On 11/23/20 1:45 PM, Toke Høiland-Jørgensen wrote:
andreas@rammhold.de writes:
While testing this patch I asked myself if there are actually still valid users of non-zeroed memory and if we shouldn't just change the default to be zeroed? That would also remove the need to patch every single call site.
Well there are other callers that already perform manual clearing, so just changing sl_alloc() would lead to double-clearing there. Not a huge issue, I suppose, but I'll leave it up to the maintainers to decide :)
An allocated structure from slab is typically intended to be used in hotpaths with fast alloc and dealloc. After going multithreaded, it may have some impact on performance. I think this is more of a documentation issue, rather than a reason to introduce double-clearing. Maria