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, 7 insertions, 0 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);
 	}