How to free unused dispmanx layer (in case of SIGSEGV)

[cmarty]

Hello.
I have opengl app (using openframeworks).
I try to solve the situation when my program receive SIGSEGV (or for debug purpose I send him SIGSEGV signal) the program do not free dispmanx layer (resource) and last state of this dispmanx layer is still present on display. When I run program again the old dispmanx layer is still present and new dispmanx layer is created. Is there any way how I can remove, delete the old unused dispmanx layer/resource ?
For list of active dispmanx layers i use command "vcgencmd dispmanx_list"

[jamesh]

Trap the SIG, do the dispmanx deallocation in the handler?

[Martin Frezman]

Trap the SIG, do the dispmanx deallocation in the handler?

Well, yes, but i assume the question is more: How do i do this after the fact? I.e., what if it has already happened?

Also:

Code: Select all

$ vcgencmd dispmanx_list 
error=1 error_msg="Command not registered"
$

[jamesh]

Trap the SIG, do the dispmanx deallocation in the handler?

Well, yes, but i assume the question is more: How do i do this after the fact? I.e., what if it has already happened?

Also:

Code: Select all

$ vcgencmd dispmanx_list 
error=1 error_msg="Command not registered"
$

Nothing possible 'after it has happened' I suspect, without updated firmware. Big red switch time.But will double check with the dispmanx experts.

[blackshard83]

AFAIK there is no documented way for a process to get the handle to a dispmanx layer of another process, so you can't free it after the leak.

Though I never tried to store the dispmanx handles somewhere (maybe on the filesystem) and let another process pick them up and do the calls to free them. From a process separation point of view dispmanx should not allow such thing, although it should automatically clean the garbage when the process dies...

[Martin Frezman]

Though I never tried to store the dispmanx handles somewhere (maybe on the filesystem) and let another process pick them up and do the calls to free them.

This sounds like a good exercise for the OP - given that he has, in his possession, a program that can crash, leaving a layer still visible. It sounds like he has most of what he needs at his fingertips.

All you need to do is add code to write out the handle to a file before the crash.

[jwatte]

last state of this dispmanx layer is still present on display

Are you sure the process has actually died? If you check for it with "ps aux | grep <procname>" is it still alive?
It sounds like a pretty bad driver bug if somehow the OS lets resources stay allocated after a process goes away.

[6by9]

Checking the firmware source there is handling for the VCHIQ service closing, and it does free up all resources.
As a quick test, taking hello_dispmanx and inserting a div0 before the sleep, the overlay comes up, and is then cleared up as the app. It is going through that VCHIQ service closing handler.

"vcgencmd dispmanx_list" was added to the firmware in September 2015. Are you running very old firmware? What does "vcgencmd version" report?
"vcgencmd commands" will list all the available commands that are registered.