Re: [PATCH 2/4] PM: hibernate: improve robustness of mapping pages in the direct map

[Mike Rapoport]

On Tue, Oct 27, 2020 at 10:44:21PM +0000, Edgecombe, Rick P wrote:
> On Tue, 2020-10-27 at 10:49 +0200, Mike Rapoport wrote:
> > On Mon, Oct 26, 2020 at 06:57:32PM +0000, Edgecombe, Rick P wrote:
> > > On Mon, 2020-10-26 at 11:15 +0200, Mike Rapoport wrote:
> > > > On Mon, Oct 26, 2020 at 12:38:32AM +0000, Edgecombe, Rick P
> > > > wrote:
> > > > > On Sun, 2020-10-25 at 12:15 +0200, Mike Rapoport wrote:
> > > > > > From: Mike Rapoport <rppt@xxxxxxxxxxxxx>
> > > > > > 
> > > > > > When DEBUG_PAGEALLOC or ARCH_HAS_SET_DIRECT_MAP is enabled a
> > > > > > page
> > > > > > may
> > > > > > be
> > > > > > not present in the direct map and has to be explicitly mapped
> > > > > > before
> > > > > > it
> > > > > > could be copied.
> > > > > > 
> > > > > > On arm64 it is possible that a page would be removed from the
> > > > > > direct
> > > > > > map
> > > > > > using set_direct_map_invalid_noflush() but
> > > > > > __kernel_map_pages()
> > > > > > will
> > > > > > refuse
> > > > > > to map this page back if DEBUG_PAGEALLOC is disabled.
> > > > > 
> > > > > It looks to me that arm64 __kernel_map_pages() will still
> > > > > attempt
> > > > > to
> > > > > map it if rodata_full is true, how does this happen?
> > > > 
> > > > Unless I misread the code, arm64 requires both rodata_full and
> > > > debug_pagealloc_enabled() to be true for __kernel_map_pages() to
> > > > do
> > > > anything.
> > > > But rodata_full condition applies to set_direct_map_*_noflush()
> > > > as
> > > > well,
> > > > so with !rodata_full the linear map won't be ever changed.
> > > 
> > > Hmm, looks to me that __kernel_map_pages() will only skip it if
> > > both
> > > debug pagealloc and rodata_full are false.
> > > 
> > > But now I'm wondering if maybe we could simplify things by just
> > > moving
> > > the hibernate unmapped page logic off of the direct map. On x86,
> > > text_poke() used to use this reserved fixmap pte thing that it
> > > could
> > > rely on to remap memory with. If hibernate had some separate pte
> > > for
> > > remapping like that, then we could not have any direct map
> > > restrictions
> > > caused by it/kernel_map_pages(), and it wouldn't have to worry
> > > about
> > > relying on anything else.
> > 
> > Well, there is map_kernel_range() that can be used by hibernation as
> > there is no requirement for particular virtual address, but that
> > would
> > be quite costly if done for every page.
> > 
> > Maybe we can do somthing like
> > 
> > 	if (kernel_page_present(s_page)) {
> > 		do_copy_page(dst, page_address(s_page));
> > 	} else {
> > 		map_kernel_range_noflush(page_address(page), PAGE_SIZE,
> > 					 PROT_READ, &page);
> > 		do_copy_page(dst, page_address(s_page));
> > 		unmap_kernel_range_noflush(page_address(page),
> > PAGE_SIZE);
> > 	}
> > 
> > But it seems that a prerequisite for changing the way a page is
> > mapped
> > in safe_copy_page() would be to teach hibernation that a mapping here
> > may fail.
> > 
> Yea that is what I meant, the direct map could still be used for mapped
> pages.
> 
> But for the unmapped case it could have a pre-setup 4k pte for some non
> direct map address. Then just change the pte to point to any unmapped
> direct map page that was encountered. The point would be to give
> hibernate some 4k pte of its own to manipulate so that it can't fail.
> 
> Yet another option would be have hibernate_map_page() just map large
> pages if it finds them.
> 
> So we could teach hibernate to handle mapping failures, OR we could
> change it so it doesn't rely on direct map page sizes in order to
> succeed. The latter seems better to me since there isn't a reason why
> it should have to fail and the resulting logic might be simpler. Both
> seem like improvements in robustness though.

That's correct, but as the purpose of this series is to prevent usage of
__kernel_map_pages() outside DEBUG_PAGALLOC, for now I'm going to update this
patch changelog and add a comment to hibernate_map_page().

-- 
Sincerely yours,
Mike.