Re: [PATCH OLK-6.6 v2 01/19] mm/hugetlb: Introduce hugetlb_insert_hugepage_pte[_by_pa]

On 2023/12/7 21:43, Wang Wensheng wrote:

ascend inclusion category: feature bugzilla: https://gitee.com/openeuler/kernel/issues/I8LNGH

---

Add hugetlb_insert_hugepage_pte[_by_pa] to insert hugepages into process page table. The by_pa version performs like remap_pfn_range() that make the pte special and can be used for reserved physical memory.

Signed-off-by: Wang Wensheng wangwensheng4@huawei.com

include/linux/hugetlb.h | 32 +++++++++++++++++++ mm/Kconfig | 15 +++++++++ mm/hugetlb.c | 70 +++++++++++++++++++++++++++++++++++++++++ 3 files changed, 117 insertions(+)

diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h index 47d25a5e1933..57c2630cb80d 100644 --- a/include/linux/hugetlb.h +++ b/include/linux/hugetlb.h @@ -282,6 +282,26 @@ long hugetlb_change_protection(struct vm_area_struct *vma, bool is_hugetlb_entry_migration(pte_t pte); void hugetlb_unshare_all_pmds(struct vm_area_struct *vma);

+#ifdef CONFIG_HUGETLB_INSERT_PAGE +int hugetlb_insert_hugepage_pte(struct mm_struct *mm, unsigned long addr,

• 		pgprot_t prot, struct page *hpage);
  

+int hugetlb_insert_hugepage_pte_by_pa(struct mm_struct *mm,

• 		unsigned long vir_addr,
  

• 		pgprot_t prot, unsigned long phy_addr);
  

+#else /* CONFIG_HUGETLB_INSERT_PAGE */ +static inline int hugetlb_insert_hugepage_pte(struct mm_struct *mm, unsigned long addr,

• 		pgprot_t prot, struct page *hpage)
  

+{

• return -EPERM;

+} +static inline int hugetlb_insert_hugepage_pte_by_pa(struct mm_struct *mm,

• 		unsigned long vir_addr,
  

• 		pgprot_t prot, unsigned long phy_addr)
  

+{

• return -EPERM;

+} +#endif /* CONFIG_HUGETLB_INSERT_PAGE */

• #else /* !CONFIG_HUGETLB_PAGE */

  static inline void hugetlb_dup_vma_private(struct vm_area_struct *vma)

@@ -491,6 +511,18 @@ static inline vm_fault_t hugetlb_fault(struct mm_struct *mm,

static inline void hugetlb_unshare_all_pmds(struct vm_area_struct *vma) { }

+static inline int hugetlb_insert_hugepage_pte(struct mm_struct *mm, unsigned long addr,

• 		pgprot_t prot, struct page *hpage)
  

+{

• return -EPERM;

+} +static inline int hugetlb_insert_hugepage_pte_by_pa(struct mm_struct *mm,

• 		unsigned long vir_addr,
  

• 		pgprot_t prot, unsigned long phy_addr)
  

+{

• return -EPERM;

+}

• #endif /* !CONFIG_HUGETLB_PAGE */ /*
  • hugepages at page global directory. If arch support

diff --git a/mm/Kconfig b/mm/Kconfig index ece4f2847e2b..3dead7328cd5 100644 --- a/mm/Kconfig +++ b/mm/Kconfig @@ -1269,6 +1269,21 @@ config LOCK_MM_AND_FIND_VMA bool depends on !STACK_GROWSUP

+menuconfig ASCEND_FEATURES

• bool "Support Ascend Features"
• depends on ARM64
• select HUGETLB_INSERT_PAGE
• help

•  The Ascend chip use the Hisilicon DaVinci architecture, and mainly
  

•  focus on AI and machine leanring area, contains many external features.
  

•  Enable this config to enable selective list of these features.
  

•  If unsure, say N
  

+config HUGETLB_INSERT_PAGE

• bool

• help

•  This allowed a driver to insert hugetlb mapping into user address space.
  

• source "mm/damon/Kconfig"

  endmenu

diff --git a/mm/hugetlb.c b/mm/hugetlb.c index 1a4d388b6a3b..a148584422d9 100644 --- a/mm/hugetlb.c +++ b/mm/hugetlb.c @@ -7552,3 +7552,73 @@ static void __init hugetlb_cma_check(void) }

#endif /* CONFIG_CMA */

+#ifdef CONFIG_HUGETLB_INSERT_PAGE +static pte_t *hugetlb_huge_pte_alloc(struct mm_struct *mm, unsigned long addr,

• 		     unsigned long size)
  

+{

• pgd_t *pgdp;
• p4d_t *p4dp;
• pud_t *pudp;
• pte_t *ptep = NULL;
• pgdp = pgd_offset(mm, addr);
• p4dp = p4d_offset(pgdp, addr);
• pudp = pud_alloc(mm, p4dp, addr);
• if (!pudp)

• return NULL;
  

• ptep = (pte_t *)pmd_alloc(mm, pudp, addr);

这里参考huge_pte_alloc

吧下面的pte的检查放到这里； pteval = ptep_get_lockless if （xxx） warn xxx 如果要warn，要释放上面的内存；

• return ptep;

+}

+static int __hugetlb_insert_hugepage(struct mm_struct *mm, unsigned long addr,

• 		     pgprot_t prot, unsigned long pfn)
  

+{

• int ret = 0;
• pte_t *ptep, entry;
• struct hstate *h;
• spinlock_t *ptl;
• h = size_to_hstate(PMD_SIZE);
• if (!h)

• return -EINVAL;
  

• ptep = hugetlb_huge_pte_alloc(mm, addr, huge_page_size(h));
• if (!ptep)

• return -ENXIO;
  

这里只要返回 nomem吧；

• if (WARN_ON(ptep && !pte_none(*ptep) && !pmd_huge(*(pmd_t *)ptep)))

• return -ENXIO;
  

移动到上面

• entry = pfn_pte(pfn, prot);
• entry = huge_pte_mkdirty(entry);
• if (!(pgprot_val(prot) & PTE_RDONLY))

• entry = huge_pte_mkwrite(entry);
  

• entry = pte_mkyoung(entry);
• entry = pte_mkhuge(entry);

这里要arch_make_huge_pte

• entry = pte_mkspecial(entry);

这个special有点特殊，最好加下注释；

• ptl = huge_pte_lockptr(h, mm, ptep);
• spin_lock(ptl);
• set_huge_pte_at(mm, addr, ptep, entry, PMD_SIZE);
• spin_unlock(ptl);
• return ret;

+}

+int hugetlb_insert_hugepage_pte(struct mm_struct *mm, unsigned long addr,

• 		pgprot_t prot, struct page *hpage)
  

+{

• return __hugetlb_insert_hugepage(mm, addr, prot, page_to_pfn(hpage));

+} +EXPORT_SYMBOL_GPL(hugetlb_insert_hugepage_pte);

+int hugetlb_insert_hugepage_pte_by_pa(struct mm_struct *mm, unsigned long addr,

• 		      pgprot_t prot, unsigned long phy_addr)
  

+{

• return __hugetlb_insert_hugepage(mm, addr, prot, phy_addr >> PAGE_SHIFT);

+} +EXPORT_SYMBOL_GPL(hugetlb_insert_hugepage_pte_by_pa); +#endif /* CONFIG_HUGETLB_INSERT_PAGE */