s390/gmap: improve THP splitting (ba925fa3) · Commits · 戴 / test

During s390_enable_sie(), we need to take care of splitting all qemu user process THP mappings. This is currently done with follow_page(FOLL_SPLIT), by simply iterating over all vma ranges, with PAGE_SIZE increment. This logic is sub-optimal and can result in a lot of unnecessary overhead, especially when using qemu and ASAN with large shadow map. Ilya reported significant system slow-down with one CPU busy for a long time and overall unresponsiveness. Fix this by using walk_page_vma() and directly calling split_huge_pmd() only for present pmds, which greatly reduces overhead. Cc: <stable@vger.kernel.org> # v5.4+ Reported-by:

Ilya Leoshkevich <iii@linux.ibm.com> Tested-by:

Ilya Leoshkevich <iii@linux.ibm.com> Acked-by:

Christian Borntraeger <borntraeger@de.ibm.com> Signed-off-by:

Gerald Schaefer <gerald.schaefer@linux.ibm.com> Signed-off-by:

Heiko Carstens <hca@linux.ibm.com>

arch/s390/mm/gmap.c

+20 −7

Original line number	Diff line number	Diff line
		@@ -2485,23 +2485,36 @@ void gmap_sync_dirty_log_pmd(struct gmap *gmap, unsigned long bitmap[4],
		}
		EXPORT_SYMBOL_GPL(gmap_sync_dirty_log_pmd);

		#ifdef CONFIG_TRANSPARENT_HUGEPAGE
		static int thp_split_walk_pmd_entry(pmd_t *pmd, unsigned long addr,
		unsigned long end, struct mm_walk *walk)
		{
		struct vm_area_struct *vma = walk->vma;

		split_huge_pmd(vma, pmd, addr);
		return 0;
		}

		static const struct mm_walk_ops thp_split_walk_ops = {
		.pmd_entry = thp_split_walk_pmd_entry,
		};

		static inline void thp_split_mm(struct mm_struct *mm)
		{
		#ifdef CONFIG_TRANSPARENT_HUGEPAGE
		struct vm_area_struct *vma;
		unsigned long addr;

		for (vma = mm->mmap; vma != NULL; vma = vma->vm_next) {
		for (addr = vma->vm_start;
		addr < vma->vm_end;
		addr += PAGE_SIZE)
		follow_page(vma, addr, FOLL_SPLIT);
		vma->vm_flags &= ~VM_HUGEPAGE;
		vma->vm_flags \|= VM_NOHUGEPAGE;
		walk_page_vma(vma, &thp_split_walk_ops, NULL);
		}
		mm->def_flags \|= VM_NOHUGEPAGE;
		#endif
		}
		#else
		static inline void thp_split_mm(struct mm_struct *mm)
		{
		}
		#endif /* CONFIG_TRANSPARENT_HUGEPAGE */

		/*
		* Remove all empty zero pages from the mapping for lazy refaulting

Admin message