Re: Boot failures with "mm/sparse: Remove CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER" on powerpc (was Re: mmotm 2018-07-10-16-50 uploaded)
- Date: Wed, 11 Jul 2018 21:12:25 +0800
- From: Baoquan He <bhe@xxxxxxxxxx>
- Subject: Re: Boot failures with "mm/sparse: Remove CONFIG_SPARSEMEM_ALLOC_MEM_MAP_TOGETHER" on powerpc (was Re: mmotm 2018-07-10-16-50 uploaded)
On 07/11/18 at 10:49pm, Michael Ellerman wrote:
> akpm@xxxxxxxxxxxxxxxxxxxx writes:
> > The mm-of-the-moment snapshot 2018-07-10-16-50 has been uploaded to
> > http://www.ozlabs.org/~akpm/mmotm/
> > * mm-sparse-add-a-static-variable-nr_present_sections.patch
> > * mm-sparsemem-defer-the-ms-section_mem_map-clearing.patch
> > * mm-sparsemem-defer-the-ms-section_mem_map-clearing-fix.patch
> > * mm-sparse-add-a-new-parameter-data_unit_size-for-alloc_usemap_and_memmap.patch
> > * mm-sparse-optimize-memmap-allocation-during-sparse_init.patch
> > * mm-sparse-optimize-memmap-allocation-during-sparse_init-checkpatch-fixes.patch
> > * mm-sparse-remove-config_sparsemem_alloc_mem_map_together.patch
> This seems to be breaking my powerpc pseries qemu boots.
> The boot log with some extra debug shows eg:
> $ make pseries_le_defconfig
> $ qemu-system-ppc64 -nographic -vga none -M pseries -m 2G -kernel vmlinux
> vmemmap_populate f000000000000000..f000000000024000, node 0
> * f000000000000000..f000000001000000 allocated at c000000076000000
> hash__vmemmap_create_mapping: start 0xf000000000000000 size 0x1000000 phys 0x76000000
> hash__vmemmap_create_mapping: failed -1
> <repeated many times>
> Then there's lots of other warnings about bad page states and eventually
> a NULL deref and we panic().
> The problem seems to be that we're calling down into
> hash__vmemmap_create_mapping() for every call to vmemmap_populate(),
> whereas previously we would only call hash__vmemmap_create_mapping()
> once because our vmemmap_populated() would return true.
> There's actually a comment in sparse_init() that says:
> * powerpc need to call sparse_init_one_section right after each
> * sparse_early_mem_map_alloc, so allocate usemap_map at first.
> So changing that behaviour does seem to be the problem.
> I assume that comment is talking about the fact that we use pfn_valid()
> in vmemmap_populated().
> I'm not clear on how to fix it though.
Have you tried reverting that patch and building kernel to test again?
Does it work?