Hi Paul, On Fri, Jun 15, 2007 at 02:25:43AM +0300, Paul Sokolovsky wrote: > Further moving towards live ramdisks stuff... WinCE alloc memory > limits strike again - unable to load some 16Mb initrd. This time I > finally coded mmap-based solution, with it I was able to load 27Mb > initrd (though wince complained with low memory dialog; all that on > hx4700 after hard reset).
The use of CreateFileMapping / MapViewOfFile looks sane to me. I think it would be worthwhile to see if that works everywhere. Note that the problems we've had in the past were not with allocation, but with finding the underlying physical pages. I don't see an issue with changing the allocation scheme. > Another issue I hit during experiments is that initrd buffer wanted > to overwrite itself. I fixed that with memmove-like solution, moving > blocks from the end instead of beginning. Be careful - if you copy the pages in reverse order, you need to make sure the pages don't overwrite the preloader. > Patch is attached, comments welcome. I'm not sure we just should > replace old alloc code with new - we faced already just too much > issues to collect and keep different allocation algorithms about. As > for selection of them and defaults, worth thinking. I'd prefer to not to have multiple allocation paths. That will increase the QA effort. Since we need to support the lowest common denominator anyway, we might as well make that the standard and ensure that it works on all platforms. -Kevin _______________________________________________ Haret mailing list [email protected] https://handhelds.org/mailman/listinfo/haret
