... > > >>> --- a/common/memsize.c > > >>> +++ b/common/memsize.c > > >>> @@ -33,6 +33,9 @@ > > >>> * Check memory range for valid RAM. A simple memory test determines > > >>> * the actually available RAM size between addresses `base' and > > >>> * `base + maxsize'. > > >>> + * > > >>> + * This function modifies the RAM. Do not use it if you're running from > > >>> + * the RAM you are going to detect! > > >>> */ > > >> > > >> Actually, I don't see how it modifies the RAM, at least permanently. The > > >> values it erase are backed up, and there's no concurrency at barebox > > >> level, so we are sure that the value saved will still be the one that > > >> would need to be backed up at the end of the function, right? > > > > > > Yes, it restores the values, but how do you make sure the function does > > > not modify the instructions you are currently executing? You need bad > > > luck to hit this, but sooner or later this will happen. > > > > Ah, yes, this would be nasty indeed. Is there a way to know the end > > address of barebox into RAM ? or the address it has been loaded to and > > the size of its binary, so that we can just check the part that doesn't > > hold barebox? > > See include/asm-generic/sections.h. You have to avoid modifying > everything between _text and __bss_stop. I haven't looked how exactly > get_dram_size works. Normally this function would have to test every > location at a power of 2, that would be: > > 1 2 4 ... 64MiB 128MiB > > It seems you have to make sure that your binary does not cross a power > of 2 boundary, then you should be safe. Let's put "get_ram_size" function in a separate section inside .text. Then we can at least make runtime warning about placing this section inside our tested region. --- _______________________________________________ barebox mailing list barebox@xxxxxxxxxxxxxxxxxxx http://lists.infradead.org/mailman/listinfo/barebox
