On Thu, 2018-03-22 at 10:14 -0600, Jens Axboe wrote:
> Hi Bart,
>
> After your commit:
>
> commit 0ffccc21fcd67d1e1d2a360e90f3fe8efc0d6b52
> Author: Bart Van Assche <bart.vanassche@xxxxxxx>
> Date: Thu Mar 8 13:41:36 2018 -0800
>
> Improve Valgrind instrumentation of memory allocations
>
> running valgrind on a fio spews a lot of warnings:
>
> ==14331== Invalid write of size 8
> ==14331== at 0x4C3451F: memset (in /usr/lib/valgrind/vgpreload_memcheck-amd64-linux.so)
> ==14331== by 0x4595F8: memset (string3.h:90)
> ==14331== by 0x4595F8: smalloc_pool (smalloc.c:434)
> ==14331== by 0x4595F8: __smalloc (smalloc.c:452)
> ==14331== by 0x4727CD: flow_init (flow.c:112)
> ==14331== by 0x421886: setup_thread_area (init.c:391)
> ==14331== by 0x42552A: get_new_job (init.c:468)
> ==14331== by 0x42552A: __parse_jobs_ini (init.c:1928)
> ==14331== by 0x42585F: parse_jobs_ini (init.c:2082)
> [...]
>
> Basically off any path that ends up in smalloc, on the memset()
> that we do:
>
> ptr = __smalloc_pool(pool, alloc_size);
> if (ptr) {
> struct block_hdr *hdr = ptr;
>
> hdr->size = alloc_size;
> fill_redzone(hdr);
>
> ptr += sizeof(*hdr);
> memset(ptr, 0, size);
> ^^^
> }
>
> for every alloc. From configure:
>
> Valgrind headers yes
That's weird. The only explanation I can think of is that Valgrind failed
to intercept the mmap() call in add_pool(). Can you check with the
Valgrind option --trace-syscalls=yes whether Valgrind is able on your
system to intercept mmap()?
Thanks,
Bart.
��.n��������+%������w��{.n�������^n�r������&��z�ޗ�zf���h���~����������_��+v���)ߣ�
