On Wed, 5 Apr 2023 21:05:15 -0700 John Hubbard jhubbard@nvidia.com wrote:
Although CONFIG_DEVICE_PRIVATE and hmm_range_fault() and related functionality was first developed on x86, it also works on arm64. However, when trying this out on an arm64 system, it turns out that there is a massive slowdown during the setup and teardown phases.
This slowdown is due to lots of calls to WARN_ON()'s that are checking for pages that are out of the physical range for the CPU. However, that's a design feature of device private pages: they are specfically chosen in order to be outside of the range of the CPU's true physical pages.
...
--- a/arch/arm64/mm/mmu.c +++ b/arch/arm64/mm/mmu.c @@ -1157,8 +1157,10 @@ int __meminit vmemmap_check_pmd(pmd_t *pmdp, int node, int __meminit vmemmap_populate(unsigned long start, unsigned long end, int node, struct vmem_altmap *altmap) { +/* Device private pages are outside of the CPU's physical page range. */ +#ifndef CONFIG_DEVICE_PRIVATE WARN_ON((start < VMEMMAP_START) || (end > VMEMMAP_END));
For a simple expression like this to cause a "massive slowdown", I assume the WARN is triggering. But changelog doesn't mention massive dmesg spewage?
Given Ard's comments, perhaps a switch to WARN_ON_ONCE() would suit?