On Tue, Aug 11, 2020 at 07:24:57PM +0100, Jessica Clarke wrote:
> IA-64 is special and treats pgd_offset_k differently from pgd_offset by
> not including the region number, and init_mm's PGD is such that it only
> points to the kernel's region's PGD. This was broken in 974b9b2c68 which
> unified the two and therefore included the region number, causing it to
> go way out of bounds of the kernel's PGD, which made the kernel hang
> during early boot. Thus, permit pgd_offset_k to be overridden like the
> other macros and override it on IA-64 with the old implementation. Also
> update the comment to clarify that this is not just an optimisation but
> a required implementation detail.
If I may suggest:
IA-64 is special and treats pgd_offset_k() differently from pgd_offset() by
using different formulas to calculate index into kernel and user PGD
tables. The index into user PGDs takes into account the region number
and the index into the kernel (init_mm) PGD always presumes predefined
kernel region number. Commit 974b9b2c68 ("mm: consolidate pte_index()
and pte_offset_*() definitions") made IA-64 to use generic
pgd_offset_k() which wrongly used pgd_index() for user page tables. As
the result, the index into kernel PGD was going out of bounds and the
kernel hang during early boot.
Allow overrides of pgd_offset_k() and use an override on IA-64 with the
old implementation that will correctly index kernel PGD.
> Fixes: 974b9b2c68 ("mm: consolidate pte_index() and pte_offset_*() definitions")
> Reported-by: John Paul Adrian Glaubitz <glaubitz@xxxxxxxxxxxxxxxxxxx>
> Signed-off-by: Jessica Clarke <jrtc27@xxxxxxxxxx>
> Tested-by: John Paul Adrian Glaubitz <glaubitz@xxxxxxxxxxxxxxxxxxx>
Thanks for the fix, I don't insist on the changelog update, so with the
nit below
Reviewed-by: Mike Rapoport <rppt@xxxxxxxxxxxxx>
> ---
> Changes since v1:
> * Fixed typo in commit message
> * Slightly reworded commit message to sound less weird
> * Included Adrian's Tested-by
>
> arch/ia64/include/asm/pgtable.h | 8 ++++++++
> include/linux/pgtable.h | 2 ++
> 2 files changed, 10 insertions(+)
>
> diff --git a/arch/ia64/include/asm/pgtable.h b/arch/ia64/include/asm/pgtable.h
> index 10850897a91c..2ac2199d99ce 100644
> --- a/arch/ia64/include/asm/pgtable.h
> +++ b/arch/ia64/include/asm/pgtable.h
> @@ -366,6 +366,14 @@ pgd_index (unsigned long address)
> }
> #define pgd_index pgd_index
>
> +/*
> + * In the kernel's mapped region we know everything is in region number 5, so
> + * as an optimisation its PGD already points to the area for that region, but
> + * that means not adding the region here is required, not just an optimisation.
> + */
How about:
/*
* In the kernel's mapped region we know everything is in region number 5, so
* as an optimisation its PGD already points to the area for that region.
* However, this also means that we cannot use pgd_index() and we never
* should add the region here.
*/
> +#define pgd_offset_k(addr) \
> + (init_mm.pgd + (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1)))
> +
> /* Look up a pgd entry in the gate area. On IA-64, the gate-area
> resides in the kernel-mapped segment, hence we use pgd_offset_k()
> here. */
> diff --git a/include/linux/pgtable.h b/include/linux/pgtable.h
> index 53e97da1e8e2..73c64fe098ba 100644
> --- a/include/linux/pgtable.h
> +++ b/include/linux/pgtable.h
> @@ -117,7 +117,9 @@ static inline pgd_t *pgd_offset_pgd(pgd_t *pgd, unsigned long address)
> * a shortcut which implies the use of the kernel's pgd, instead
> * of a process's
> */
> +#ifndef pgd_offset_k
> #define pgd_offset_k(address) pgd_offset(&init_mm, (address))
> +#endif
>
> /*
> * In many cases it is known that a virtual address is mapped at PMD or PTE
> --
> 2.23.0
>
--
Sincerely yours,
Mike.
