parisc: Optimize per-pagetable spinlocks

On parisc a spinlock is stored in the next page behind the pgd which
protects against parallel accesses to the pgd. That's why one additional
page (PGD_ALLOC_ORDER) is allocated for the pgd.

Matthew Wilcox suggested that we instead should use a pointer in the
struct page table for this spinlock and noted, that the comments for the
PGD_ORDER and PMD_ORDER defines were wrong.

Both suggestions are addressed with this patch. Instead of having an own
spinlock to protect the pgd, we now switch to use the existing
page_table_lock.  Additionally, beside loading the pgd into cr25 in
switch_mm_irqs_off(), the physical address of this lock is loaded into
cr28 (tr4), so that we can avoid implementing a complicated lookup in
assembly for this lock in the TLB fault handlers.

The existing Hybrid L2/L3 page table scheme (where the pmd is adjacent
to the pgd) has been dropped with this patch.

Remove the locking in set_pte() and the huge-page pte functions too.
They trigger a spinlock recursion on 32bit machines and seem unnecessary.

Suggested-by: Matthew Wilcox <willy@infradead.org>
Fixes: b37d1c1898b2 ("parisc: Use per-pagetable spinlock")
Signed-off-by: John David Anglin <dave.anglin@bell.net>
Signed-off-by: Helge Deller <deller@gmx.de>

1 files changed, 18 insertions, 58 deletions

diff --git a/arch/parisc/include/asm/pgalloc.h b/arch/parisc/include/asm/pgalloc.h
index a6482b2ce0ea..dda557085311 100644
--- a/arch/parisc/include/asm/pgalloc.h
+++ b/arch/parisc/include/asm/pgalloc.h
@@ -15,47 +15,23 @@
 #define __HAVE_ARCH_PGD_FREE
 #include <asm-generic/pgalloc.h>
 
-/* Allocate the top level pgd (page directory)
- *
- * Here (for 64 bit kernels) we implement a Hybrid L2/L3 scheme: we
- * allocate the first pmd adjacent to the pgd.  This means that we can
- * subtract a constant offset to get to it.  The pmd and pgd sizes are
- * arranged so that a single pmd covers 4GB (giving a full 64-bit
- * process access to 8TB) so our lookups are effectively L2 for the
- * first 4GB of the kernel (i.e. for all ILP32 processes and all the
- * kernel for machines with under 4GB of memory) */
+/* Allocate the top level pgd (page directory) */
 static inline pgd_t *pgd_alloc(struct mm_struct *mm)
 {
-	pgd_t *pgd = (pgd_t *)__get_free_pages(GFP_KERNEL,
-					       PGD_ALLOC_ORDER);
-	pgd_t *actual_pgd = pgd;
+	pgd_t *pgd;
 
-	if (likely(pgd != NULL)) {
-		memset(pgd, 0, PAGE_SIZE<<PGD_ALLOC_ORDER);
-#if CONFIG_PGTABLE_LEVELS == 3
-		actual_pgd += PTRS_PER_PGD;
-		/* Populate first pmd with allocated memory.  We mark it
-		 * with PxD_FLAG_ATTACHED as a signal to the system that this
-		 * pmd entry may not be cleared. */
-		set_pgd(actual_pgd, __pgd((PxD_FLAG_PRESENT |
-				        PxD_FLAG_VALID |
-					PxD_FLAG_ATTACHED)
-			+ (__u32)(__pa((unsigned long)pgd) >> PxD_VALUE_SHIFT)));
-		/* The first pmd entry also is marked with PxD_FLAG_ATTACHED as
-		 * a signal that this pmd may not be freed */
-		set_pgd(pgd, __pgd(PxD_FLAG_ATTACHED));
-#endif
-	}
-	spin_lock_init(pgd_spinlock(actual_pgd));
-	return actual_pgd;
+	pgd = (pgd_t *) __get_free_pages(GFP_KERNEL, PGD_ORDER);
+	if (unlikely(pgd == NULL))
+		return NULL;
+
+	memset(pgd, 0, PAGE_SIZE << PGD_ORDER);
+
+	return pgd;
 }
 
 static inline void pgd_free(struct mm_struct *mm, pgd_t *pgd)
 {
-#if CONFIG_PGTABLE_LEVELS == 3
-	pgd -= PTRS_PER_PGD;
-#endif
-	free_pages((unsigned long)pgd, PGD_ALLOC_ORDER);
+	free_pages((unsigned long)pgd, PGD_ORDER);
 }
 
 #if CONFIG_PGTABLE_LEVELS == 3
@@ -70,41 +46,25 @@ static inline void pud_populate(struct mm_struct *mm, pud_t *pud, pmd_t *pmd)
 
 static inline pmd_t *pmd_alloc_one(struct mm_struct *mm, unsigned long address)
 {
-	return (pmd_t *)__get_free_pages(GFP_PGTABLE_KERNEL, PMD_ORDER);
+	pmd_t *pmd;
+
+	pmd = (pmd_t *)__get_free_pages(GFP_PGTABLE_KERNEL, PMD_ORDER);
+	if (likely(pmd))
+		memset ((void *)pmd, 0, PAGE_SIZE << PMD_ORDER);
+	return pmd;
 }
 
 static inline void pmd_free(struct mm_struct *mm, pmd_t *pmd)
 {
-	if (pmd_flag(*pmd) & PxD_FLAG_ATTACHED) {
-		/*
-		 * This is the permanent pmd attached to the pgd;
-		 * cannot free it.
-		 * Increment the counter to compensate for the decrement
-		 * done by generic mm code.
-		 */
-		mm_inc_nr_pmds(mm);
-		return;
-	}
 	free_pages((unsigned long)pmd, PMD_ORDER);
 }
-
 #endif
 
 static inline void
 pmd_populate_kernel(struct mm_struct *mm, pmd_t *pmd, pte_t *pte)
 {
-#if CONFIG_PGTABLE_LEVELS == 3
-	/* preserve the gateway marker if this is the beginning of
-	 * the permanent pmd */
-	if(pmd_flag(*pmd) & PxD_FLAG_ATTACHED)
-		set_pmd(pmd, __pmd((PxD_FLAG_PRESENT |
-				PxD_FLAG_VALID |
-				PxD_FLAG_ATTACHED)
-			+ (__u32)(__pa((unsigned long)pte) >> PxD_VALUE_SHIFT)));
-	else
-#endif
-		set_pmd(pmd, __pmd((PxD_FLAG_PRESENT | PxD_FLAG_VALID)
-			+ (__u32)(__pa((unsigned long)pte) >> PxD_VALUE_SHIFT)));
+	set_pmd(pmd, __pmd((PxD_FLAG_PRESENT | PxD_FLAG_VALID)
+		+ (__u32)(__pa((unsigned long)pte) >> PxD_VALUE_SHIFT)));
 }
 
 #define pmd_populate(mm, pmd, pte_page) \