| Message ID | 1425350171-26333-1-git-send-email-xerofoify@gmail.com (mailing list archive) |
|---|---|
| State | Rejected |
| Headers | show |
On 3 March 2015 at 03:36, Nicholas Krause <xerofoify@gmail.com> wrote: > Releases resources and deregisters the stucture pointer ar passed by the caller to the function, ath10k_core_register_work > if unable to probe the structure pointer successfully with a call to ath10k_core_probe_fw. Further more if this happerns > we must first jump to the label err for the goto statement required to jump to handle this particular error in the function, > ath10k_core_register_work. After we are in the correct error section we must free the resources for the structure pointer,ar > with a call to the function, ath10k_core_unregister to free resources allocated for the structure pointer,ar. > > Signed-off-by: Nicholas Krause <xerofoify@gmail.com> > --- > drivers/net/wireless/ath/ath10k/core.c | 4 +--- > 1 file changed, 1 insertion(+), 3 deletions(-) > > diff --git a/drivers/net/wireless/ath/ath10k/core.c b/drivers/net/wireless/ath/ath10k/core.c > index 310e12b..8b2ca25 100644 > --- a/drivers/net/wireless/ath/ath10k/core.c > +++ b/drivers/net/wireless/ath/ath10k/core.c > @@ -1307,9 +1307,7 @@ err_unregister_mac: > err_release_fw: > ath10k_core_free_firmware_files(ar); > err: > - /* TODO: It's probably a good idea to release device from the driver > - * but calling device_release_driver() here will cause a deadlock. > - */ > + ath10k_core_unregister(ar); > return; > } Did you test this? This will deadlock. ath10k_core_unregister() tries to cancel ar->register_work. This won't work if you call it from the worker itself. Moreover if I ignore the deadlock ath10k_core_unregister() would do nothing else in this context because ATH10K_FLAG_CORE_REGISTERED wouldn't be even set. If you're interested in dealing with this TODO I suggest you read through the original thread which led to the current state of affairs: http://www.spinics.net/lists/linux-wireless/msg124004.html Micha?
On 2015-03-03 01:40 AM, Michal Kazior wrote: > On 3 March 2015 at 03:36, Nicholas Krause <xerofoify@gmail.com> wrote: >> Releases resources and deregisters the stucture pointer ar passed by the caller to the function, ath10k_core_register_work >> if unable to probe the structure pointer successfully with a call to ath10k_core_probe_fw. Further more if this happerns >> we must first jump to the label err for the goto statement required to jump to handle this particular error in the function, >> ath10k_core_register_work. After we are in the correct error section we must free the resources for the structure pointer,ar >> with a call to the function, ath10k_core_unregister to free resources allocated for the structure pointer,ar. >> >> Signed-off-by: Nicholas Krause <xerofoify@gmail.com> >> --- >> drivers/net/wireless/ath/ath10k/core.c | 4 +--- >> 1 file changed, 1 insertion(+), 3 deletions(-) >> >> diff --git a/drivers/net/wireless/ath/ath10k/core.c b/drivers/net/wireless/ath/ath10k/core.c >> index 310e12b..8b2ca25 100644 >> --- a/drivers/net/wireless/ath/ath10k/core.c >> +++ b/drivers/net/wireless/ath/ath10k/core.c >> @@ -1307,9 +1307,7 @@ err_unregister_mac: >> err_release_fw: >> ath10k_core_free_firmware_files(ar); >> err: >> - /* TODO: It's probably a good idea to release device from the driver >> - * but calling device_release_driver() here will cause a deadlock. >> - */ >> + ath10k_core_unregister(ar); >> return; >> } > > Did you test this? This will deadlock. ath10k_core_unregister() tries > to cancel ar->register_work. This won't work if you call it from the > worker itself. Moreover if I ignore the deadlock > ath10k_core_unregister() would do nothing else in this context because > ATH10K_FLAG_CORE_REGISTERED wouldn't be even set. > > If you're interested in dealing with this TODO I suggest you read > through the original thread which led to the current state of affairs: > > http://www.spinics.net/lists/linux-wireless/msg124004.html > > > Micha? > Michal, Thanks for the info and the thread related to this issue. After reading it I was wondering if we can test my patch on actual hardware using this driver with my patch included in order to see what issues are still there. I don't have the hardware lying around for this but if someone can test my patch out and see where there are still deadlock issues or other bugs to my patch I am certain I will be able to figure this out. Thanks, Nick P.S. I am a student so my budget for test hardware is a little thin. :(
diff --git a/drivers/net/wireless/ath/ath10k/core.c b/drivers/net/wireless/ath/ath10k/core.c index 310e12b..8b2ca25 100644 --- a/drivers/net/wireless/ath/ath10k/core.c +++ b/drivers/net/wireless/ath/ath10k/core.c @@ -1307,9 +1307,7 @@ err_unregister_mac: err_release_fw: ath10k_core_free_firmware_files(ar); err: - /* TODO: It's probably a good idea to release device from the driver - * but calling device_release_driver() here will cause a deadlock. - */ + ath10k_core_unregister(ar); return; }
Releases resources and deregisters the stucture pointer ar passed by the caller to the function, ath10k_core_register_work if unable to probe the structure pointer successfully with a call to ath10k_core_probe_fw. Further more if this happerns we must first jump to the label err for the goto statement required to jump to handle this particular error in the function, ath10k_core_register_work. After we are in the correct error section we must free the resources for the structure pointer,ar with a call to the function, ath10k_core_unregister to free resources allocated for the structure pointer,ar. Signed-off-by: Nicholas Krause <xerofoify@gmail.com> --- drivers/net/wireless/ath/ath10k/core.c | 4 +--- 1 file changed, 1 insertion(+), 3 deletions(-)