Xenomai 3.1 missing munmap Posix wrapper

Paal Tamas paal_to at freemail.hu
Mon Apr 6 19:15:16 CEST 2020


>>>> Hello Everybody,
>>>>
>>>> I just migrated from 2.6.2.1. to 3.1 and run into an issue in my user mode application.
>>>> The Posix wrapper for the munmap() function is missing. I mean the cobalt.wrappers file does not have a --wrap munmap line and the library does not have the function either. Why is that?
>>>> The mmap wrapper is there. I need to unmap the memory before closing the file descriptor. At least I did it this way in my old sources. Can I skip the munmap() call in this new Xenomai version before closing the descriptor?
>>>
>>> You mmap memory against an RTDM device handle, the only difference to
>>> mmap against a Linux device is who provides the memory. Releasing that
>>> mapping is just the same as with Linux: call munmap.
>>>
>>> If your driver has anything to do on release, register vmops with the
>>> backing vma. See e.g. spi_master_mmap.
>>>
>>> Jan
>>>
>> 
>> This is an ARM system (MX6) where I use the external parallel bus of the CPU to communicate with an FPGA in a real-time fashion in user mode. I need to access memory mapped registers of the CPU during the configuration of the CPU external bus interface, and the address range I use to "access" the FPGA is mapped to memory as well. In the old 2.6.2.1 system I was using the mmap() and munmap() wrapper functions for this purpose. I linked only Xenomai libraries to my executable.
>> Do you mean I should not use the mmap() wrapper from the cobalt library anymore. Use the Linux equivalent in 3.1? Will it not lead to any trouble? I mean will I not switch to secondary mode when I access the memory I mapped this way for example?
>
>Mapping an unmapping were never time sensitive operations. No magic 
>happened when they were executed, memory locking was the magic to ensure 
>that backing memory is always available on access (but even that is 
>unrelated when we talk about device memory, e.g. MMIO).
>
>The only reason for mmap being wrapped is that it can be used on 
>resources (represented by RTDM file descriptors) the Linux kernel itself 
>has no knowledge about, only Xenomai. If you use mmap to map /dev/mem, 
>ie. without a proper kernel driver, this will continue to work as before.
>
>Jan
>

Understood. Thank you very much for the explanation. 
What I did technically is removed the mmap and munmap wrappers from my makefile. This led to an error-free compilation. I assume the Linux libraries got into the business this way...
Thanks again,
Tamas



More information about the Xenomai mailing list