Hi Alan,
On Thu, Jan 29, 2015 at 5:56 PM, Alan Stern <[email protected]> wrote:
> On Thu, 29 Jan 2015, Ruslan Bilovol wrote:
>
>> Change behavior during registration of gadgets and
>> gadget drivers in udc-core. Instead of previous
>> approach when for successful probe of usb gadget driver
>> at least one usb gadget should be already registered
>> use another one where gadget drivers and gadgets
>> can be registered in udc-core independently.
>>
>> Independent registration of gadgets and gadget drivers
>> is useful for built-in into kernel gadget and gadget
>> driver case - because it's possible that gadget is
>> really probed only on late_init stage (due to deferred
>> probe) whereas gadget driver's probe is silently failed
>> on module_init stage due to no any UDC added.
>>
>> Also it is useful for modules case - now there is no
>> difference what module to insert first: gadget module
>> or gadget driver one.
>
> It's possible to do all this much more simply. In fact, I posted a
> patch some time ago to do exactly this (but I can't find a copy of it
> anywhere).
Unfortunately I didn't find your patch.
>
>> Signed-off-by: Ruslan Bilovol <[email protected]>
>> ---
>> drivers/usb/gadget/udc/udc-core.c | 113 +++++++++++++++++++++++++++++++++++---
>> 1 file changed, 105 insertions(+), 8 deletions(-)
>>
>> diff --git a/drivers/usb/gadget/udc/udc-core.c b/drivers/usb/gadget/udc/udc-core.c
>> index e31d574..4c9412b 100644
>> --- a/drivers/usb/gadget/udc/udc-core.c
>> +++ b/drivers/usb/gadget/udc/udc-core.c
>> @@ -43,13 +43,23 @@ struct usb_udc {
>> struct usb_gadget_driver *driver;
>> struct usb_gadget *gadget;
>> struct device dev;
>> + bool bind_by_name;
>> + struct list_head list;
>> +};
>> +
>> +struct pending_gadget_driver {
>> + struct usb_gadget_driver *driver;
>> + char *udc_name;
>> struct list_head list;
>> };
>
> You don't need all this stuff. What's the point of keeping track of
> names? If there are any unbound gadget drivers pending, a newly
> registered UDC should bind to the first one available.
It's because gadget driver may be bound to usb_gadget in two ways:
- standard way - in this case any available udc will be picked up
- by name of udc, in this case only matching udc will be picked up
Main feature of my path is not only deferred binding of gadget driver,
but also possibility to register/unregister udc at any time.
This is useful for user who can load, for example, udc module
if needed and unload it safely, not touching gadget driver.
Another example is USB device controllers that consist of pair of
HS+SS controllers, each one having its own udc driver. In this case
it's possible to switch SS/HS by registering/unregistering corresponding
udc and not touching gadget driver.
I did all of this inside of udc-core because it looks like to be best place for
udc <-> gadget driver housekeeping. Also it is verified on lot of combinations
of udc and gadget drivers that can be built-in or build as modules
Best regards,
Ruslan
>
> Just add a "pending" list_head into the usb_gadget_driver structure and
> forget about all the rest. (Or try to find my patch in the mailing
> list archives somehow see if you think it needs to be changed.)
>
> Alan Stern
>
On Sun, Feb 08, 2015 at 09:04:32PM +0200, Ruslan Bilovol wrote:
> Hi Alan,
>
> On Thu, Jan 29, 2015 at 5:56 PM, Alan Stern <[email protected]> wrote:
> > On Thu, 29 Jan 2015, Ruslan Bilovol wrote:
> >
> >> Change behavior during registration of gadgets and
> >> gadget drivers in udc-core. Instead of previous
> >> approach when for successful probe of usb gadget driver
> >> at least one usb gadget should be already registered
> >> use another one where gadget drivers and gadgets
> >> can be registered in udc-core independently.
> >>
> >> Independent registration of gadgets and gadget drivers
> >> is useful for built-in into kernel gadget and gadget
> >> driver case - because it's possible that gadget is
> >> really probed only on late_init stage (due to deferred
> >> probe) whereas gadget driver's probe is silently failed
> >> on module_init stage due to no any UDC added.
> >>
> >> Also it is useful for modules case - now there is no
> >> difference what module to insert first: gadget module
> >> or gadget driver one.
> >
> > It's possible to do all this much more simply. In fact, I posted a
> > patch some time ago to do exactly this (but I can't find a copy of it
> > anywhere).
>
> Unfortunately I didn't find your patch.
>
> >
> >> Signed-off-by: Ruslan Bilovol <[email protected]>
> >> ---
> >> drivers/usb/gadget/udc/udc-core.c | 113 +++++++++++++++++++++++++++++++++++---
> >> 1 file changed, 105 insertions(+), 8 deletions(-)
> >>
> >> diff --git a/drivers/usb/gadget/udc/udc-core.c b/drivers/usb/gadget/udc/udc-core.c
> >> index e31d574..4c9412b 100644
> >> --- a/drivers/usb/gadget/udc/udc-core.c
> >> +++ b/drivers/usb/gadget/udc/udc-core.c
> >> @@ -43,13 +43,23 @@ struct usb_udc {
> >> struct usb_gadget_driver *driver;
> >> struct usb_gadget *gadget;
> >> struct device dev;
> >> + bool bind_by_name;
> >> + struct list_head list;
> >> +};
> >> +
> >> +struct pending_gadget_driver {
> >> + struct usb_gadget_driver *driver;
> >> + char *udc_name;
> >> struct list_head list;
> >> };
> >
> > You don't need all this stuff. What's the point of keeping track of
> > names? If there are any unbound gadget drivers pending, a newly
> > registered UDC should bind to the first one available.
>
> It's because gadget driver may be bound to usb_gadget in two ways:
> - standard way - in this case any available udc will be picked up
> - by name of udc, in this case only matching udc will be picked up
>
> Main feature of my path is not only deferred binding of gadget driver,
> but also possibility to register/unregister udc at any time.
> This is useful for user who can load, for example, udc module
> if needed and unload it safely, not touching gadget driver.
> Another example is USB device controllers that consist of pair of
> HS+SS controllers, each one having its own udc driver. In this case
> it's possible to switch SS/HS by registering/unregistering corresponding
> udc and not touching gadget driver.
>
> I did all of this inside of udc-core because it looks like to be best place for
> udc <-> gadget driver housekeeping. Also it is verified on lot of combinations
> of udc and gadget drivers that can be built-in or build as modules
>
In fact, both I and Robert Baldyga posted patches to try fix this
problem.
http://marc.info/?l=linux-usb&m=139287784610046&w=2
http://lwn.net/Articles/601839/
I tried to use Robert's solution (fix some bugs) in internal tree, but
the mass storage gadget still has problems to work if unload udc first.
The possible reason should be: it has two places to call usb_composite_unregister.
--
Best Regards,
Peter Chen
On Sun, 8 Feb 2015, Ruslan Bilovol wrote:
> Hi Alan,
>
> On Thu, Jan 29, 2015 at 5:56 PM, Alan Stern <[email protected]> wrote:
> > On Thu, 29 Jan 2015, Ruslan Bilovol wrote:
> >
> >> Change behavior during registration of gadgets and
> >> gadget drivers in udc-core. Instead of previous
> >> approach when for successful probe of usb gadget driver
> >> at least one usb gadget should be already registered
> >> use another one where gadget drivers and gadgets
> >> can be registered in udc-core independently.
> >>
> >> Independent registration of gadgets and gadget drivers
> >> is useful for built-in into kernel gadget and gadget
> >> driver case - because it's possible that gadget is
> >> really probed only on late_init stage (due to deferred
> >> probe) whereas gadget driver's probe is silently failed
> >> on module_init stage due to no any UDC added.
> >>
> >> Also it is useful for modules case - now there is no
> >> difference what module to insert first: gadget module
> >> or gadget driver one.
> >
> > It's possible to do all this much more simply. In fact, I posted a
> > patch some time ago to do exactly this (but I can't find a copy of it
> > anywhere).
>
> Unfortunately I didn't find your patch.
> > You don't need all this stuff. What's the point of keeping track of
> > names? If there are any unbound gadget drivers pending, a newly
> > registered UDC should bind to the first one available.
>
> It's because gadget driver may be bound to usb_gadget in two ways:
> - standard way - in this case any available udc will be picked up
> - by name of udc, in this case only matching udc will be picked up
Where did this "by name" feature come from? You did not mention it in
the patch description.
Why bother matching by name? Why not simply take the first available
UDC?
> Main feature of my path is not only deferred binding of gadget driver,
> but also possibility to register/unregister udc at any time.
> This is useful for user who can load, for example, udc module
> if needed and unload it safely, not touching gadget driver.
We can already do that with the existing code. There's no need for a
patch.
Also, it's not clear that the existing gadget drivers will work
properly if they are unbound from one UDC and then bound again to
another one. They were not written with that sort of thing in mind.
On Mon, 9 Feb 2015, Peter Chen wrote:
> In fact, both I and Robert Baldyga posted patches to try fix this
> problem.
>
> http://marc.info/?l=linux-usb&m=139287784610046&w=2
> http://lwn.net/Articles/601839/
That's right. The patch I wrote was a lot like Robert's patch (the
marc.info URL above). His approach can be simplified a little; the
"attached" field isn't needed if the driver_list holds only unbound
gadget drivers.
> I tried to use Robert's solution (fix some bugs) in internal tree, but
> the mass storage gadget still has problems to work if unload udc first.
> The possible reason should be: it has two places to call
> usb_composite_unregister.
The mass-storage gadget driver can be fixed, if necessary.
Alan Stern
Hi,
(... snip ...)
>
> > > You don't need all this stuff. What's the point of keeping
> track of
> > > names? If there are any unbound gadget drivers pending, a
> newly
> > > registered UDC should bind to the first one available.
> >
> > It's because gadget driver may be bound to usb_gadget in two
> ways:
> > - standard way - in this case any available udc will be picked
> up
> > - by name of udc, in this case only matching udc will be picked
> up
>
> Where did this "by name" feature come from? You did not mention it
> in
> the patch description.
>
> Why bother matching by name? Why not simply take the first
> available
> UDC?
Because you may have more than one udc. This would allow to pick one by
name just like using configfs interface.
>
> > Main feature of my path is not only deferred binding of gadget
> driver,
> > but also possibility to register/unregister udc at any time.
> > This is useful for user who can load, for example, udc module
> > if needed and unload it safely, not touching gadget driver.
>
> We can already do that with the existing code. There's no need for
> a
> patch.
>
> Also, it's not clear that the existing gadget drivers will work
> properly if they are unbound from one UDC and then bound again to
> another one. They were not written with that sort of thing in
> mind.
>
What you have described is one of basics configfs features.
You should be able to bind and unbind your gadget whenever you want
and it should work properly after doing:
## create gadget
$ echo "udc.0" > UDC
$ echo "" > UDC
$ echo "udc.1" > UDC
Function shouldn't care which udc it has been bound previously.
Only current one is important and on each unbind each function
should cleanup its state and prepare to be bound to another udc.
Configfs interface doesn't prohibit this and I haven't seen any
info about such restriction. If some function is not working in
such situation there is a bug in that function and it should be fixed.
I have tried to test this on my odroid with dwc2 and dummy_hcd.
Most of functions seems to be working but for example ecm isn't.
After some digging with Robert we found that it's always reusing
endpoints received in first bind(). Once again in my opinion it's
a bug which should be fixed and not treated as general assumption.
--
Krzysztof Opasiak
Samsung R&D Institute Poland
Samsung Electronics
[email protected]
> -----Original Message-----
> From: [email protected] [mailto:linux-usb-
> [email protected]] On Behalf Of Krzysztof Opasiak
> Sent: Monday, February 09, 2015 7:06 PM
> To: 'Alan Stern'; 'Ruslan Bilovol'; 'Peter Chen'
> Cc: [email protected]; [email protected];
> 'Balbi, Felipe'; [email protected]; Andrzej Pietrasiewicz
> Subject: RE: [PATCH 1/2] usb: gadget: udc-core: independent
> registration of gadgets and gadget drivers
>
> Hi,
>
> (... snip ...)
>
> >
> > > > You don't need all this stuff. What's the point of keeping
> > track of
> > > > names? If there are any unbound gadget drivers pending, a
> > newly
> > > > registered UDC should bind to the first one available.
> > >
> > > It's because gadget driver may be bound to usb_gadget in two
> > ways:
> > > - standard way - in this case any available udc will be picked
> > up
> > > - by name of udc, in this case only matching udc will be
> picked
> > up
> >
> > Where did this "by name" feature come from? You did not mention
> it
> > in
> > the patch description.
> >
> > Why bother matching by name? Why not simply take the first
> > available
> > UDC?
>
> Because you may have more than one udc. This would allow to pick
> one by
> name just like using configfs interface.
>
> >
> > > Main feature of my path is not only deferred binding of gadget
> > driver,
> > > but also possibility to register/unregister udc at any time.
> > > This is useful for user who can load, for example, udc module
> > > if needed and unload it safely, not touching gadget driver.
> >
> > We can already do that with the existing code. There's no need
> for
> > a
> > patch.
> >
> > Also, it's not clear that the existing gadget drivers will work
> > properly if they are unbound from one UDC and then bound again to
> > another one. They were not written with that sort of thing in
> > mind.
> >
>
> What you have described is one of basics configfs features.
> You should be able to bind and unbind your gadget whenever you want
> and it should work properly after doing:
>
> ## create gadget
> $ echo "udc.0" > UDC
> $ echo "" > UDC
> $ echo "udc.1" > UDC
>
> Function shouldn't care which udc it has been bound previously.
> Only current one is important and on each unbind each function
> should cleanup its state and prepare to be bound to another udc.
> Configfs interface doesn't prohibit this and I haven't seen any
> info about such restriction. If some function is not working in
> such situation there is a bug in that function and it should be
> fixed.
>
> I have tried to test this on my odroid with dwc2 and dummy_hcd.
> Most of functions seems to be working but for example ecm isn't.
^ above is ok
> After some digging with Robert we found that it's always reusing
> endpoints received in first bind().
Fixup:
That's bullshit ignore it please. ecm_opts->bound is not used to take
endpoints but only to register net device. Went too far after short
reading.
All in all, ecm is not working when binding form one udc to another.
Don't know exact reason, but in my opinion it's more a bug than common
assumption.
--
Krzysztof Opasiak
Samsung R&D Institute Poland
Samsung Electronics
[email protected]
On Mon, 9 Feb 2015, Krzysztof Opasiak wrote:
> > Why bother matching by name? Why not simply take the first
> > available
> > UDC?
>
> Because you may have more than one udc. This would allow to pick one by
> name just like using configfs interface.
Clearly it would be more flexible to allow the user to do the matching,
the way configfs does (sysfs too).
> > > Main feature of my path is not only deferred binding of gadget
> > driver,
> > > but also possibility to register/unregister udc at any time.
> > > This is useful for user who can load, for example, udc module
> > > if needed and unload it safely, not touching gadget driver.
> >
> > We can already do that with the existing code. There's no need for
> > a
> > patch.
> >
> > Also, it's not clear that the existing gadget drivers will work
> > properly if they are unbound from one UDC and then bound again to
> > another one. They were not written with that sort of thing in
> > mind.
> >
>
> What you have described is one of basics configfs features.
> You should be able to bind and unbind your gadget whenever you want
> and it should work properly after doing:
>
> ## create gadget
> $ echo "udc.0" > UDC
> $ echo "" > UDC
> $ echo "udc.1" > UDC
>
> Function shouldn't care which udc it has been bound previously.
> Only current one is important and on each unbind each function
> should cleanup its state and prepare to be bound to another udc.
> Configfs interface doesn't prohibit this and I haven't seen any
> info about such restriction.
It's not prohibited, but it also was never required. Therefore it may
not be implemented in all gadget drivers.
> If some function is not working in
> such situation there is a bug in that function and it should be fixed.
That's fine. I wasn't pointing out a fundamental limitation, just a
fact that will have to be taken into account.
Anyway, instead of going through all this, why not do what I suggested
earlier (see http://marc.info/?l=linux-usb&m=139888691230119&w=2) and
create a "gadget" bus type? That would give userspace explicit control
over which gadget driver was bound to which UDC.
Or maybe that's not a very good fit with the existing code, since most
gadget drivers assume they will be bound to only one UDC at a time. So
instead, why not create a sysfs interface that allows userspace to
control which gadget drivers are bound to which UDCs?
As I recall, the original problem people were complaining about was
deferred probing. They didn't need fancy matching; all they wanted was
the ability to bind a gadget driver to a UDC some time after the gadget
driver was loaded and initialized. Something simple like Robert
Baldyga's patch is enough to do that.
Alan Stern
Hi guys,
On Mon, Feb 9, 2015 at 10:00 PM, Alan Stern <[email protected]> wrote:
> On Mon, 9 Feb 2015, Krzysztof Opasiak wrote:
>
>> > Why bother matching by name? Why not simply take the first
>> > available
>> > UDC?
>>
>> Because you may have more than one udc. This would allow to pick one by
>> name just like using configfs interface.
>
> Clearly it would be more flexible to allow the user to do the matching,
> the way configfs does (sysfs too).
>
>> > > Main feature of my path is not only deferred binding of gadget
>> > driver,
>> > > but also possibility to register/unregister udc at any time.
>> > > This is useful for user who can load, for example, udc module
>> > > if needed and unload it safely, not touching gadget driver.
>> >
>> > We can already do that with the existing code. There's no need for
>> > a
>> > patch.
>> >
>> > Also, it's not clear that the existing gadget drivers will work
>> > properly if they are unbound from one UDC and then bound again to
>> > another one. They were not written with that sort of thing in
>> > mind.
>> >
>>
>> What you have described is one of basics configfs features.
>> You should be able to bind and unbind your gadget whenever you want
>> and it should work properly after doing:
>>
>> ## create gadget
>> $ echo "udc.0" > UDC
>> $ echo "" > UDC
>> $ echo "udc.1" > UDC
>>
>> Function shouldn't care which udc it has been bound previously.
>> Only current one is important and on each unbind each function
>> should cleanup its state and prepare to be bound to another udc.
>> Configfs interface doesn't prohibit this and I haven't seen any
>> info about such restriction.
Thank you Krzysztof for this explanation that makes things more clear
for reviewers.
>
> It's not prohibited, but it also was never required. Therefore it may
> not be implemented in all gadget drivers.
>
>> If some function is not working in
>> such situation there is a bug in that function and it should be fixed.
>
> That's fine. I wasn't pointing out a fundamental limitation, just a
> fact that will have to be taken into account.
I also don't see any restrictions to bind/unbind gadget driver few times
and in case of such bug we just can fix it. I didn't see any issue with
gadget drivers that I used for verification, however, to be honest, I didn't
test it with all possible ones.
Anyway, it should work in legacy way (one gadget driver is bound to one uds)
so current behavior at least is not broken.
>
> Anyway, instead of going through all this, why not do what I suggested
> earlier (see http://marc.info/?l=linux-usb&m=139888691230119&w=2) and
> create a "gadget" bus type? That would give userspace explicit control
> over which gadget driver was bound to which UDC.
Exactly, my patch transforms udc-core to something like tiny bus with very basic
functionality. But in mentioned mailthread I see good ideas that is possible to
implement with small overhead.
>
> Or maybe that's not a very good fit with the existing code, since most
> gadget drivers assume they will be bound to only one UDC at a time. So
> instead, why not create a sysfs interface that allows userspace to
> control which gadget drivers are bound to which UDCs?
>
> As I recall, the original problem people were complaining about was
> deferred probing. They didn't need fancy matching; all they wanted was
> the ability to bind a gadget driver to a UDC some time after the gadget
> driver was loaded and initialized. Something simple like Robert
> Baldyga's patch is enough to do that.
This simplicity is also a limitation. As I mentioned before, sometimes it is
needed to be able to bind/unbind gadget driver multiple times to different UDCs.
A real case I faced recently is SoC with HighSpeed-only and SuperSpeed-only
UDCs. SuperSpeed-only UDC can't work on USB 2.0 speeds and vice versa.
The system has USB3.0 USB connector with soldered HS lines from
mentioned HS-only and SS lines from SS-only UDCs. Each UDC has it's own
device driver, so if we want to use both of them with one gadget driver, we
must be able to bind/unbind it multiple times to different UDCs.
Another one usecase is users who can unload udc drivers without carrying
about gadget drivers.
Third usecase is, for example, developers who can switch between dummy_hcd
and real UDC hardware without unloading gadget driver.
I'll work on improved version and will send new patch soon...
Best regards,
Ruslan
>
> Alan Stern
>
> -----Original Message-----
> From: Ruslan Bilovol [mailto:[email protected]]
> Sent: Tuesday, February 10, 2015 12:46 AM
> To: Alan Stern
> Cc: Krzysztof Opasiak; Peter Chen; [email protected];
> [email protected]; Balbi, Felipe;
> [email protected]; Andrzej Pietrasiewicz
> Subject: Re: [PATCH 1/2] usb: gadget: udc-core: independent
> registration of gadgets and gadget drivers
>
> Hi guys,
>
> On Mon, Feb 9, 2015 at 10:00 PM, Alan Stern
> <[email protected]> wrote:
> > On Mon, 9 Feb 2015, Krzysztof Opasiak wrote:
> >
> >> > Why bother matching by name? Why not simply take the first
> >> > available
> >> > UDC?
> >>
> >> Because you may have more than one udc. This would allow to pick
> one by
> >> name just like using configfs interface.
> >
> > Clearly it would be more flexible to allow the user to do the
> matching,
> > the way configfs does (sysfs too).
> >
> >> > > Main feature of my path is not only deferred binding of
> gadget
> >> > driver,
> >> > > but also possibility to register/unregister udc at any time.
> >> > > This is useful for user who can load, for example, udc
> module
> >> > > if needed and unload it safely, not touching gadget driver.
> >> >
> >> > We can already do that with the existing code. There's no
> need for
> >> > a
> >> > patch.
> >> >
> >> > Also, it's not clear that the existing gadget drivers will
> work
> >> > properly if they are unbound from one UDC and then bound again
> to
> >> > another one. They were not written with that sort of thing in
> >> > mind.
> >> >
> >>
> >> What you have described is one of basics configfs features.
> >> You should be able to bind and unbind your gadget whenever you
> want
> >> and it should work properly after doing:
> >>
> >> ## create gadget
> >> $ echo "udc.0" > UDC
> >> $ echo "" > UDC
> >> $ echo "udc.1" > UDC
> >>
> >> Function shouldn't care which udc it has been bound previously.
> >> Only current one is important and on each unbind each function
> >> should cleanup its state and prepare to be bound to another udc.
> >> Configfs interface doesn't prohibit this and I haven't seen any
> >> info about such restriction.
>
> Thank you Krzysztof for this explanation that makes things more
> clear
> for reviewers.
>
> >
> > It's not prohibited, but it also was never required. Therefore
> it may
> > not be implemented in all gadget drivers.
> >
> >> If some function is not working in
> >> such situation there is a bug in that function and it should be
> fixed.
> >
> > That's fine. I wasn't pointing out a fundamental limitation,
> just a
> > fact that will have to be taken into account.
>
> I also don't see any restrictions to bind/unbind gadget driver few
> times
> and in case of such bug we just can fix it. I didn't see any issue
> with
> gadget drivers that I used for verification, however, to be honest,
> I didn't
> test it with all possible ones.
> Anyway, it should work in legacy way (one gadget driver is bound to
> one uds)
> so current behavior at least is not broken.
>
> >
> > Anyway, instead of going through all this, why not do what I
> suggested
> > earlier (see http://marc.info/?l=linux-usb&m=139888691230119&w=2)
> and
> > create a "gadget" bus type? That would give userspace explicit
> control
> > over which gadget driver was bound to which UDC.
>
> Exactly, my patch transforms udc-core to something like tiny bus
> with very basic
> functionality. But in mentioned mailthread I see good ideas that is
> possible to
> implement with small overhead.
>
> >
> > Or maybe that's not a very good fit with the existing code, since
> most
> > gadget drivers assume they will be bound to only one UDC at a
> time. So
> > instead, why not create a sysfs interface that allows userspace
> to
> > control which gadget drivers are bound to which UDCs?
> >
> > As I recall, the original problem people were complaining about
> was
> > deferred probing. They didn't need fancy matching; all they
> wanted was
> > the ability to bind a gadget driver to a UDC some time after the
> gadget
> > driver was loaded and initialized. Something simple like Robert
> > Baldyga's patch is enough to do that.
>
> This simplicity is also a limitation. As I mentioned before,
> sometimes it is
> needed to be able to bind/unbind gadget driver multiple times to
> different UDCs.
> A real case I faced recently is SoC with HighSpeed-only and
> SuperSpeed-only
> UDCs. SuperSpeed-only UDC can't work on USB 2.0 speeds and vice
> versa.
> The system has USB3.0 USB connector with soldered HS lines from
> mentioned HS-only and SS lines from SS-only UDCs. Each UDC has it's
> own
> device driver, so if we want to use both of them with one gadget
> driver, we
> must be able to bind/unbind it multiple times to different UDCs.
> Another one usecase is users who can unload udc drivers without
> carrying
> about gadget drivers.
> Third usecase is, for example, developers who can switch between
> dummy_hcd
> and real UDC hardware without unloading gadget driver.
>
Just a stupid question. Why don't you use configfs composite gadget
instead of legacy gadgets?
In my opinion all things which you have described are working out-of-box
when you use configfs interface. It's mostly ready so you may create
equivalent of most legacy gadgets (apart from printer and tcm) and
just bind from one udc to another whenever you want.
Cheers,
--
Krzysztof Opasiak
Samsung R&D Institute Poland
Samsung Electronics
[email protected]
Hi Alan,
On Mon, Feb 9, 2015 at 10:00 PM, Alan Stern <[email protected]> wrote:
> On Mon, 9 Feb 2015, Krzysztof Opasiak wrote:
>
>> > Why bother matching by name? Why not simply take the first
>> > available
>> > UDC?
>>
>> Because you may have more than one udc. This would allow to pick one by
>> name just like using configfs interface.
>
> Clearly it would be more flexible to allow the user to do the matching,
> the way configfs does (sysfs too).
>
>> > > Main feature of my path is not only deferred binding of gadget
>> > driver,
>> > > but also possibility to register/unregister udc at any time.
>> > > This is useful for user who can load, for example, udc module
>> > > if needed and unload it safely, not touching gadget driver.
>> >
>> > We can already do that with the existing code. There's no need for
>> > a
>> > patch.
>> >
>> > Also, it's not clear that the existing gadget drivers will work
>> > properly if they are unbound from one UDC and then bound again to
>> > another one. They were not written with that sort of thing in
>> > mind.
>> >
>>
>> What you have described is one of basics configfs features.
>> You should be able to bind and unbind your gadget whenever you want
>> and it should work properly after doing:
>>
>> ## create gadget
>> $ echo "udc.0" > UDC
>> $ echo "" > UDC
>> $ echo "udc.1" > UDC
>>
>> Function shouldn't care which udc it has been bound previously.
>> Only current one is important and on each unbind each function
>> should cleanup its state and prepare to be bound to another udc.
>> Configfs interface doesn't prohibit this and I haven't seen any
>> info about such restriction.
>
> It's not prohibited, but it also was never required. Therefore it may
> not be implemented in all gadget drivers.
>
>> If some function is not working in
>> such situation there is a bug in that function and it should be fixed.
>
> That's fine. I wasn't pointing out a fundamental limitation, just a
> fact that will have to be taken into account.
>
> Anyway, instead of going through all this, why not do what I suggested
> earlier (see http://marc.info/?l=linux-usb&m=139888691230119&w=2) and
> create a "gadget" bus type? That would give userspace explicit control
> over which gadget driver was bound to which UDC.
>
> Or maybe that's not a very good fit with the existing code, since most
> gadget drivers assume they will be bound to only one UDC at a time. So
> instead, why not create a sysfs interface that allows userspace to
> control which gadget drivers are bound to which UDCs?
>
> As I recall, the original problem people were complaining about was
> deferred probing. They didn't need fancy matching; all they wanted was
> the ability to bind a gadget driver to a UDC some time after the gadget
> driver was loaded and initialized. Something simple like Robert
> Baldyga's patch is enough to do that.
I looked over my patch and see that it doesn't automatically rebind
gadget driver to another available UDC after previous UDC is unbound.
The Gadget Bus mentioned previously is good thing but so far it seems there
is no users of such approach. So I left only deferred registration
from my patch.
I still keep it inside of udc-core since we also need to keep tracking UDC name
if somebody wanted to bind gadget driver to specific UDC and it looks
like good place to maintain this. I'll send second version of patches soon
--
Best regards,
Ruslan Bilvol
Hi Krzysztof,
On Tue, Feb 10, 2015 at 10:47 AM, Krzysztof Opasiak
<[email protected]> wrote:
>
>
>> -----Original Message-----
>> From: Ruslan Bilovol [mailto:[email protected]]
>> Sent: Tuesday, February 10, 2015 12:46 AM
>> To: Alan Stern
>> Cc: Krzysztof Opasiak; Peter Chen; [email protected];
>> [email protected]; Balbi, Felipe;
>> [email protected]; Andrzej Pietrasiewicz
>> Subject: Re: [PATCH 1/2] usb: gadget: udc-core: independent
>> registration of gadgets and gadget drivers
>>
>> Hi guys,
>>
>> On Mon, Feb 9, 2015 at 10:00 PM, Alan Stern
>> <[email protected]> wrote:
>> > On Mon, 9 Feb 2015, Krzysztof Opasiak wrote:
>> >
>> >> > Why bother matching by name? Why not simply take the first
>> >> > available
>> >> > UDC?
>> >>
>> >> Because you may have more than one udc. This would allow to pick
>> one by
>> >> name just like using configfs interface.
>> >
>> > Clearly it would be more flexible to allow the user to do the
>> matching,
>> > the way configfs does (sysfs too).
>> >
>> >> > > Main feature of my path is not only deferred binding of
>> gadget
>> >> > driver,
>> >> > > but also possibility to register/unregister udc at any time.
>> >> > > This is useful for user who can load, for example, udc
>> module
>> >> > > if needed and unload it safely, not touching gadget driver.
>> >> >
>> >> > We can already do that with the existing code. There's no
>> need for
>> >> > a
>> >> > patch.
>> >> >
>> >> > Also, it's not clear that the existing gadget drivers will
>> work
>> >> > properly if they are unbound from one UDC and then bound again
>> to
>> >> > another one. They were not written with that sort of thing in
>> >> > mind.
>> >> >
>> >>
>> >> What you have described is one of basics configfs features.
>> >> You should be able to bind and unbind your gadget whenever you
>> want
>> >> and it should work properly after doing:
>> >>
>> >> ## create gadget
>> >> $ echo "udc.0" > UDC
>> >> $ echo "" > UDC
>> >> $ echo "udc.1" > UDC
>> >>
>> >> Function shouldn't care which udc it has been bound previously.
>> >> Only current one is important and on each unbind each function
>> >> should cleanup its state and prepare to be bound to another udc.
>> >> Configfs interface doesn't prohibit this and I haven't seen any
>> >> info about such restriction.
>>
>> Thank you Krzysztof for this explanation that makes things more
>> clear
>> for reviewers.
>>
>> >
>> > It's not prohibited, but it also was never required. Therefore
>> it may
>> > not be implemented in all gadget drivers.
>> >
>> >> If some function is not working in
>> >> such situation there is a bug in that function and it should be
>> fixed.
>> >
>> > That's fine. I wasn't pointing out a fundamental limitation,
>> just a
>> > fact that will have to be taken into account.
>>
>> I also don't see any restrictions to bind/unbind gadget driver few
>> times
>> and in case of such bug we just can fix it. I didn't see any issue
>> with
>> gadget drivers that I used for verification, however, to be honest,
>> I didn't
>> test it with all possible ones.
>> Anyway, it should work in legacy way (one gadget driver is bound to
>> one uds)
>> so current behavior at least is not broken.
>>
>> >
>> > Anyway, instead of going through all this, why not do what I
>> suggested
>> > earlier (see http://marc.info/?l=linux-usb&m=139888691230119&w=2)
>> and
>> > create a "gadget" bus type? That would give userspace explicit
>> control
>> > over which gadget driver was bound to which UDC.
>>
>> Exactly, my patch transforms udc-core to something like tiny bus
>> with very basic
>> functionality. But in mentioned mailthread I see good ideas that is
>> possible to
>> implement with small overhead.
>>
>> >
>> > Or maybe that's not a very good fit with the existing code, since
>> most
>> > gadget drivers assume they will be bound to only one UDC at a
>> time. So
>> > instead, why not create a sysfs interface that allows userspace
>> to
>> > control which gadget drivers are bound to which UDCs?
>> >
>> > As I recall, the original problem people were complaining about
>> was
>> > deferred probing. They didn't need fancy matching; all they
>> wanted was
>> > the ability to bind a gadget driver to a UDC some time after the
>> gadget
>> > driver was loaded and initialized. Something simple like Robert
>> > Baldyga's patch is enough to do that.
>>
>> This simplicity is also a limitation. As I mentioned before,
>> sometimes it is
>> needed to be able to bind/unbind gadget driver multiple times to
>> different UDCs.
>> A real case I faced recently is SoC with HighSpeed-only and
>> SuperSpeed-only
>> UDCs. SuperSpeed-only UDC can't work on USB 2.0 speeds and vice
>> versa.
>> The system has USB3.0 USB connector with soldered HS lines from
>> mentioned HS-only and SS lines from SS-only UDCs. Each UDC has it's
>> own
>> device driver, so if we want to use both of them with one gadget
>> driver, we
>> must be able to bind/unbind it multiple times to different UDCs.
>> Another one usecase is users who can unload udc drivers without
>> carrying
>> about gadget drivers.
>> Third usecase is, for example, developers who can switch between
>> dummy_hcd
>> and real UDC hardware without unloading gadget driver.
>>
>
> Just a stupid question. Why don't you use configfs composite gadget
> instead of legacy gadgets?
>
> In my opinion all things which you have described are working out-of-box
> when you use configfs interface. It's mostly ready so you may create
> equivalent of most legacy gadgets (apart from printer and tcm) and
> just bind from one udc to another whenever you want.
It's because legacy gadgets are easy to use and usually don't need any
userspace-side configuration. Are there any plans to remove legacy
drivers in the future?
Best regards,
Ruslan
>
> Cheers,
>
> --
> Krzysztof Opasiak
> Samsung R&D Institute Poland
> Samsung Electronics
> [email protected]
>
>
>
>
--
Best regards,
Ruslan Bilvol
W dniu 15.02.2015 o 23:43, Ruslan Bilovol pisze:
<snip>
>>
>> In my opinion all things which you have described are working out-of-box
>> when you use configfs interface. It's mostly ready so you may create
>> equivalent of most legacy gadgets (apart from printer and tcm) and
>> just bind from one udc to another whenever you want.
>
> It's because legacy gadgets are easy to use and usually don't need any
> userspace-side configuration. Are there any plans to remove legacy
> drivers in the future?
>
I'm not going to express strong opinions here, but their name implies
that this can happen, some time in the future.
And I also think it will not happen before the userspace part
(libusbg, gt, gadgetd etc) is mature enough. My personal opinion
in that matter is that it will take at least a couple of years
to remove legacy gadgets entirely.
AP
Hi Andrzej,
On Mon, Feb 16, 2015 at 10:07 AM, Andrzej Pietrasiewicz
<[email protected]> wrote:
> W dniu 15.02.2015 o 23:43, Ruslan Bilovol pisze:
>
> <snip>
>
>>>
>>> In my opinion all things which you have described are working out-of-box
>>> when you use configfs interface. It's mostly ready so you may create
>>> equivalent of most legacy gadgets (apart from printer and tcm) and
>>> just bind from one udc to another whenever you want.
>>
>>
>> It's because legacy gadgets are easy to use and usually don't need any
>> userspace-side configuration. Are there any plans to remove legacy
>> drivers in the future?
>>
>
> I'm not going to express strong opinions here, but their name implies
> that this can happen, some time in the future.
>
> And I also think it will not happen before the userspace part
> (libusbg, gt, gadgetd etc) is mature enough. My personal opinion
> in that matter is that it will take at least a couple of years
> to remove legacy gadgets entirely.
OK, so it looks like there is a sense even to add new gadget/functions
with legacy support
Thanks,
Ruslan
>
> AP
>
W dniu 17.02.2015 o 22:02, Ruslan Bilovol pisze:
> Hi Andrzej,
>
> On Mon, Feb 16, 2015 at 10:07 AM, Andrzej Pietrasiewicz
> <[email protected]> wrote:
>> W dniu 15.02.2015 o 23:43, Ruslan Bilovol pisze:
>>
>> <snip>
>>
>>>>
>>>> In my opinion all things which you have described are working out-of-box
>>>> when you use configfs interface. It's mostly ready so you may create
>>>> equivalent of most legacy gadgets (apart from printer and tcm) and
>>>> just bind from one udc to another whenever you want.
>>>
>>>
>>> It's because legacy gadgets are easy to use and usually don't need any
>>> userspace-side configuration. Are there any plans to remove legacy
>>> drivers in the future?
>>>
>>
>> I'm not going to express strong opinions here, but their name implies
>> that this can happen, some time in the future.
>>
>> And I also think it will not happen before the userspace part
>> (libusbg, gt, gadgetd etc) is mature enough. My personal opinion
>> in that matter is that it will take at least a couple of years
>> to remove legacy gadgets entirely.
>
> OK, so it looks like there is a sense even to add new gadget/functions
> with legacy support
>
I'm not sure what you mean exactly.
For sure legacy gadgets are supported as long as they are
a part of the mainline kernel. So any changes you make
to the kernel must not affect the legacy gadgets, or you
need to modify the legacy gadgets too and have them working.
But adding new legacy-style gadgets is a completely different
story. IMHO you need a _very_ good reason to succeed,
but I remember Felipe expressing an opinion that chances
or merging another legacy gadget were zero.
AP