On Mon, Nov 11, 2019 at 11:12:41AM +0100, Janosch Frank wrote: > On 11/4/19 12:29 PM, Paolo Bonzini wrote: > > On 04/11/19 11:54, Andrew Jones wrote: > >> > >> diff --git a/lib/alloc.c b/lib/alloc.c > >> index ecdbbc44dbf9..ed8f5f94c9b0 100644 > >> --- a/lib/alloc.c > >> +++ b/lib/alloc.c > >> @@ -46,15 +46,17 @@ void *memalign(size_t alignment, size_t size) > >> uintptr_t blkalign; > >> uintptr_t mem; > >> > >> + if (!size) > >> + return NULL; > >> + > >> + assert(alignment >= sizeof(void *) && is_power_of_2(alignment)); > >> assert(alloc_ops && alloc_ops->memalign); > >> - if (alignment <= sizeof(uintptr_t)) > >> - alignment = sizeof(uintptr_t); > >> - else > >> - size += alignment - 1; > >> > >> + size += alignment - 1; > >> blkalign = MAX(alignment, alloc_ops->align_min); > >> size = ALIGN(size + METADATA_EXTRA, alloc_ops->align_min); > >> p = alloc_ops->memalign(blkalign, size); > >> + assert(p); > > I had some more time to think about and test this and I think returning > NULL here is more useful. My usecase is a limit test where I allocate > until I get a NULL and then free everything afterwards. I think I prefer the assert here, since most users of this will not be expecting to run out of memory, and therefore not checking for it. I think you may want to just write your own allocator in your unit test that's based on alloc_pages(). Thanks, drew
