On 23 March 2018 at 15:15, Bart Van Assche <Bart.VanAssche@xxxxxxx> wrote: > On Fri, 2018-03-23 at 08:49 -0600, Jens Axboe wrote: >> Right, this is identical to what I reported. I get tons of these with >> the patch, not without. > > Since I can reproduce this I will have a look at what is going on. I wonder if the manner that fio allocates redzones is incompatible with valgrind's due to fio's redzone's potentially being smaller than the size of a pointer and the pre-redzone being partially inside the header structure. I'm attaching a patch I made but it still seems to generate valgrind warnings that aren't there when falling back to malloc/free without pools... -- Sitsofe | http://sucs.org/~sits/
Attachment:
smalloc.patch
Description: Binary data
