Merge patch series "svnapot fixes"

Alexandre Ghiti <alexghiti@rivosinc.com> says:

While merging riscv napot and arm64 contpte support, I noticed we did
not abide by the specification which states that we should clear a
napot mapping before setting a new one, called "break before make" in
arm64 (patch 1). And also that we did not add the new hugetlb page size
added by napot in hugetlb_mask_last_page() (patch 2).

* b4-shazam-merge:
  riscv: Fix hugetlb_mask_last_page() when NAPOT is enabled
  riscv: Fix set_huge_pte_at() for NAPOT mapping

Link: https://lore.kernel.org/r/20240117195741.1926459-1-alexghiti@rivosinc.com
Signed-off-by: Palmer Dabbelt <palmer@rivosinc.com>

Author: palmer-dabbelt

arch/riscv/mm/hugetlbpage.c

@@ -125,6 +125,26 @@ pte_t *huge_pte_offset(struct mm_struct *mm,
 	return pte;
 }
 
+unsigned long hugetlb_mask_last_page(struct hstate *h)
+{
+	unsigned long hp_size = huge_page_size(h);
+
+	switch (hp_size) {
+#ifndef __PAGETABLE_PMD_FOLDED
+	case PUD_SIZE:
+		return P4D_SIZE - PUD_SIZE;
+#endif
+	case PMD_SIZE:
+		return PUD_SIZE - PMD_SIZE;
+	case napot_cont_size(NAPOT_CONT64KB_ORDER):
+		return PMD_SIZE - napot_cont_size(NAPOT_CONT64KB_ORDER);
+	default:
+		break;
+	}
+
+	return 0UL;
+}
+
 static pte_t get_clear_contig(struct mm_struct *mm,
 			      unsigned long addr,
 			      pte_t *ptep,
@@ -177,13 +197,36 @@ pte_t arch_make_huge_pte(pte_t entry, unsigned int shift, vm_flags_t flags)
 	return entry;
 }
 
+static void clear_flush(struct mm_struct *mm,
+			unsigned long addr,
+			pte_t *ptep,
+			unsigned long pgsize,
+			unsigned long ncontig)
+{
+	struct vm_area_struct vma = TLB_FLUSH_VMA(mm, 0);
+	unsigned long i, saddr = addr;
+
+	for (i = 0; i < ncontig; i++, addr += pgsize, ptep++)
+		ptep_get_and_clear(mm, addr, ptep);
+
+	flush_tlb_range(&vma, saddr, addr);
+}
+
+/*
+ * When dealing with NAPOT mappings, the privileged specification indicates that
+ * "if an update needs to be made, the OS generally should first mark all of the
+ * PTEs invalid, then issue SFENCE.VMA instruction(s) covering all 4 KiB regions
+ * within the range, [...] then update the PTE(s), as described in Section
+ * 4.2.1.". That's the equivalent of the Break-Before-Make approach used by
+ * arm64.
+ */
 void set_huge_pte_at(struct mm_struct *mm,
 		     unsigned long addr,
 		     pte_t *ptep,
 		     pte_t pte,
 		     unsigned long sz)
 {
-	unsigned long hugepage_shift;
+	unsigned long hugepage_shift, pgsize;
 	int i, pte_num;
 
 	if (sz >= PGDIR_SIZE)
@@ -198,7 +241,22 @@ void set_huge_pte_at(struct mm_struct *mm,
 		hugepage_shift = PAGE_SHIFT;
 
 	pte_num = sz >> hugepage_shift;
-	for (i = 0; i < pte_num; i++, ptep++, addr += (1 << hugepage_shift))
+	pgsize = 1 << hugepage_shift;
+
+	if (!pte_present(pte)) {
+		for (i = 0; i < pte_num; i++, ptep++, addr += pgsize)
+			set_ptes(mm, addr, ptep, pte, 1);
+		return;
+	}
+
+	if (!pte_napot(pte)) {
+		set_ptes(mm, addr, ptep, pte, 1);
+		return;
+	}
+
+	clear_flush(mm, addr, ptep, pgsize, pte_num);
+
+	for (i = 0; i < pte_num; i++, ptep++, addr += pgsize)
 		set_pte_at(mm, addr, ptep, pte);
 }