On 25/10/21 12:39, David Woodhouse wrote:
Not every single time, only if the cache is absent, stale or not
initialized.
Hm, my reading of it suggests that it will fail even when the cache is
valid, on IOMEM PFNs for which pfn_valid() is not set:
if (pfn_valid(pfn)) {
page = pfn_to_page(pfn);
if (atomic)
hva = kmap_atomic(page);
else
hva = kmap(page);
#ifdef CONFIG_HAS_IOMEM
} else if (!atomic) {
hva = memremap(pfn_to_hpa(pfn), PAGE_SIZE,
MEMREMAP_WB);
} else {
return -EINVAL;
#endif
}
Yeah, you're right. That's the "if" above.
For this use case I'm not even sure why I'd *want* to cache the PFN and
explicitly kmap/memremap it, when surely by *definition* there's a
perfectly serviceable HVA which already points to it?
The point of the gfn_to_pfn cache would be to know in advance that there
won't be a page fault in atomic context. You certainly don't want to
memremap/memunmap it here, it would be awfully slow, but pulling the
kmap/memremap to the MMU notifier would make sense.
Paolo
