On Sat, Mar 28, 2020 at 9:52 AM Christoph Hellwig <hch@xxxxxx> wrote: > > On Fri, Mar 27, 2020 at 07:46:08PM +0100, Alexander Potapenko wrote: > > > > Do you mean comparing the address to TASK_SIZE, or is there a more > > > > portable way to check that? > > > > > > !PageHighMem(page) implies the page has a kernel direct mapping. > > > > I tried adding this check and started seeing false positives because > > the virtio_ring driver actually uses highmem pages for DMA, and data > > from those pages is later copied to the kernel. > > Guess it's easier to just allow handling highmem pages? What problems > > do you anticipate? > > For PageHighMem(page), page_address(page) is not actually valid, so І'm > not sure how your code in this patch even worked at all. Note that > all drivers (well, except for a few buggy legacy ones with workarounds) > can DMA from/to highmem. Hm, skipping PageHighMem pages works now, I must've been doing something wrong. Thanks, will add that check to v6. I found a bunch of other places in mm/kmsan/kmsan_shadow.c where we're using page_address(), but metadata pages are not supposed to reside in high memory. -- Alexander Potapenko Software Engineer Google Germany GmbH Erika-Mann-Straße, 33 80636 München Geschäftsführer: Paul Manicle, Halimah DeLaine Prado Registergericht und -nummer: Hamburg, HRB 86891 Sitz der Gesellschaft: Hamburg
