On Wed, 30 Sept 2026 at 23:38, Uladzislau Rezki <[email protected]> wrote:
>
> On Wed, Sep 23, 2026 at 02:28:27PM +0800, Wen Jiang wrote:
> > From: "Barry Song (Xiaomi)" <[email protected]>
> >
> > Allow arch_vmap_pte_range_map_size to batch across multiple CONT_PTE
> > blocks, reducing both PTE setup and TLB flush iterations.
> >
> > For CONT_PTE_SIZE-aligned ranges, return a mapping size that may cover
> > multiple CONT_PTE blocks, capped below PMD_SIZE. These sizes are vmalloc
> > mapping spans, not HugeTLB hstate sizes.
> >
> > Signed-off-by: Barry Song (Xiaomi) <[email protected]>
> > Signed-off-by: Wen Jiang <[email protected]>
> > Tested-by: Xueyuan Chen <[email protected]>
> > Tested-by: Leo Yan <[email protected]>
> > ---
> > arch/arm64/include/asm/vmalloc.h | 8 +++++++-
> > 1 file changed, 7 insertions(+), 1 deletion(-)
> >
> > diff --git a/arch/arm64/include/asm/vmalloc.h
> > b/arch/arm64/include/asm/vmalloc.h
> > index 4ec1acd3c1b34..4053b1ec1902e 100644
> > --- a/arch/arm64/include/asm/vmalloc.h
> > +++ b/arch/arm64/include/asm/vmalloc.h
> > @@ -23,10 +23,14 @@ static inline unsigned long
> > arch_vmap_pte_range_map_size(unsigned long addr,
> > unsigned long end, u64 pfn,
> > unsigned int max_page_shift)
> > {
> > + unsigned long size;
> > +
> > /*
> > * If the block is at least CONT_PTE_SIZE in size, and is naturally
> > * aligned in both virtual and physical space, then we can pte-map the
> > * block using the PTE_CONT bit for more efficient use of the TLB.
> > + * The returned mapping size may cover multiple CONT_PTE_SIZE blocks,
> > + * capped below PMD_SIZE.
> > */
> > if (max_page_shift < CONT_PTE_SHIFT)
> > return PAGE_SIZE;
> > @@ -40,7 +44,9 @@ static inline unsigned long
> > arch_vmap_pte_range_map_size(unsigned long addr,
> > if (!IS_ALIGNED(PFN_PHYS(pfn), CONT_PTE_SIZE))
> > return PAGE_SIZE;
> >
> > - return CONT_PTE_SIZE;
> > + size = min3(end - addr, 1UL << max_page_shift, PMD_SIZE >> 1);
> >
> Capped below PMD_SIZE? It is half of PMD_SIZE. Where is that limitation
> comes from? I think the commit message should be improved in that sense.
>
Hi Uladzislau,
You're right, and the "half" is a leftover that can be dropped entirely.
Before this series decoupled vmalloc from the HugeTLB, the value
returned here was fed into ilog2, which reconstructs pagesize as
1UL << shift. That forced the size to be a power of two, and PMD_SIZE >> 1
was then the largest power of two strictly below PMD_SIZE,
How that pte_set_huge() takes the size directly, the size only needs
CONT_PTE_SIZE alignment. And this function is only ever called for
a range that the caller has already confined to a single PMD via
pmd_addr_end(), so no explicit PMD cap is needed at all. I'll
simplify it to:
size = min2(end -addr, 1UL << max_page_shift);
and update the commit message accordingly.
Thanks,
Wen
