On 8/2/21 5:12 pm, Sebastian Huber wrote: > On 08/02/2021 06:50, Chris Johns wrote: > >>> Another approach is to remove the greedy allocation functions and test the >>> no >>> memory conditions differently. We could wrap the allocator function and let >>> if >>> fail every n-the call. With this you can write generic tests like: >>> >>> for i=1,2,... >>> >>> let allocate fail in i-th call >>> >>> p = test() >>> >>> if p != NULL: >>> >>> done >> I think greedy should have a separate test as it is an interface. A test >> means >> sbrk can be tested because I am wondering if it is. Removing it and playing >> with >> allocators not an option for me. There are too many powerpc BSP we have not >> been >> able to test and have sat broken for a decade now. They are a tangle of >> interconnected pieces that are fragile and the ISA bridge IRQ bug is an >> example. >> If we cannot test all powerpc BSPs we need to step carefully. My efforts are >> focused on a specific working PowerPC BSP. > > This approach would be independent of the BSP. The problem is that this stuff > is > used in a couple of tests:
I am not sure about this approach. The fix I have is simple and easy and apart from something in _Heap_Free_block taking what I think is too much time it looks OK. FYI It seems the sbrk heap extend slows down in _Heap_Free_block. No idea why that would have a slow down. Chris _______________________________________________ devel mailing list devel@rtems.org http://lists.rtems.org/mailman/listinfo/devel
