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>

4 files changed, 44 insertions, 130 deletions

diff --git a/arch/parisc/include/asm/mmu_context.h b/arch/parisc/include/asm/mmu_context.h
index 46f8c22c5977..726257648d9f 100644
--- a/arch/parisc/include/asm/mmu_context.h
+++ b/arch/parisc/include/asm/mmu_context.h
@@ -5,6 +5,7 @@
 #include <linux/mm.h>
 #include <linux/sched.h>
 #include <linux/atomic.h>
+#include <linux/spinlock.h>
 #include <asm-generic/mm_hooks.h>
 
 /* on PA-RISC, we actually have enough contexts to justify an allocator
@@ -50,6 +51,12 @@ static inline void switch_mm_irqs_off(struct mm_struct *prev,
 		struct mm_struct *next, struct task_struct *tsk)
 {
 	if (prev != next) {
+#ifdef CONFIG_TLB_PTLOCK
+		/* put physical address of page_table_lock in cr28 (tr4)
+		   for TLB faults */
+		spinlock_t *pgd_lock = &next->page_table_lock;
+		mtctl(__pa(__ldcw_align(&pgd_lock->rlock.raw_lock)), 28);
+#endif
 		mtctl(__pa(next->pgd), 25);
 		load_context(next->context);
 	}
diff --git a/arch/parisc/include/asm/page.h b/arch/parisc/include/asm/page.h
index 6b3f6740a6a6..d00313d1274e 100644
--- a/arch/parisc/include/asm/page.h
+++ b/arch/parisc/include/asm/page.h
@@ -112,7 +112,7 @@ extern int npmem_ranges;
 #else
 #define BITS_PER_PTE_ENTRY	2
 #define BITS_PER_PMD_ENTRY	2
-#define BITS_PER_PGD_ENTRY	BITS_PER_PMD_ENTRY
+#define BITS_PER_PGD_ENTRY	2
 #endif
 #define PGD_ENTRY_SIZE	(1UL << BITS_PER_PGD_ENTRY)
 #define PMD_ENTRY_SIZE	(1UL << BITS_PER_PMD_ENTRY)
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) \
diff --git a/arch/parisc/include/asm/pgtable.h b/arch/parisc/include/asm/pgtable.h
index 75cf84070fc9..39017210dbf0 100644
--- a/arch/parisc/include/asm/pgtable.h
+++ b/arch/parisc/include/asm/pgtable.h
@@ -23,8 +23,6 @@
 #include <asm/processor.h>
 #include <asm/cache.h>
 
-static inline spinlock_t *pgd_spinlock(pgd_t *);
-
 /*
  * kern_addr_valid(ADDR) tests if ADDR is pointing to valid kernel
  * memory.  For the return value to be meaningful, ADDR must be >=
@@ -42,12 +40,8 @@ static inline spinlock_t *pgd_spinlock(pgd_t *);
 
 /* This is for the serialization of PxTLB broadcasts. At least on the N class
  * systems, only one PxTLB inter processor broadcast can be active at any one
- * time on the Merced bus.
-
- * PTE updates are protected by locks in the PMD.
- */
+ * time on the Merced bus. */
 extern spinlock_t pa_tlb_flush_lock;
-extern spinlock_t pa_swapper_pg_lock;
 #if defined(CONFIG_64BIT) && defined(CONFIG_SMP)
 extern int pa_serialize_tlb_flushes;
 #else
@@ -86,18 +80,16 @@ static inline void purge_tlb_entries(struct mm_struct *mm, unsigned long addr)
  * within a page table are directly modified.  Thus, the following
  * hook is made available.
  */
-#define set_pte(pteptr, pteval)                                 \
-        do{                                                     \
-                *(pteptr) = (pteval);                           \
-        } while(0)
-
-#define set_pte_at(mm, addr, ptep, pteval)			\
-	do {							\
-		unsigned long flags;				\
-		spin_lock_irqsave(pgd_spinlock((mm)->pgd), flags);\
-		set_pte(ptep, pteval);				\
-		purge_tlb_entries(mm, addr);			\
-		spin_unlock_irqrestore(pgd_spinlock((mm)->pgd), flags);\
+#define set_pte(pteptr, pteval)			\
+	do {					\
+		*(pteptr) = (pteval);		\
+		barrier();			\
+	} while(0)
+
+#define set_pte_at(mm, addr, pteptr, pteval)	\
+	do {					\
+		*(pteptr) = (pteval);		\
+		purge_tlb_entries(mm, addr);	\
 	} while (0)
 
 #endif /* !__ASSEMBLY__ */
@@ -120,12 +112,10 @@ static inline void purge_tlb_entries(struct mm_struct *mm, unsigned long addr)
 #define KERNEL_INITIAL_SIZE	(1 << KERNEL_INITIAL_ORDER)
 
 #if CONFIG_PGTABLE_LEVELS == 3
-#define PGD_ORDER	1 /* Number of pages per pgd */
-#define PMD_ORDER	1 /* Number of pages per pmd */
-#define PGD_ALLOC_ORDER	(2 + 1) /* first pgd contains pmd */
+#define PMD_ORDER	1
+#define PGD_ORDER	0
 #else
-#define PGD_ORDER	1 /* Number of pages per pgd */
-#define PGD_ALLOC_ORDER	(PGD_ORDER + 1)
+#define PGD_ORDER	1
 #endif
 
 /* Definitions for 3rd level (we use PLD here for Page Lower directory
@@ -240,11 +230,9 @@ static inline void purge_tlb_entries(struct mm_struct *mm, unsigned long addr)
  * able to effectively address 40/42/44-bits of physical address space
  * depending on 4k/16k/64k PAGE_SIZE */
 #define _PxD_PRESENT_BIT   31
-#define _PxD_ATTACHED_BIT  30
-#define _PxD_VALID_BIT     29
+#define _PxD_VALID_BIT     30
 
 #define PxD_FLAG_PRESENT  (1 << xlate_pabit(_PxD_PRESENT_BIT))
-#define PxD_FLAG_ATTACHED (1 << xlate_pabit(_PxD_ATTACHED_BIT))
 #define PxD_FLAG_VALID    (1 << xlate_pabit(_PxD_VALID_BIT))
 #define PxD_FLAG_MASK     (0xf)
 #define PxD_FLAG_SHIFT    (4)
@@ -326,23 +314,10 @@ extern unsigned long *empty_zero_page;
 #define pgd_flag(x)	(pgd_val(x) & PxD_FLAG_MASK)
 #define pgd_address(x)	((unsigned long)(pgd_val(x) &~ PxD_FLAG_MASK) << PxD_VALUE_SHIFT)
 
-#if CONFIG_PGTABLE_LEVELS == 3
-/* The first entry of the permanent pmd is not there if it contains
- * the gateway marker */
-#define pmd_none(x)	(!pmd_val(x) || pmd_flag(x) == PxD_FLAG_ATTACHED)
-#else
 #define pmd_none(x)	(!pmd_val(x))
-#endif
 #define pmd_bad(x)	(!(pmd_flag(x) & PxD_FLAG_VALID))
 #define pmd_present(x)	(pmd_flag(x) & PxD_FLAG_PRESENT)
 static inline void pmd_clear(pmd_t *pmd) {
-#if CONFIG_PGTABLE_LEVELS == 3
-	if (pmd_flag(*pmd) & PxD_FLAG_ATTACHED)
-		/* This is the entry pointing to the permanent pmd
-		 * attached to the pgd; cannot clear it */
-		set_pmd(pmd, __pmd(PxD_FLAG_ATTACHED));
-	else
-#endif
 		set_pmd(pmd,  __pmd(0));
 }
 
@@ -358,12 +333,6 @@ static inline void pmd_clear(pmd_t *pmd) {
 #define pud_bad(x)      (!(pud_flag(x) & PxD_FLAG_VALID))
 #define pud_present(x)  (pud_flag(x) & PxD_FLAG_PRESENT)
 static inline void pud_clear(pud_t *pud) {
-#if CONFIG_PGTABLE_LEVELS == 3
-	if(pud_flag(*pud) & PxD_FLAG_ATTACHED)
-		/* This is the permanent pmd attached to the pud; cannot
-		 * free it */
-		return;
-#endif
 	set_pud(pud, __pud(0));
 }
 #endif
@@ -456,32 +425,18 @@ extern void update_mmu_cache(struct vm_area_struct *, unsigned long, pte_t *);
 #define __pte_to_swp_entry(pte)		((swp_entry_t) { pte_val(pte) })
 #define __swp_entry_to_pte(x)		((pte_t) { (x).val })
 
-
-static inline spinlock_t *pgd_spinlock(pgd_t *pgd)
-{
-	if (unlikely(pgd == swapper_pg_dir))
-		return &pa_swapper_pg_lock;
-	return (spinlock_t *)((char *)pgd + (PAGE_SIZE << (PGD_ALLOC_ORDER - 1)));
-}
-
-
 static inline int ptep_test_and_clear_young(struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
 {
 	pte_t pte;
-	unsigned long flags;
 
 	if (!pte_young(*ptep))
 		return 0;
 
-	spin_lock_irqsave(pgd_spinlock(vma->vm_mm->pgd), flags);
 	pte = *ptep;
 	if (!pte_young(pte)) {
-		spin_unlock_irqrestore(pgd_spinlock(vma->vm_mm->pgd), flags);
 		return 0;
 	}
-	set_pte(ptep, pte_mkold(pte));
-	purge_tlb_entries(vma->vm_mm, addr);
-	spin_unlock_irqrestore(pgd_spinlock(vma->vm_mm->pgd), flags);
+	set_pte_at(vma->vm_mm, addr, ptep, pte_mkold(pte));
 	return 1;
 }
 
@@ -489,24 +444,16 @@ struct mm_struct;
 static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 {
 	pte_t old_pte;
-	unsigned long flags;
 
-	spin_lock_irqsave(pgd_spinlock(mm->pgd), flags);
 	old_pte = *ptep;
-	set_pte(ptep, __pte(0));
-	purge_tlb_entries(mm, addr);
-	spin_unlock_irqrestore(pgd_spinlock(mm->pgd), flags);
+	set_pte_at(mm, addr, ptep, __pte(0));
 
 	return old_pte;
 }
 
 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
 {
-	unsigned long flags;
-	spin_lock_irqsave(pgd_spinlock(mm->pgd), flags);
-	set_pte(ptep, pte_wrprotect(*ptep));
-	purge_tlb_entries(mm, addr);
-	spin_unlock_irqrestore(pgd_spinlock(mm->pgd), flags);
+	set_pte_at(mm, addr, ptep, pte_wrprotect(*ptep));
 }
 
 #define pte_same(A,B)	(pte_val(A) == pte_val(B))