Revisiting "too small to fail"
Revisiting "too small to fail"
Posted May 23, 2017 9:43 UTC (Tue) by dvrabel (subscriber, #9500)In reply to: Revisiting "too small to fail" by corbet
Parent article: Revisiting "too small to fail"
32 KiB doesn't seem like a "small" allocation to me -- anything more than PAGE_SIZE can fail (or trigger the OOM killer) due to fragmentation. Perhaps the thing to do here is to reduce the size that is considered small, until small is a page or less?
By way of example, the Xen hypervisor has taken considerable effort to ensure all memory allocation are a page or less, or have fallback paths that do so.
