On Mon, Jul 13, 2020 at 3:07 PM Christoph Hellwig <h...@infradead.org> wrote: > On Fri, Jul 10, 2020 at 08:55:42PM -0700, Palmer Dabbelt wrote: > > +ssize_t copy_oldmem_page(unsigned long pfn, char *buf, > > + size_t csize, unsigned long offset, > > + int userbuf) > > +{ > > + void *vaddr; > > + > > + if (!csize) > > + return 0; > > + > > + vaddr = memremap(__pfn_to_phys(pfn), PAGE_SIZE, MEMREMAP_WB); > > + if (!vaddr) > > + return -ENOMEM; > > Doing a memremap for every page is very inefficient. Also I don't see > why you'd want to even do that. All memory is in the direct mapping > for RISC-V. For other architecture that support highmem kmap_atomic_pfn > would do the job, which is what I'd use in a generic version.
I would expect the 'oldmem' data to not have a 'struct page', which would be a problem at least for the generic implementation of kmap_atomic_pfn() include/linux/highmem.h:#define kmap_atomic_pfn(pfn) kmap_atomic(pfn_to_page(pfn)) kmap_atomic() might still work with a bogus page pointer if it only transforms it back into a pfn, but some implementations just have this one that cannot work: static inline void *page_address(const struct page *page) { return page->virtual; } I have not checked how the crash dump code works though, maybe it does work after all. Arnd