driverfs support for USB - take 2
Subsystems (3)
PATCH 1/1 driverfs support for USB - take 2
Hi,
Well after determining that the last version of this patch doesn't show the USB tree properly, here's another patch against 2.5.3-pre6 that fixes this issue.
With this patch (the driver core changes were from Pat Mochel, thanks Pat for putting up with my endless questions) my machine now shows the following tree:
[greg@soap x]$ tree
.
`-- root
|-- pci0
| |-- 00:00.0
| | |-- power
| | `-- status
| |-- 00:01.0
| | |-- 02:00.0
| | | |-- power
| | | `-- status
| | |-- power
| | `-- status
| |-- 00:1e.0
| | |-- 01:08.0
| | | |-- power
| | | `-- status
| | |-- 01:0d.0
| | | |-- power
| | | |-- status
| | | `-- usb_bus
| | | |-- power
| | | `-- status
| | |-- 01:0d.1
| | | |-- power
| | | |-- status
| | | `-- usb_bus
| | | |-- 003
| | | | |-- power
| | | | `-- status
| | | |-- power
| | | `-- status
| | |-- 01:0d.2
| | | |-- power
| | | |-- status
| | | `-- usb_bus
| | | |-- power
| | | `-- status
| | |-- power
| | `-- status
| |-- 00:1f.0
| | |-- power
| | `-- status
| |-- 00:1f.1
| | |-- power
| | `-- status
| |-- 00:1f.2
| | |-- power
| | |-- status
| | `-- usb_bus
| | |-- 002
| | | |-- 003
| | | | |-- power
| | | | `-- status
| | | |-- power
| | | `-- status
| | |-- power
| | `-- status
| |-- 00:1f.3
| | |-- power
| | `-- status
| |-- 00:1f.5
| | |-- power
| | `-- status
| |-- power
| `-- status
|-- power
`-- statusI have 2 USB controllers in this system, 1 UHCI controller on the motherboard that shows up under PCI device 00:1f.2. There are two devices plugged into this bus, a hub, and a trackball plugged into the hub. This topology is now shown properly.
The other controller is a USB 2.0 controller that shows up in 3 places on the pci bus: 01:0d.0, 01:0d.1, and 01:0d.2. The 01:0d.1 device is controlled by the ohci-hcd driver, and a USB hub is plugged into it. The ehci-hcd driver controls the other 2 pci devices on the card.
Yes, I need to have better names for the devices than just "usb_bus", any suggestions? These devices nodes are really the USB root hubs in the USB controller, so they could just have the USB number as the name like the other USB devices (001), but that's pretty boring :)
greg k-h
drivers/base/core.c +23 -78
@@ -17,10 +17,7 @@ # define DBG(x...)#endif -static struct iobus device_root = { - bus_id:"root" , - name:"Logical System Root" , -}; +static struct device * device_root;int (*platform_notify)(struct device * dev) =NULL ;int (*platform_notify_remove)(struct device * dev) =NULL ;
@@ -28,9 +25,6 @@extern int device_make_dir (struct device * dev);extern void device_remove_dir (struct device * dev); -extern int iobus_make_dir (struct iobus * iobus); -extern void iobus_remove_dir (struct iobus * iobus); -static spinlock_t device_lock;/**
@@ -47,19 +41,25 @@if (!dev || !strlen (dev->bus_id))return -EINVAL ; -BUG_ON (!dev->parent);spin_lock (&device_lock);INIT_LIST_HEAD (&dev->node); +INIT_LIST_HEAD (&dev->children);spin_lock_init (&dev->lock);atomic_set (&dev->refcount,2 ); -get_iobus (dev->parent); -list_add_tail (&dev->node,&dev->parent->devices); +if (dev != device_root) { +struct device * parent; +if (!dev->parent) + dev->parent = device_root; + parent = dev->parent; +get_device (parent); +list_add_tail (&dev->node,&parent->children); + }spin_unlock (&device_lock); -DBG ("DEV: registering device: ID = '%s', name = %s, parent = %s\n" , - dev->bus_id, dev->name, parent->bus_id); +DBG ("DEV: registering device: ID = '%s', name = %s\n" , + dev->bus_id, dev->name);if ((error =device_make_dir (dev))) goto register_done;
@@ -70,8 +70,8 @@ register_done:put_device (dev); -if (error) -put_iobus (dev->parent); +if (error && dev->parent) +put_device (dev->parent);return error; }
@@ -100,9 +100,6 @@/* remove the driverfs directory */ device_remove_dir (dev); -if (dev->subordinate) -iobus_remove_dir (dev->subordinate); -/* Notify the platform of the removal, in case they * need to do anything... */
@@ -116,70 +113,18 @@if (dev->driver && dev->driver->remove) dev->driver->remove (dev,REMOVE_FREE_RESOURCES ); -put_iobus (dev->parent); -} - -int iobus_register (struct iobus *bus) -{ -int error; - -if (!bus || !strlen (bus->bus_id)) -return -EINVAL ; - -spin_lock (&device_lock); -atomic_set (&bus->refcount,2 ); -spin_lock_init (&bus->lock); -INIT_LIST_HEAD (&bus->node); -INIT_LIST_HEAD (&bus->devices); -INIT_LIST_HEAD (&bus->children); - -if (bus != &device_root) { -if (!bus->parent) - bus->parent = &device_root; -get_iobus (bus->parent); -list_add_tail (&bus->node,&bus->parent->children); - } -spin_unlock (&device_lock); - -DBG ("DEV: registering bus. ID = '%s' name = '%s' parent = %p\n" , - bus->bus_id,bus->name,bus->parent); - - error =iobus_make_dir (bus); - -put_iobus (bus); -if (error && bus->parent) -put_iobus (bus->parent); -return error; -} - -/** -* iobus_unregister - remove bus and children from device tree -* @bus: pointer to bus structure -* -* Remove device from parent's list of children and decrement -* reference count on controlling device. That should take care of -* the rest of the cleanup. -*/ -void put_iobus (struct iobus * iobus) -{ -if (!atomic_dec_and_lock (&iobus->refcount,&device_lock)) -return ; -list_del_init (&iobus->node); -spin_unlock (&device_lock); - -if (!list_empty (&iobus->devices) || - !list_empty (&iobus->children)) -BUG (); - -put_iobus (iobus->parent); -/* unregister itself */ -put_device (iobus->self); +put_device (dev->parent); }static int __init device_init_root (void ) { -/* initialize parent bus lists */ -return iobus_register (&device_root); + device_root =kmalloc (sizeof (*device_root),GFP_KERNEL ); +if (!device_root) +return -ENOMEM ; +memset (device_root,0 ,sizeof (*device_root)); +strcpy (device_root->bus_id,"root" ); +strcpy (device_root->name,"System Root" ); +return device_register (device_root); }int __init device_driver_init (void )
@@ -208,5 +153,5 @@ }EXPORT_SYMBOL (device_register); -EXPORT_SYMBOL (iobus_register); +EXPORT_SYMBOL (put_device);EXPORT_SYMBOL (device_driver_init);
drivers/base/fs.c +0 -22
@@ -102,28 +102,6 @@ } }return 0 ; -} - -void iobus_remove_dir (struct iobus * iobus) -{ -if (iobus) -driverfs_remove_dir (&iobus->dir); -} - -int iobus_make_dir (struct iobus * iobus) -{ -struct driver_dir_entry * parent =NULL ; -int error; - -INIT_LIST_HEAD (&iobus->dir.files); - iobus->dir.mode = (S_IFDIR |S_IRWXU |S_IRUGO |S_IXUGO ); - iobus->dir.name = iobus->bus_id; - -if (iobus->parent) - parent = &iobus->parent->dir; - - error =driverfs_create_dir (&iobus->dir,parent); -return error; }EXPORT_SYMBOL (device_create_file);
drivers/pci/pci.c +7 -16
@@ -1086,13 +1086,7 @@ child->parent = parent; child->ops = parent->ops; child->sysdata = parent->sysdata; - -/* init generic fields */ - child->iobus.self = &dev->dev; - child->iobus.parent = &parent->iobus; - dev->dev.subordinate = &child->iobus; - -strcpy (child->iobus.name,dev->dev.name); + child->dev = &dev->dev;/* * Set up the primary, secondary and subordinate
@@ -1361,16 +1355,11 @@DBG ("Scanning bus %02x\n" , bus->number); max = bus->secondary; -/* we should know for sure what the bus number is, so set the bus ID -* for the bus and make sure it's registered in the device tree */ -sprintf (bus->iobus.bus_id,"pci%d" ,bus->number); -iobus_register (&bus->iobus); -/* Create a device template */ memset (&dev0,0 ,sizeof (dev0)); dev0.bus = bus; dev0.sysdata = bus->sysdata; - dev0.dev.parent = &bus->iobus; + dev0.dev.parent = bus->dev; dev0.dev.driver = &pci_device_driver;/* Go find them, Rover! */
@@ -1430,9 +1419,11 @@return NULL ;list_add_tail (&b->node, &pci_root_buses); -sprintf (b->iobus.bus_id,"pci%d" ,bus); -strcpy (b->iobus.name,"Host/PCI Bridge" ); -iobus_register (&b->iobus); + b->dev =kmalloc (sizeof (*(b->dev)),GFP_KERNEL ); +memset (b->dev,0 ,sizeof (*(b->dev))); +sprintf (b->dev->bus_id,"pci%d" ,bus); +strcpy (b->dev->name,"Host/PCI Bridge" ); +device_register (b->dev); b->number = b->secondary = bus; b->resource[0 ] = &ioport_resource;
drivers/usb/hcd/ehci-hcd.c +6 -0
@@ -290,6 +290,12 @@ goto done2; } +/* hook up the root hub to the pci controller in the device tree */ + udev->dev.parent = &ehci->hcd.pdev->dev; +strcpy (udev->dev.name,"usb_name" ); +strcpy (udev->dev.bus_id,"usb_bus" ); +device_register (&udev->dev); +return 0 ; }
drivers/usb/hcd/ohci-hcd.c +7 -1
@@ -475,7 +475,13 @@// FIXME cleanup return -ENODEV ; } - + +/* hook up the root hub to the pci controller in the device tree */ + udev->dev.parent = &ohci->hcd.pdev->dev; +strcpy (udev->dev.name,"usb_name" ); +strcpy (udev->dev.bus_id,"usb_bus" ); +device_register (&udev->dev); +return 0 ; }
drivers/usb/hub.c +9 -1
@@ -722,8 +722,16 @@ dev->bus->busnum, dev->devpath, dev->devnum);/* Run it through the hoops (find a driver, etc) */ -if (!usb_new_device (dev)) +if (!usb_new_device (dev)) { +/* put the device in the global device tree */ + dev->dev.parent = &dev->parent->dev; +sprintf (&dev->dev.name[0 ],"USB device %04x:%04x" , + dev->descriptor.idVendor, + dev->descriptor.idProduct); +sprintf (&dev->dev.bus_id[0 ],"%03d" , dev->devnum); +device_register (&dev->dev); goto done; + }/* Free the configuration if there was an error */ usb_free_dev (dev);
drivers/usb/uhci.c +6 -0
@@ -2848,6 +2848,12 @@ goto err_start_root_hub; } +/* hook the root hub up to the pci controller in the device tree */ + uhci->rh.dev->dev.parent = &dev->dev; +strcpy (uhci->rh.dev->dev.name,"usb_name" ); +strcpy (uhci->rh.dev->dev.bus_id,"usb_bus" ); +device_register (&uhci->rh.dev->dev); +return 0 ;/*
drivers/usb/usb-ohci.c +7 -0
@@ -2508,6 +2508,13 @@#ifdef DEBUG ohci_dump (ohci,1 );#endif + +/* hook the root hub up to the pci controller in the device tree */ + ohci->bus->root_hub->dev.parent = &dev->dev; +strcpy (ohci->bus->root_hub->dev.name,"usb_name" ); +strcpy (ohci->bus->root_hub->dev.bus_id,"usb_bus" ); +device_register (&ohci->bus->root_hub->dev); +return 0 ; }
drivers/usb/usb-uhci.c +6 -0
@@ -3028,6 +3028,12 @@pci_set_drvdata (dev, s); devs=s; +/* hook the root hub up to the pci controller in the device tree */ + s->bus->root_hub->dev.parent = &dev->dev; +strcpy (s->bus->root_hub->dev.name,"usb_name" ); +strcpy (s->bus->root_hub->dev.bus_id,"usb_bus" ); +device_register (&s->bus->root_hub->dev); +return 0 ; }
drivers/usb/usb.c +1 -0
@@ -1925,6 +1951,7 @@if (dev->devnum >0 ) {clear_bit (dev->devnum, &dev->bus->devmap.devicemap);usbfs_remove_device (dev); +put_device (&dev->dev); }/* Free up the device itself */
include/linux/device.h +2 -4
@@ -66,10 +66,8 @@struct device {struct list_head node;/* node in sibling list */ -struct iobus *parent;/* parent bus */ - -struct iobus *subordinate;/* only valid if this device is a -bridge device */ +struct list_head children; +struct device *parent;char name[DEVICE_NAME_SIZE ];/* descriptive ascii string */ char bus_id[BUS_ID_SIZE ];/* position on parent bus */
include/linux/pci.h +1 -2
@@ -432,8 +432,7 @@unsigned char productver;/* product version */ unsigned char checksum;/* if zero - checksum passed */ unsigned char pad1; - -struct iobus iobus;/* Generic device interface */ +struct device * dev; };#define pci_bus_b (n) list_entry(n, struct pci_bus, node)
include/linux/usb.h +4 -0
@@ -1,6 +1,8 @@#ifndef __LINUX_USB_H#define __LINUX_USB_H +#include <linux/device.h> +/* USB constants */ /*
@@ -1037,6 +1042,8 @@struct usb_device *parent;struct usb_bus *bus;/* Bus we're part of */ + +struct device dev;/* Generic device interface */ struct usb_device_descriptor descriptor;/* Descriptor */ struct usb_config_descriptor *config;/* All of the configs */
Quoted 9 lines · Greg KH
Hi,
Well after determining that the last version of this patch doesn't show the USB tree properly, here's another patch against 2.5.3-pre6 that fixes this issue.
With this patch (the driver core changes were from Pat Mochel, thanks Pat for putting up with my endless questions) my machine now shows the following tree:
Looks good, now when the PM bits can power down from the leaves, and turn off USB devices /before/ the USB controller and PCI bridges, which sounds much more sensible.
Quoted 4 lines
Yes, I need to have better names for the devices than just "
usb_bus", any suggestions? These devices nodes are really the USB root hubs in the USB controller, so they could just have the USB number as the name like the other USB devices (001), but that's pretty boring :)
"usb_root0" .. "usb_rootN" ?
btw, a script to marry the busid's from driverfs to lspci/lsusb output may be useful in the future especially if combined somehow with tree(1). Could be very handy when it gets time to debug those "My system won't suspend to disk" "What does /driver look like?" situations.
Signature · 2 lines
| Dave Jones. http://www.codemonkey.org.uk | SuSE Labs
Quoted 7 lines · Dave Jones
Quoted 5 lines
Yes, I need to have better names for the devices than just "
usb_bus", any suggestions? These devices nodes are really the USB root hubs in the USB controller, so they could just have the USB number as the name like the other USB devices (001), but that's pretty boring :)"
usb_root0" .. "usb_rootN" ?
Hm, that's a good idea, it would match the usbfs bus numbers which should keep people happy.
Quoted 5 lines
btw, a script to marry the busid's from driverfs to lspci/lsusb output may be useful in the future especially if combined somehow with tree(1). Could be very handy when it gets time to debug those "My system won't suspend to disk" "What does /driver look like?" situations.
Ah, a lsdrivers program is needed! :)
thanks,
greg k-h
Quoted 11 lines · Greg KH
Quoted 7 lines · Dave Jones
Quoted 5 lines
Yes, I need to have better names for the devices than just "
usb_bus", any suggestions? These devices nodes are really the USB root hubs in the USB controller, so they could just have the USB number as the name like the other USB devices (001), but that's pretty boring :)"
usb_root0" .. "usb_rootN" ?Hm, that's a good idea, it would match the usbfs bus numbers which should keep people happy.
Would it be usb_rootN or usb_busN?
Quoted 7 lines
Quoted 5 lines
btw, a script to marry the busid's from driverfs to lspci/lsusb output may be useful in the future especially if combined somehow with tree(1). Could be very handy when it gets time to debug those "My system won't suspend to disk" "What does /driver look like?" situations.
Ah, a lsdrivers program is needed! :)
I started writing an 'lsdev' program a while back. If I ever get some free time and feel like playing userspace again, I'll finish it along with a couple of other utilities.
One of the things I fantasized about was making different bus 'personalities' for it. So, you emulate lspci behavior with the same executable. And, extend it to other buses, so you could view all the devices of a particular bus type (and only those devices).
This information will be known by devices as they are registered in the tree, and will be easy to export to userspace. So, one could do lspcmcia, lssbus, and *drum roll* lsisa.
Now for the best part: instead of having to do lsdev -t pci etc, I was just going to hard link ls{$bus type} to lsdev and make it check what argv[0] was to decide how to behave. :)
People hate that idea, and I admit it's twisted. But, there's something kinda cute about it.
-pat
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Yes, I need to have better names for the devices than just "
usb_bus", any suggestions? These devices nodes are really the USB root hubs in the USB controller, so they could just have the USB number as the name like the other USB devices (001), but that's pretty boring :)
Actually one of my criticisms of Greg's patch is that
it hides the actual device tree. The root hub is easily
distinguishable, it's the topmost one in the tree! There should be no need to name it specially.
I'd really rather move away from the model which exposes a USB bus as a flat non-hierarchical setup, and move instead to a model reflects the actual topology of the USB devices and hubs.
Quoted 6 lines
Quoted 4 lines
Quoted 1 line
"
usb_root0" .. "usb_rootN" ?Hm, that's a good idea, it would match the usbfs bus numbers which should keep people happy.
Would it be
usb_rootNorusb_busN?
I'd rather see neither, and have the device names reflect physical topology ... so they could make sense to users.
For example, if you plug a USB device into a particular USB socket it would have a particular name, and that name would show up in diagnostics. So when something goes flakey about the device, the diagnostic will be able to completely identify it. And likewise, when userspace tools need to do something, they should be able to use the same pathname each time, unless the devices got re-cabled ... re-enumerating shouldn't affect those names.
The notion of a "bus number" is bothersome, since it's a function only of the order of driver initialization, and that isn't a "stable" way to identify anything. Re-ordering driver initialization shouldn't change device name.
- Dave
Quoted 10 lines · David Brownell
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Quoted 4 lines
Yes, I need to have better names for the devices than just "
usb_bus", any suggestions? These devices nodes are really the USB root hubs in the USB controller, so they could just have the USB number as the name like the other USB devices (001), but that's pretty boring :)Actually one of my criticisms of Greg's patch is that
it hides the actual device tree. The root hub is easily
distinguishable, it's the topmost one in the tree! There should be no need to name it specially.
Quoted 4 lines
I'd really rather move away from the model which exposes a USB bus as a flat non-hierarchical setup, and move instead to a model reflects the actual topology of the USB devices and hubs.
<good description of what the topology tree should show snipped>
I completly agree. My main concern with these first patches, was getting the driverfs - usb subsystem linkage working properly. I wasn't trying to work on the names yet (you should have seen some of my testing patches, the names there were pretty horrible :)
As we wait for the rest of Pat's patches to be merged into the kernel, we can work on nailing down the proper naming scheme for the USB trees.
However, the root hub name _does_ propose a problem. I feel we have two solutions:
- use the bus number (usb_bus_00x)
Pros:
matches the usbfs naming and directory structure
Cons:
depends on the initialization order of the busses.
- use a generic name like I did (usb_bus)
Pros:
does not depend on the init order, and relies on the location in the entire pci topology tree to show its uniqueness.
Cons:
boring :)
And also remember, the status file in a device's directory also provides a _lot_ of information. We haven't even started to fill up the fields there...
thanks,
greg k-h
"> " == "Greg KH" <[email protected]>
Quoted 1 line
I completly agree.
As I see -- some mailer daemons were adding delays and fomenting confusion!
Quoted 7 lines
However, the root hub name _does_ propose a problem. I feel we have two solutions:
- use the bus number (usb_bus_00x)
Pros:
matches the usbfs naming and directory structure
Cons:
depends on the initialization order of the busses.
At this point, I'd propose that "driverfs" should adopt a policy that naming dependencies on initialization order are a virus that should be obliterated to the degree it's possible ... which in this case is completely! :)
Quoted 7 lines
- use a generic name like I did (
usb_bus) Pros:
does not depend on the init order, and relies on the
location in the entire pci topology tree to show its uniqueness.
Cons:
boring :)
Or a third option is to get rid of the extra level of indirection in the current driverfs naming policy, as Patrick suggested in a separate email. I'll follow up separately.
Quoted 3 lines
And also remember, the status file in a device's directory also provides a _lot_ of information. We haven't even started to fill up the fields there...
And there can be a lot more such files. Though that 4KB limit may become an issue at some point.
What sort of USB information were you thinking should show up? Current configuration and altsetting? Power consumption for hubs (not that we budget that yet :)? There really aren't any examples of this in the kernel yet.
Also, one could argue that each USB function ("interface") should be presented as an individual device, just like each PCI function is handled ... after all, USB drivers bind to interfaces, not devices, and this is the "driver" FS! :)
- Dave
Quoted 6 lines · David Brownell
Quoted 3 lines
And also remember, the status file in a device's directory also provides a _lot_ of information. We haven't even started to fill up the fields there...
And there can be a lot more such files. Though that 4KB limit may become an issue at some point.
I doubt it, we are talking one value per file here. I can't see any USB driver wanting to go over 4Kb for 1 value (and if it does, I'll change it :)
Quoted 4 lines
What sort of USB information were you thinking should show up? Current configuration and altsetting? Power consumption for hubs (not that we budget that yet :)? There really aren't any examples of this in the kernel yet.
Bandwidth is a good one.
Current config and altsetting might be useful. I haven't really thought about the different files yet.
Quoted 4 lines
Also, one could argue that each USB function ("interface") should be presented as an individual device, just like each PCI function is handled ... after all, USB drivers bind to interfaces, not devices, and this is the "driver" FS! :)
No, I'll say that we need to stay one physical device per device in the tree. If you want to do an interface tree, let's put that in usbfs, where it belongs :)
thanks,
greg k-h
Quoted 6 lines
Quoted 2 lines
And there can be a lot more such files. Though that 4KB limit may become an issue at some point.
I doubt it, we are talking one value per file here. I can't see any USB driver wanting to go over 4Kb for 1 value (and if it does, I'll change it :)
I could see descriptors getting that large, particularly if they're turned from binary form into text. I seem to recall Patrick was anticipating troubles with that 4K limit, at some point.
Quoted 7 lines
Quoted 4 lines
Also, one could argue that each USB function ("interface") should be presented as an individual device, just like each PCI function is handled ... after all, USB drivers bind to interfaces, not devices, and this is the "driver" FS! :)
No, I'll say that we need to stay one physical device per device in the tree.
But we aren't that way today. Examples:
- Take a multifunction PCI card ("physical device") and plug it in. Each function shows up as another "logical device" in the tree. Each such logical device gets one driver.
- Take a composite USB device ("physical device", like a keyboard with hub) and plug it in. Each logical device shows up separately. Each such logical device gets one driver.
The issue with USB is that it's got a much more complex configuration
model, not all of which is well supported yet in Linux. There's a type
of device which is handled inconsistently:
- Take a multiple-interface USB device ("physical device") and plug it in. It's presented as one logical device. BUT (!!) such devices need MULTIPLE drivers!! (Example: speaker with built in volume control, needs audio and HID drivers.)
I was observing that we have a chance to make things consistent. In my experience, that's normally a good thing. Similarly, I think USB should handle configurations more as first class entities. Changing a device's config doesn't trigger driver rebinding, for example; we're in luck, so far, that most devices don't have many configurations.
Quoted 2 lines
If you want to do an interface tree, let's put that in usbfs,
where it belongs :)
Ah, but changing usbfs is impractical at this point since lots of userspace programs rely on it not changing. Which is why I was pointing this out in the context of driverfs, which can still be improved in such ways ... "usbdevfs" was always advertised as "preliminary", anyway! :)
- Dave
Quoted 4 lines · David Brownell
Quoted 2 lines
No, I'll say that we need to stay one physical device per device in the tree.
But we aren't that way today. Examples:
<snip>
Ok, you're right. We want to tell the drivers to shut down (remember, the original goal of driverfs was for power management), so all drivers that attach to a device need to be shown.
I'll play with the code some more and make this kind of change.
Quoted 8 lines
Quoted 2 lines
If you want to do an interface tree, let's put that in usbfs,
where it belongs :)Ah, but changing usbfs is impractical at this point since lots of userspace programs rely on it not changing. Which is why I was pointing this out in the context of driverfs, which can still be improved in such ways ... "usbdevfs" was always advertised as "preliminary", anyway! :)
Heh, I took the "preliminary" tag off of it a short while ago, as so many different userspace programs were using it. Maybe usbfs2? :)
Seriously, I've had some ideas of a different way to implement the functionality of usbfs, possibly without all of the ioctl calls, but I have not had the time to experiment with it...
thanks,
greg k-h
Hi Greg,
Not that it matters for this discussion, but I thought the USB2.0
cards had two OHCI and one EHCI controllers?
regards,
Stephen.
Quoted 516 lines · Greg KH
Hi,
Well after determining that the last version of this patch doesn't show the USB tree properly, here's another patch against 2.5.3-pre6 that fixes this issue.
With this patch (the driver core changes were from Pat Mochel, thanks Pat for putting up with my endless questions) my machine now shows the following tree:
[greg@soap x]$ tree
.
`-- root|-- pci0 | |-- 00:00.0 | | |-- power | | `-- status | |-- 00:01.0 | | |-- 02:00.0 | | | |-- power | | | `-- status | | |-- power | | `-- status | |-- 00:1e.0 | | |-- 01:08.0 | | | |-- power | | | `-- status | | |-- 01:0d.0 | | | |-- power | | | |-- status | | | `-- usb_bus | | | |-- power | | | `-- status | | |-- 01:0d.1 | | | |-- power | | | |-- status | | | `-- usb_bus | | | |-- 003 | | | | |-- power | | | | `-- status | | | |-- power | | | `-- status | | |-- 01:0d.2 | | | |-- power | | | |-- status | | | `-- usb_bus | | | |-- power | | | `-- status | | |-- power | | `-- status | |-- 00:1f.0 | | |-- power | | `-- status | |-- 00:1f.1 | | |-- power | | `-- status | |-- 00:1f.2 | | |-- power | | |-- status | | `-- usb_bus | | |-- 002 | | | |-- 003 | | | | |-- power | | | | `-- status | | | |-- power | | | `-- status | | |-- power | | `-- status | |-- 00:1f.3 | | |-- power | | `-- status | |-- 00:1f.5 | | |-- power | | `-- status | |-- power | `-- status |-- power `-- statusI have 2 USB controllers in this system, 1 UHCI controller on the motherboard that shows up under PCI device 00:1f.2. There are two devices plugged into this bus, a hub, and a trackball plugged into the hub. This topology is now shown properly.
The other controller is a USB 2.0 controller that shows up in 3 places on the pci bus: 01:0d.0, 01:0d.1, and 01:0d.2. The 01:0d.1 device is controlled by the ohci-hcd driver, and a USB hub is plugged into it. The ehci-hcd driver controls the other 2 pci devices on the card.
Yes, I need to have better names for the devices than just "
usb_bus", any suggestions? These devices nodes are really the USB root hubs in the USB controller, so they could just have the USB number as the name like the other USB devices (001), but that's pretty boring :)greg k-h
diff -Nru
a/drivers/base/core.cb/drivers/base/core.cdrivers/base/core.c +23
-78@@ -17,10 +17,7 @@ # define DBG(x...)#endif -static struct iobus device_root = { - bus_id:"root" , - name:"Logical System Root" , -}; +static struct device * device_root;int (*platform_notify)(struct device * dev) =NULL ;int (*platform_notify_remove)(struct device * dev) =NULL ;@@ -28,9 +25,6 @@extern int device_make_dir (struct device * dev);extern void device_remove_dir (struct device * dev); -extern int iobus_make_dir (struct iobus * iobus); -extern void iobus_remove_dir (struct iobus * iobus); -static spinlock_t device_lock;/** @@ -47,19 +41,25 @@if (!dev || !strlen (dev->bus_id))return -EINVAL ; -BUG_ON (!dev->parent);spin_lock (&device_lock);INIT_LIST_HEAD (&dev->node); +INIT_LIST_HEAD (&dev->children);spin_lock_init (&dev->lock);atomic_set (&dev->refcount,2 ); -get_iobus (dev->parent); -list_add_tail (&dev->node,&dev->parent->devices); +if (dev != device_root) { +struct device * parent; +if (!dev->parent) + dev->parent = device_root; + parent = dev->parent; +get_device (parent); +list_add_tail (&dev->node,&parent->children); + }spin_unlock (&device_lock); -DBG ("DEV: registering device: ID = '%s', name = %s, parent = %s\n" , - dev->bus_id, dev->name, parent->bus_id); +DBG ("DEV: registering device: ID = '%s', name = %s\n" , + dev->bus_id, dev->name);if ((error =device_make_dir (dev))) goto register_done;@@ -70,8 +70,8 @@ register_done:put_device (dev); -if (error) -put_iobus (dev->parent); +if (error && dev->parent) +put_device (dev->parent);return error; }@@ -100,9 +100,6 @@/* remove the driverfs directory */ device_remove_dir (dev); -if (dev->subordinate) -iobus_remove_dir (dev->subordinate); -/* Notify the platform of the removal, in case they * need to do anything... */ @@ -116,70 +113,18 @@if (dev->driver && dev->driver->remove) dev->driver->remove (dev,REMOVE_FREE_RESOURCES ); -put_iobus (dev->parent); -} - -int iobus_register (struct iobus *bus) -{ -int error; - -if (!bus || !strlen (bus->bus_id)) -return -EINVAL ; - -spin_lock (&device_lock); -atomic_set (&bus->refcount,2 ); -spin_lock_init (&bus->lock); -INIT_LIST_HEAD (&bus->node); -INIT_LIST_HEAD (&bus->devices); -INIT_LIST_HEAD (&bus->children); - -if (bus != &device_root) { -if (!bus->parent) - bus->parent = &device_root; -get_iobus (bus->parent); -list_add_tail (&bus->node,&bus->parent->children); - } -spin_unlock (&device_lock); - -DBG ("DEV: registering bus. ID = '%s' name = '%s' parent = %p\n" , - bus->bus_id,bus->name,bus->parent); - - error =iobus_make_dir (bus); - -put_iobus (bus); -if (error && bus->parent) -put_iobus (bus->parent); -return error; -} - -/** -* iobus_unregister - remove bus and children from device tree -* @bus: pointer to bus structure -* -* Remove device from parent's list of children and decrement -* reference count on controlling device. That should take care of -* the rest of the cleanup. -*/ -void put_iobus (struct iobus * iobus) -{ -if (!atomic_dec_and_lock (&iobus->refcount,&device_lock)) -return ; -list_del_init (&iobus->node); -spin_unlock (&device_lock); - -if (!list_empty (&iobus->devices) || - !list_empty (&iobus->children)) -BUG (); - -put_iobus (iobus->parent); -/* unregister itself */ -put_device (iobus->self); +put_device (dev->parent); }static int __init device_init_root (void ) { -/* initialize parent bus lists */ -return iobus_register (&device_root); + device_root =kmalloc (sizeof (*device_root),GFP_KERNEL ); +if (!device_root) +return -ENOMEM ; +memset (device_root,0 ,sizeof (*device_root)); +strcpy (device_root->bus_id,"root" ); +strcpy (device_root->name,"System Root" ); +return device_register (device_root); }int __init device_driver_init (void )@@ -208,5 +153,5 @@ }EXPORT_SYMBOL (device_register); -EXPORT_SYMBOL (iobus_register); +EXPORT_SYMBOL (put_device);EXPORT_SYMBOL (device_driver_init);diff -Nru
a/drivers/base/fs.cb/drivers/base/fs.cdrivers/base/fs.c +0
-22@@ -102,28 +102,6 @@ } }return 0 ; -} - -void iobus_remove_dir (struct iobus * iobus) -{ -if (iobus) -driverfs_remove_dir (&iobus->dir); -} - -int iobus_make_dir (struct iobus * iobus) -{ -struct driver_dir_entry * parent =NULL ; -int error; - -INIT_LIST_HEAD (&iobus->dir.files); - iobus->dir.mode = (S_IFDIR |S_IRWXU |S_IRUGO |S_IXUGO ); - iobus->dir.name = iobus->bus_id; - -if (iobus->parent) - parent = &iobus->parent->dir; - - error =driverfs_create_dir (&iobus->dir,parent); -return error; }EXPORT_SYMBOL (device_create_file);diff -Nru
a/drivers/pci/pci.cb/drivers/pci/pci.cdrivers/pci/pci.c +7
-16@@ -1086,13 +1086,7 @@ child->parent = parent; child->ops = parent->ops; child->sysdata = parent->sysdata; - -/* init generic fields */ - child->iobus.self = &dev->dev; - child->iobus.parent = &parent->iobus; - dev->dev.subordinate = &child->iobus; - -strcpy (child->iobus.name,dev->dev.name); + child->dev = &dev->dev;/* * Set up the primary, secondary and subordinate @@ -1361,16 +1355,11 @@DBG ("Scanning bus %02x\n" , bus->number); max = bus->secondary; -/* we should know for sure what the bus number is, so set the bus ID -* for the bus and make sure it's registered in the device tree */ -sprintf (bus->iobus.bus_id,"pci%d" ,bus->number); -iobus_register (&bus->iobus); -/* Create a device template */ memset (&dev0,0 ,sizeof (dev0)); dev0.bus = bus; dev0.sysdata = bus->sysdata; - dev0.dev.parent = &bus->iobus; + dev0.dev.parent = bus->dev; dev0.dev.driver = &pci_device_driver;/* Go find them, Rover! */ @@ -1430,9 +1419,11 @@return NULL ;list_add_tail (&b->node, &pci_root_buses); -sprintf (b->iobus.bus_id,"pci%d" ,bus); -strcpy (b->iobus.name,"Host/PCI Bridge" ); -iobus_register (&b->iobus); + b->dev =kmalloc (sizeof (*(b->dev)),GFP_KERNEL ); +memset (b->dev,0 ,sizeof (*(b->dev))); +sprintf (b->dev->bus_id,"pci%d" ,bus); +strcpy (b->dev->name,"Host/PCI Bridge" ); +device_register (b->dev); b->number = b->secondary = bus; b->resource[0 ] = &ioport_resource;diff -Nru
a/drivers/usb/hcd/ehci-hcd.cb/drivers/usb/hcd/ehci-hcd.cdrivers/usb/hcd/ehci-hcd.c +6
-0@@ -290,6 +290,12 @@ goto done2; } +/* hook up the root hub to the pci controller in the device tree */ + udev->dev.parent = &ehci->hcd.pdev->dev; +strcpy (udev->dev.name,"usb_name" ); +strcpy (udev->dev.bus_id,"usb_bus" ); +device_register (&udev->dev); +return 0 ; }diff -Nru
a/drivers/usb/hcd/ohci-hcd.cb/drivers/usb/hcd/ohci-hcd.cdrivers/usb/hcd/ohci-hcd.c +7
-1@@ -475,7 +475,13 @@// FIXME cleanup return -ENODEV ; } - + +/* hook up the root hub to the pci controller in the device tree */ + udev->dev.parent = &ohci->hcd.pdev->dev; +strcpy (udev->dev.name,"usb_name" ); +strcpy (udev->dev.bus_id,"usb_bus" ); +device_register (&udev->dev); +return 0 ; }diff -Nru
a/drivers/usb/hub.cb/drivers/usb/hub.cdrivers/usb/hub.c +9
-1@@ -722,8 +722,16 @@ dev->bus->busnum, dev->devpath, dev->devnum);/* Run it through the hoops (find a driver, etc) */ -if (!usb_new_device (dev)) +if (!usb_new_device (dev)) { +/* put the device in the global device tree */ + dev->dev.parent = &dev->parent->dev; +sprintf (&dev->dev.name[0 ],"USB device %04x:%04x" , + dev->descriptor.idVendor, + dev->descriptor.idProduct); +sprintf (&dev->dev.bus_id[0 ],"%03d" , dev->devnum); +device_register (&dev->dev); goto done; + }/* Free the configuration if there was an error */ usb_free_dev (dev);diff -Nru
a/drivers/usb/uhci.cb/drivers/usb/uhci.cdrivers/usb/uhci.c +6
-0@@ -2848,6 +2848,12 @@ goto err_start_root_hub; } +/* hook the root hub up to the pci controller in the device tree */ + uhci->rh.dev->dev.parent = &dev->dev; +strcpy (uhci->rh.dev->dev.name,"usb_name" ); +strcpy (uhci->rh.dev->dev.bus_id,"usb_bus" ); +device_register (&uhci->rh.dev->dev); +return 0 ;/* diff -Nru
a/drivers/usb/usb-ohci.cb/drivers/usb/usb-ohci.cdrivers/usb/usb-ohci.c +7
-0@@ -2508,6 +2508,13 @@#ifdef DEBUG ohci_dump (ohci,1 );#endif + +/* hook the root hub up to the pci controller in the device tree */ + ohci->bus->root_hub->dev.parent = &dev->dev; +strcpy (ohci->bus->root_hub->dev.name,"usb_name" ); +strcpy (ohci->bus->root_hub->dev.bus_id,"usb_bus" ); +device_register (&ohci->bus->root_hub->dev); +return 0 ; }diff -Nru
a/drivers/usb/usb-uhci.cb/drivers/usb/usb-uhci.cdrivers/usb/usb-uhci.c +6
-0@@ -3028,6 +3028,12 @@pci_set_drvdata (dev, s); devs=s; +/* hook the root hub up to the pci controller in the device tree */ + s->bus->root_hub->dev.parent = &dev->dev; +strcpy (s->bus->root_hub->dev.name,"usb_name" ); +strcpy (s->bus->root_hub->dev.bus_id,"usb_bus" ); +device_register (&s->bus->root_hub->dev); +return 0 ; }diff -Nru
a/drivers/usb/usb.cb/drivers/usb/usb.cdrivers/usb/usb.c +1
-0@@ -1925,6 +1951,7 @@if (dev->devnum >0 ) {clear_bit (dev->devnum, &dev->bus->devmap.devicemap);usbfs_remove_device (dev); +put_device (&dev->dev); }/* Free up the device itself */ diff -Nru
a/include/linux/device.hb/include/linux/device.hinclude/linux/device.h +2
-4@@ -66,10 +66,8 @@struct device {struct list_head node;/* node in sibling list */ -struct iobus *parent;/* parent bus */ - -struct iobus *subordinate;/* only valid if this device is a -bridge device */ +struct list_head children; +struct device *parent;char name[DEVICE_NAME_SIZE ];/* descriptive ascii string */ char bus_id[BUS_ID_SIZE ];/* position on parent bus */ diff -Nru
a/include/linux/pci.hb/include/linux/pci.hinclude/linux/pci.h +1
-2@@ -432,8 +432,7 @@unsigned char productver;/* product version */ unsigned char checksum;/* if zero - checksum passed */ unsigned char pad1; - -struct iobus iobus;/* Generic device interface */ +struct device * dev; };#define pci_bus_b (n) list_entry(n, struct pci_bus, node)diff -Nru
a/include/linux/usb.hb/include/linux/usb.hinclude/linux/usb.h +4
-0@@ -1,6 +1,8 @@#ifndef __LINUX_USB_H#define __LINUX_USB_H +#include <linux/device.h> +/* USB constants */ /* @@ -1037,6 +1042,8 @@struct usb_device *parent;struct usb_bus *bus;/* Bus we're part of */ + +struct device dev;/* Generic device interface */ struct usb_device_descriptor descriptor;/* Descriptor */ struct usb_config_descriptor *config;/* All of the configs */ _______________________________________________
[email protected]
To unsubscribe, use the last form field at:
https://lists.sourceforge.net/lists/listinfo/linux-usb-devel
Signature · 6 lines
/------------------------------------+-------------------------\ |Stephen J. Gowdy | SLAC, MailStop 17, | |http://www.slac.stanford.edu/~gowdy/ | 2575 Sand Hill Road, | | | Menlo Park CA 94025, USA | |EMail: [email protected] | Tel: +1 650 926 3144 | \------------------------------------+-------------------------/
Quoted 3 lines · Stephen J. Gowdy
Hi Greg,
Not that it matters for this discussion, but I thought the USB2.0
cards had two OHCI and one EHCI controllers?
I think you're right.
greg k-h
Quoted 9 lines · Greg KH
Hi,
Well after determining that the last version of this patch doesn't show the USB tree properly, here's another patch against 2.5.3-pre6 that fixes this issue.
With this patch (the driver core changes were from Pat Mochel, thanks Pat for putting up with my endless questions) my machine now shows the following tree:
This is great! Thanks to Greg for testing and providing feedback.
One question:
Quoted 9 lines
| | |-- 01:0d.1
| | | |-- power
| | | |-- status
| | | `-- usb_bus
| | | |-- 003
| | | | |-- power
| | | | `-- status
| | | |-- power
| | | `-- statusYou have a PCI device that is the USB controller. You create a child of that represents the USB bus. Then, devices are added as children of that.
Logically, couldn't you skip that extra layer of indirection and make USB devices children of the USB controller? Or, do you see benefit in the explicit distinction?
include/linux/device.h line 70
Quoted 13 lines
diff -Nru
a/include/linux/device.hb/include/linux/device.hinclude/linux/device.h +2
-4@@ -66,10 +66,8 @@struct device {struct list_head node;/* node in sibling list */ -struct iobus *parent;/* parent bus */ - -struct iobus *subordinate;/* only valid if this device is a -bridge device */ +struct list_head children; +struct device *parent;
(To everyone else)
Note this change: This patch removed struct iobus, so that everything becomes a device and only a device. A 'bus' is simply a device that has children.
This concept is something that I have argued both for and against since I started on this. Initially, the goal was to separate the two because they followed such different semantics.
But, I've found it, IMO, it creates more problems than it solves. There was/is a lot of code replication just because you want to do the same thing to each object, but have two different pointer types to deal with.
And, it's quite easy to add extra semantics for devices that have children.
-pat
"> " == "Patrick Mochel" <[email protected]>
Quoted 6 lines
You have a PCI device that is the USB controller. You create a child of that represents the USB bus. Then, devices are added as children of that.
Logically, couldn't you skip that extra layer of indirection and make USB devices children of the USB controller? Or, do you see benefit in the explicit distinction?
Since I don't see a benefit from that extra indirection, I was going to ask almost that same question ... :)
But it's broader than that: Why shouldn't that apply to _every_ kind of bridge, not just USB controllers ("PCI-to-USB bridges")?
For example, with PCI why should there ever be "pci0" directories, with children "00:??.?" and "pci1"?
Quoted 7 lines
... A 'bus' is simply a device that has children.
This concept is something that I have argued both for and against since I started on this. Initially, the goal was to separate the two because they followed such different semantics.
But, I've found it, IMO, it creates more problems than it solves. ...
If the model that driverfs exposes is that directories denote devices (of any type including bridge) and (text) files expose device properties, then my initial suspicion is that the additional level of indirection would not be necessary. Maybe some "type" property would be desirable though, to better separate namespace structure from typing issues. (It might need to be provided by the device/bridge driver.)
When I've seen corresponding problems in other areas, the extra indirection has not been necessary. The simpler naming policies have been a lot easier to work with.
I suspect it'd be better to simplify the naming policy for bridges and busses everywhere in driverfs, not just for USB, but I'd like to know any arguments for keeping that extra level of indirection.
- Dave
Quoted 11 lines · David Brownell
"> " == "Patrick Mochel" <[email protected]>
Quoted 6 lines
You have a PCI device that is the USB controller. You create a child of that represents the USB bus. Then, devices are added as children of that.
Logically, couldn't you skip that extra layer of indirection and make USB devices children of the USB controller? Or, do you see benefit in the explicit distinction?
Since I don't see a benefit from that extra indirection, I was going to ask almost that same question ... :)
But that device _is_ a USB device, it's a root hub. It has bandwidth allocation, strings, a configuration, and other stuff. It operates a bit differently from other hubs, but it's pretty close to the real thing.
Quoted 5 lines
But it's broader than that: Why shouldn't that apply to _every_ kind of bridge, not just USB controllers ("PCI-to-USB bridges")?
For example, with PCI why should there ever be "pci0" directories, with children "00:??.?" and "pci1"?
It's information that is useful to the user. If presented with a tree that doesn't have the pci? and usb directories, it just looks like a random tree of different numbers. If we did that, we would really need a lsdevice program just to determine what the different devices are easily :)
I want to be able to easily see where the pci root buses, usb root buses, ieee1394 root buses, etc. are in the system, by just looking at the tree.
I'll play around more with the naming scheme next week and post the results.
thanks,
greg k-h
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
You have a PCI device that is the USB controller. You create a child of that represents the USB bus. Then, devices are added as children of that.
Logically, couldn't you skip that extra layer of indirection and make USB devices children of the USB controller? Or, do you see benefit in the explicit distinction?
Quoted 7 lines
Quoted 2 lines
Since I don't see a benefit from that extra indirection, I was going to ask almost that same question ... :)
But that device _is_ a USB device, it's a root hub. It has bandwidth allocation, strings, a configuration, and other stuff. It operates a bit differently from other hubs, but it's pretty close to the real thing.
Sure it's a hub (with the firmware provided by Linux :) but it's also a PCI device. The extra indirection is the one which makes the root hub be a different device than the PCI device. It'll have some attributes that relate to its PCI device role, others that relate to the more specialized "USB root hub" role.
Quoted 7 lines
Quoted 5 lines
But it's broader than that: Why shouldn't that apply to _every_ kind of bridge, not just USB controllers ("PCI-to-USB bridges")?
For example, with PCI why should there ever be "pci0" directories, with children "00:??.?" and "pci1"?
It's information that is useful to the user.
Not useful to this user ... and I'm unclear why it'd be useful to any others. Initialization order can change, and embedding it in device naming is basically just trouble.
Quoted 3 lines
If presented with a tree
that doesn't have the pci? and usb directories, it just looks like a random tree of different numbers.
Just like most directories. That's why I commented that it might be desirable to have any typing information just show up as one of the directory attributes.
- Dave
Hi!
drivers/base/core.c line 128
Quoted 12 lines
static int
__initdevice_init_root(void)
{drivers/base/core.c +7
-2-/* initialize parent bus lists */ -return iobus_register (&device_root); + device_root =kmalloc (sizeof (*device_root),GFP_KERNEL ); +if (!device_root) +return -ENOMEM ; +memset (device_root,0 ,sizeof (*device_root)); +strcpy (device_root->bus_id,"root" ); +strcpy (device_root->name,"System Root" ); +return device_register (device_root); }
Why don't you leave device_root allocated statically?
drivers/pci/pci.c line 1422
Quoted 8 lines
drivers/pci/pci.c +1
-3@@ -1430,9 +1419,11 @@return NULL ;list_add_tail (&b->node, &pci_root_buses); -sprintf (b->iobus.bus_id,"pci%d" ,bus); -strcpy (b->iobus.name,"Host/PCI Bridge" ); -iobus_register (&b->iobus); + b->dev =kmalloc (sizeof (*(b->dev)),GFP_KERNEL );
Uff... ~~~~~~~~~ would not "struct device" (or
what should it be) look better?
drivers/pci/pci.c line 1423
Quoted 1 line
drivers/pci/pci.c +1
-0+memset (b->dev,0 ,sizeof (*(b->dev)));
Pavel
Signature · 2 lines
Philips Velo 1: 1"x4"x8", 300gram, 60, 12MB, 40bogomips, linux, mutt, details at http://atrey.karlin.mff.cuni.cz/~pavel/velo/index.html.
Pavel Machek <[email protected]> said:
[...]
Quoted 4 lines
drivers/pci/pci.c line 1422
Quoted 2 lines
drivers/pci/pci.c +1
-1-iobus_register (&b->iobus); + b->dev =kmalloc (sizeof (*(b->dev)),GFP_KERNEL );Uff... ~~~~~~~~~ would not "struct device" (or
what should it be) look better?
Yep, but there is _no_ chance to screw up ("or what should it be" ;-) this way.
Signature · 1 line
Horst von Brand http://counter.li.org # 22616