:28AM +0000, Paul Brook wrote:
> > This patch enables USB UHCI global suspend/resume feature. The OS will
> > stop the HC once all ports are suspended. If there is activity on the
> > port(s), an interrupt signalling remote wakeup will be triggered.
> 
> I'm pretty sure this is wrong.  Suspend/resume works based on physical 
> topology, i.e. the resume notification should go to the the port/hub to which 
> the device is connected, not directly to the host controller.

If the host controller is in global suspend state, and resume is
received on any of its root hub ports (given that remote wakeup is
enabled for the given port), the system will be interrupted if interrupt
enable bit is set.

2.1.2 USB STATUS REGISTER
I/O Address: Base + (02-03h)
Default Value: 0000h
Attribute: Read/Write Clear
size: 16 bits
This register indicates pending interrupts and various states of the
Host Controller

Resume Detect. The Host Controller sets this bit to 1 when it receives
a “RESUME” signal from a USB device. This is only valid if the Host
Controller is in a global suspend state (bit 3 of Command register = 1).

2.1.7.1 Behavior Under Global or Selective Suspend Scenarios

Resume will be recognized in the USBSTS register (bit 2) if resume is
received on a suspended or enabled port when the Host Controller is in
the global suspend state (USBCMD register bit 3 is set).

4.2.1 RESUME RECEIVED
This event indicates that the HC received a RESUME signal from a device
on the USB during a global suspend. If
this interrupt is enabled in the HC Interrupt Enable register, a
hardware interrupt will be signaled to the system
allowing the USB to be brought out of the suspend state and returned to
normal operation.

You are correct in that USB HUB emulation does not propagate resume, but
this does not make this patch incorrect.
Gerd Hoffmann - Nov. 26, 2010, 8:49 a.m.
On 11/26/10 03:15, Marcelo Tosatti wrote:
> On Fri, Nov 26, 2010 at 12:38:28AM +0000, Paul Brook wrote:
>>> This patch enables USB UHCI global suspend/resume feature. The OS will
>>> stop the HC once all ports are suspended. If there is activity on the
>>> port(s), an interrupt signalling remote wakeup will be triggered.
>>
>> I'm pretty sure this is wrong.  Suspend/resume works based on physical
>> topology, i.e. the resume notification should go to the the port/hub to which
>> the device is connected, not directly to the host controller.

> You are correct in that USB HUB emulation does not propagate resume, but
> this does not make this patch incorrect.

Well, it does.  When the notification is port based our software model 
should better reflect that, so we have the chance to add resume 
propagation to the hub emulation later on.

I guess the Ops should be moved from the USBBus to the USBPort to 
reflect that.  This way the hub emulation and the uhci root hub can have 
different callbacks, which is needed to get this correct.

cheers,
   Gerd
Paul Brook - Nov. 26, 2010, 12:09 p.m.
> On 11/26/10 03:15, Marcelo Tosatti wrote:
> > On Fri, Nov 26, 2010 at 12:38:28AM +0000, Paul Brook wrote:
> >>> This patch enables USB UHCI global suspend/resume feature. The OS will
> >>> stop the HC once all ports are suspended. If there is activity on the
> >>> port(s), an interrupt signalling remote wakeup will be triggered.
> >> 
> >> I'm pretty sure this is wrong.  Suspend/resume works based on physical
> >> topology, i.e. the resume notification should go to the the port/hub to
> >> which the device is connected, not directly to the host controller.
> > 
> > You are correct in that USB HUB emulation does not propagate resume, but
> > this does not make this patch incorrect.
> 
> Well, it does.  When the notification is port based our software model
> should better reflect that, so we have the chance to add resume
> propagation to the hub emulation later on.

Exactly. The patch assumes the device is connected to a root hub port. This 
assumption is incorrect.

The device should be sending the resume signal to the port/hub to which it is 
connected. If that hub is still active it will reactivate the port, and flag a 
port change notification in the normal manner. If the hub is also suspended it 
will propagate the resume notification upstream (which may or may not be the 
root hub).

Paul
Marcelo Tosatti - Dec. 1, 2010, 4:47 p.m.
v3:
- Move remote wakeup callback to USBPort
- Add subsection

Patch

diff --git a/hw/usb-hid.c b/hw/usb-hid.c
index 882d933..b7a4dc1 100644
--- a/hw/usb-hid.c
+++ b/hw/usb-hid.c
@@ -412,6 +412,9 @@  static void usb_hid_changed(USBHIDState *hs)
 
     if (hs->datain)
         hs->datain(hs->datain_opaque);
+
+    if (hs->dev.remote_wakeup)
+        usb_remote_wakeup(&hs->dev);
 }
 
 static void usb_mouse_event(void *opaque,
diff --git a/hw/usb-uhci.c b/hw/usb-uhci.c
index 1d83400..674cb0c 100644
--- a/hw/usb-uhci.c
+++ b/hw/usb-uhci.c
@@ -59,6 +59,7 @@ 
 
 #define UHCI_PORT_RESET (1 << 9)
 #define UHCI_PORT_LSDA  (1 << 8)
+#define UHCI_PORT_RD    (1 << 6)
 #define UHCI_PORT_ENC   (1 << 3)
 #define UHCI_PORT_EN    (1 << 2)
 #define UHCI_PORT_CSC   (1 << 1)
@@ -501,6 +502,7 @@  static void uhci_ioport_writew(void *opaque, uint32_t addr, uint32_t val)
             port->ctrl = (port->ctrl & 0x01fb) | (val & ~0x01fb);
             /* some bits are reset when a '1' is written to them */
             port->ctrl &= ~(val & 0x000a);
+            port->ctrl &= ~(port->ctrl & 0x0040); /* clear port resume detected */
         }
         break;
     }
@@ -593,6 +595,43 @@  static void uhci_resume (void *opaque)
     }
 }
 
+static UHCIPort *find_port(UHCIState *s, USBDevice *d)
+{
+    int i;
+
+    for (i = 0; i < NB_PORTS; i++) {
+        UHCIPort *port = &s->ports[i];
+
+        if (d == port->port.dev) {
+            return port;
+        }
+    }
+
+    return NULL;
+}
+
+static void uhci_event(USBDevice *dev)
+{
+    USBBus *bus = usb_bus_from_device(dev);
+    UHCIState *s = container_of(bus, UHCIState, bus);
+
+    if (s->cmd & UHCI_CMD_EGSM) {
+        UHCIPort *port = find_port(s, dev);
+
+        if (!port) {
+            return;
+        }
+
+        if (port->ctrl & UHCI_PORT_RD) {
+            return;
+        }
+
+        port->ctrl |= UHCI_PORT_RD;
+
+        uhci_resume(s);
+    }
+}
+
 static void uhci_attach(USBPort *port1, USBDevice *dev)
 {
     UHCIState *s = port1->opaque;
@@ -602,6 +641,7 @@  static void uhci_attach(USBPort *port1, USBDevice *dev)
         if (port->port.dev) {
             usb_attach(port1, NULL);
         }
+        dev->info->remote_wakeup_cb = uhci_event;
         /* set connect status */
         port->ctrl |= UHCI_PORT_CCS | UHCI_PORT_CSC;
 
diff --git a/hw/usb.c b/hw/usb.c
index a326bcf..9b24d49 100644
--- a/hw/usb.c
+++ b/hw/usb.c
@@ -229,3 +229,9 @@  void usb_send_msg(USBDevice *dev, int msg)
 
     /* This _must_ be synchronous */
 }
+
+void usb_remote_wakeup(USBDevice *dev)
+{
+    if (dev->info->remote_wakeup_cb)
+        dev->info->remote_wakeup_cb(dev);
+}
diff --git a/hw/usb.h b/hw/usb.h
index 00d2802..16de1c9 100644
--- a/hw/usb.h
+++ b/hw/usb.h
@@ -189,6 +189,11 @@  struct USBDeviceInfo {
      */
     int (*handle_data)(USBDevice *dev, USBPacket *p);
 
+    /*
+     * Process remote wakeup request.
+     */
+    void (*remote_wakeup_cb)(USBDevice *dev);
+
     const char *product_desc;
 
     /* handle legacy -usbdevice command line options */
@@ -317,6 +322,7 @@  void usb_unregister_port(USBBus *bus, USBPort *port);
 int usb_device_attach(USBDevice *dev);
 int usb_device_detach(USBDevice *dev);
 int usb_device_delete_addr(int busnr, int addr);
+void usb_remote_wakeup(USBDevice *dev);
 
 static inline USBBus *usb_bus_from_device(USBDevice *d)
 {