usb: gadget: storage: Remove warning message
PATCH v2 1/1 usb: gadget: storage: Remove warning message
This change is to fix below warning message in following scenario:usb_composite_setup_continue: Unexpected call
When system tried to enter suspend, the fsg_disable() will be called to disable fsg driver and send a signal to fsg_main_thread. However, at this point, the fsg_main_thread has already been frozen and can not respond to this signal. So, this signal will be pended until fsg_main_thread wakes up.
Once system resumes from suspend, fsg_main_thread will detect a signal pended and do some corresponding action (in handle_exception()). Then, host will send some setup requests (get descriptor, set configuration...) to UDC driver trying to enumerate this device. During the handling of "set configuration" request, it will try to sync up with fsg_main_thread by sending a signal (which is the same as the signal sent by fsg_disable) to it. In a similar manner, once the fsg_main_thread receives this signal, it will call handle_exception() to handle the request.
However, if the fsg_main_thread wakes up from suspend a little late and "set configuration" request from Host arrives a little earlier, fsg_main_thread might come across the request from "set configuration" when it handles the signal from fsg_disable(). In this case, it will handle this request as well. So, when fsg_main_thread tries to handle the signal sent from "set configuration" later, there will nothing left to do and warning message "Unexpected call" is printed.
---
v2: remove the copyright info
--- drivers/usb/gadget/function/f_mass_storage.c | 17 ++++++++++++++--- drivers/usb/gadget/function/storage_common.h | 1 + 2 files changed, 15 insertions(+), 3 deletions(-)
drivers/usb/gadget/function/f_mass_storage.c +14 -3
@@ -2294,7 +2294,7 @@ static void fsg_disable(struct usb_function *f) {struct fsg_dev *fsg =fsg_from_func (f); fsg->common->new_fsg =NULL ; -raise_exception (fsg->common,FSG_STATE_CONFIG_CHANGE ); +raise_exception (fsg->common,FSG_STATE_DISCONNECT ); }
@@ -2307,6 +2307,7 @@ static void handle_exception(struct fsg_common *common)enum fsg_state old_state;struct fsg_lun *curlun;unsigned int exception_req_tag; +struct fsg_dev *fsg;/* * Clear the existing signals. Anything but SIGUSR1 is converted
@@ -2413,9 +2414,19 @@ static void handle_exception(struct fsg_common *common)break ;case FSG_STATE_CONFIG_CHANGE : -do_set_interface (common, common->new_fsg); -if (common->new_fsg) + fsg = common->new_fsg; +/* +* Add a check here to double confirm if a disconnect event +* occurs and common->new_fsg has been cleared. +*/ +if (fsg) { +do_set_interface (common, fsg);usb_composite_setup_continue (common->cdev); + } +break ; + +case FSG_STATE_DISCONNECT : +do_set_interface (common,NULL );break ;case FSG_STATE_EXIT :
drivers/usb/gadget/function/storage_common.h +1 -0
@@ -161,6 +161,7 @@ enum fsg_state {FSG_STATE_ABORT_BULK_OUT ,FSG_STATE_PROTOCOL_RESET ,FSG_STATE_CONFIG_CHANGE , +FSG_STATE_DISCONNECT ,FSG_STATE_EXIT , FSG_STATE_TERMINATED };
Signature · 1 line
2.1.4
Quoted 39 lines · EJ Hsu
This change is to fix below warning message in following scenario:
usb_composite_setup_continue: Unexpected callWhen system tried to enter suspend, the
fsg_disable()will be called to disable fsg driver and send a signal tofsg_main_thread. However, at this point, thefsg_main_threadhas already been frozen and can not respond to this signal. So, this signal will be pended untilfsg_main_threadwakes up.Once system resumes from suspend,
fsg_main_threadwill detect a signal pended and do some corresponding action (inhandle_exception()). Then, host will send some setup requests (get descriptor, set configuration...) to UDC driver trying to enumerate this device. During the handling of "set configuration" request, it will try to sync up withfsg_main_threadby sending a signal (which is the same as the signal sent byfsg_disable) to it. In a similar manner, once thefsg_main_threadreceives this signal, it will callhandle_exception()to handle the request.However, if the
fsg_main_threadwakes up from suspend a little late and "set configuration" request from Host arrives a little earlier,fsg_main_threadmight come across the request from "set configuration" when it handles the signal fromfsg_disable(). In this case, it will handle this request as well. So, whenfsg_main_threadtries to handle the signal sent from "set configuration" later, there will nothing left to do and warning message "Unexpected call" is printed.Signed-off-by EJ Hsu <[email protected]>---
v2: remove the copyright info--- drivers/usb/gadget/function/f_mass_storage.c | 17 ++++++++++++++--- drivers/usb/gadget/function/storage_common.h | 1 + 2 files changed, 15 insertions(+), 3 deletions(-)drivers/usb/gadget/function/f_mass_storage.c +0
-0
drivers/usb/gadget/function/f_mass_storage.c static void fsg_disable(struct usb_function *f) · line 2296
Quoted 3 lines
drivers/usb/gadget/function/f_mass_storage.c +0
-0@@ -2294,7 +2294,7 @@ static void fsg_disable(struct usb_function *f) {struct fsg_dev *fsg =fsg_from_func (f); fsg->common->new_fsg =NULL ;
That line is no longer needed.
Quoted 40 lines
drivers/usb/gadget/function/f_mass_storage.c +14
-3-raise_exception (fsg->common,FSG_STATE_CONFIG_CHANGE ); +raise_exception (fsg->common,FSG_STATE_DISCONNECT ); }@@ -2307,6 +2307,7 @@ static void handle_exception(struct fsg_common *common)enum fsg_state old_state;struct fsg_lun *curlun;unsigned int exception_req_tag; +struct fsg_dev *fsg;/* * Clear the existing signals. Anything but SIGUSR1 is converted @@ -2413,9 +2414,19 @@ static void handle_exception(struct fsg_common *common)break ;case FSG_STATE_CONFIG_CHANGE : -do_set_interface (common, common->new_fsg); -if (common->new_fsg) + fsg = common->new_fsg; +/* +* Add a check here to double confirm if a disconnect event +* occurs and common->new_fsg has been cleared. +*/ +if (fsg) { +do_set_interface (common, fsg);usb_composite_setup_continue (common->cdev); + } +break ; + +case FSG_STATE_DISCONNECT : +do_set_interface (common,NULL );break ;case FSG_STATE_EXIT :drivers/usb/gadget/function/storage_common.h +0
-0
drivers/usb/gadget/function/storage_common.h enum fsg_state { · line 167
Quoted 7 lines
drivers/usb/gadget/function/storage_common.h +1
-0@@ -161,6 +161,7 @@ enum fsg_state {FSG_STATE_ABORT_BULK_OUT ,FSG_STATE_PROTOCOL_RESET ,FSG_STATE_CONFIG_CHANGE , +FSG_STATE_DISCONNECT ,FSG_STATE_EXIT , FSG_STATE_TERMINATED };
You forgot to change fsg_unbind() to use FSG_STATE_DISCONNECT. And when that's done, it will no longer need to set common->new_fsg to NULL either.
This is sort of a band-aid approach. The real problem is that the original design of the driver never intended for there to be more than one outstanding CONFIG_CHANGE event, so naturally there are scenarios where it doesn't work right when that assumption is violated. This patch addresses one of those scenarios, but there may be others remaining.
Ultimately the problem boils down to synchronization with the composite core. Some of the callbacks made by the core take time to fully process, so what should happen if the core makes a second callback before the first one is completely processed?
On the other hand, I can't think of anything better at the moment.
Alan Stern
Quoted 4 lines
You forgot to change
fsg_unbind()to useFSG_STATE_DISCONNECT. And when that's done, it will no longer need to setcommon->new_fsgto NULL either.
Yes, we should change fsg_unbind() as well.
Quoted 7 lines
This is sort of a band-aid approach. The real problem is that the original design of the driver never intended for there to be more than one outstanding
CONFIG_CHANGEevent, so naturally there are scenarios where it doesn't work right when that assumption is violated. This patch addresses one of those scenarios, but there may be others remaining.Ultimately the problem boils down to synchronization with the composite core.
Thanks for your reviewing, I agree with your point.
Quoted 7 lines
Some of the callbacks made by the core take time to fully process, so what should happen if the core makes a second callback before the first one is completely processed?
On the other hand, I can't think of anything better at the moment.
Alan Stern
Actually, composite core have tried to deal with this case by using a spinlock. (please refer to the cdev->lock)
The problem here is that the callback of fsg will delegate the request to fsg_main_thread. That is, the handling of fsg request will run in parallel with composite core.
In my opinion, this issue can be avoided if we handle these request directly in the callback of fsg instead of handing it over to fsg_main_thread. But this might take too much time in the interrupt context, which is not good for system performance.
Please correct me if there is anything wrong. Thanks
EJ
--nvpublic
Quoted 34 lines · EJ Hsu
Quoted 3 lines
You forgot to change
fsg_unbind()to useFSG_STATE_DISCONNECT. And when that's done, it will no longer need to setcommon->new_fsgto NULL either.Yes, we should change
fsg_unbind()as well.Quoted 7 lines
This is sort of a band-aid approach. The real problem is that the original design of the driver never intended for there to be more than one outstanding
CONFIG_CHANGEevent, so naturally there are scenarios where it doesn't work right when that assumption is violated. This patch addresses one of those scenarios, but there may be others remaining.Ultimately the problem boils down to synchronization with the composite core.
Thanks for your reviewing, I agree with your point.
Quoted 7 lines
Some of the callbacks made by the core take time to fully process, so what should happen if the core makes a second callback before the first one is completely processed?
On the other hand, I can't think of anything better at the moment.
Alan Stern
Actually, composite core have tried to deal with this case by using a spinlock. (please refer to the
cdev->lock)
The problem here is that the callback of fsg will delegate the request tofsg_main_thread. That is, the handling of fsg request will run in parallel with composite core.
In my opinion, this issue can be avoided if we handle these request directly in the callback of fsg instead of handing it over tofsg_main_thread. But this might take too much time in the interrupt context, which is not good for system performance.
Quoted 1 line
Please correct me if there is anything wrong. Thanks
In theory, the mass-storage function could also be fixed to be more careful about locking and synchronization. For example, never set or read common->next_fsg unless you're holding the fsg lock, and don't raise a CONFIG_CHANGE exception if another one is already pending.
But I think your patch will be good enough for now, once you have fixed up the two issues mentioned earlier.
Alan Stern