On Tue, Aug 09, 2022 at 04:45:32PM +0800, Huang, Ying wrote:
> Peter Xu <peterx@xxxxxxxxxx> writes:
>
> > On Thu, Aug 04, 2022 at 03:40:57PM -0700, Nadav Amit wrote:
> >> On Aug 4, 2022, at 1:39 PM, Peter Xu <peterx@xxxxxxxxxx> wrote:
> >> > +
> >> > static inline bool is_pfn_swap_entry(swp_entry_t entry);
> >> >
> >> > /* Clear all flags but only keep swp_entry_t related information */
> >> > @@ -265,6 +285,57 @@ static inline swp_entry_t make_writable_migration_entry(pgoff_t offset)
> >> > return swp_entry(SWP_MIGRATION_WRITE, offset);
> >> > }
> >> >
> >> > +/*
> >> > + * Returns whether the host has large enough swap offset field to support
> >> > + * carrying over pgtable A/D bits for page migrations. The result is
> >> > + * pretty much arch specific.
> >> > + */
> >> > +static inline bool migration_entry_supports_ad(void)
> >> > +{
> >> > + /*
> >> > + * max_swapfile_size() returns the max supported swp-offset plus 1.
> >> > + * We can support the migration A/D bits iff the pfn swap entry has
> >> > + * the offset large enough to cover all of them (PFN, A & D bits).
> >> > + */
> >> > +#ifdef CONFIG_SWAP
> >> > + return max_swapfile_size() >= (1UL << SWP_MIG_TOTAL_BITS);
> >>
> >> This is an actual a function call (unless LTO has some trick). A bit of a
> >> shame it cannot be at least memoized.
> >>
> >> Or at least mark max_swapfile_size() as __attribute_const__ so it would not
> >> be called twice for make_migration_entry_young() and
> >> make_migration_entry_dirty().
> >
> > I didn't take too much effort on this one since we're on swap path and I
> > assumed that's not a super hot path. But __attribute_const__ sounds good
> > and easy to get, thanks.
> >
> > Perhaps I should mark it on migration_entry_supports_ad() as a whole? Note
> > that unfortunately SWP_MIG_TOTAL_BITS may not be a const either (see how
> > that define roots back to MAX_PHYSMEM_BITS, where on x86_64 it needs to
> > check 5-lvl).
>
> I think it's possible to memorize max_swapfile_size() or
> migration_entry_supports_ad(). Although they are not constant, they are
> not changed after initialized. The challenge is to find a clean way to
> initialize it.
I checked it up today, the conclusion is I think it's safe we initialize ad
bits for migration in swapfile_init() and I'll do that in the next version.
Longer proof material below.
Generic max_swapfile_size() is pretty much a constant except x86. x86 has
two dependency, on (1) X86_BUG_L1TF, and (2) l1tf_mitigation.
The other challenge is the reference to MAX_PHYSMEM_BITS, which is also
only complicated on x86 with 5 level page tables.
Luckily the cpu bits are all setup within early_cpu_init(). The setup of
l1tf_mitigation variable is later but still earlier than most of the init
calls (which swapfile_init belongs to level 4).
A full graph for reference:
- start_kernel
- setup_arch
- early_cpu_init
- get_cpu_cap --> fetch from CPUID (including X86_FEATURE_LA57)
- early_identify_cpu --> setup X86_BUG_L1TF
- early_identify_cpu --> clear X86_FEATURE_LA57 (if early lvl5 not enabled (USE_EARLY_PGTABLE_L5))
- parse_early_param
- l1tf_cmdline --> set l1tf_mitigation
- check_bugs
- l1tf_select_mitigation --> set l1tf_mitigation
- arch_call_rest_init
- rest_init
- kernel_init
- kernel_init_freeable
- do_basic_setup
- do_initcalls --> calls swapfile_init() (initcall level 4)
I'll add one (or maybe >1?) patch on top in next post to optimize the
fetching of these states.
Thanks,
--
Peter Xu
