Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit a741e679 authored by Benjamin Herrenschmidt's avatar Benjamin Herrenschmidt Committed by Paul Mackerras
Browse files

[POWERPC] Make tlb flush batch use lazy MMU mode 



The current tlb flush code on powerpc 64 bits has a subtle race since we
lost the page table lock due to the possible faulting in of new PTEs
after a previous one has been removed but before the corresponding hash
entry has been evicted, which can leads to all sort of fatal problems.

This patch reworks the batch code completely. It doesn't use the mmu_gather
stuff anymore. Instead, we use the lazy mmu hooks that were added by the
paravirt code. They have the nice property that the enter/leave lazy mmu
mode pair is always fully contained by the PTE lock for a given range
of PTEs. Thus we can guarantee that all batches are flushed on a given
CPU before it drops that lock.

We also generalize batching for any PTE update that require a flush.

Batching is now enabled on a CPU by arch_enter_lazy_mmu_mode() and
disabled by arch_leave_lazy_mmu_mode(). The code epects that this is
always contained within a PTE lock section so no preemption can happen
and no PTE insertion in that range from another CPU. When batching
is enabled on a CPU, every PTE updates that need a hash flush will
use the batch for that flush.

Signed-off-by: default avatarBenjamin Herrenschmidt <benh@kernel.crashing.org>
Signed-off-by: default avatarPaul Mackerras <paulus@samba.org>
parent e4ee3891

arch/powerpc/kernel/process.c

+1 −3
Original line number Original line Diff line number Diff line
@@ -305,9 +305,7 @@ struct task_struct *__switch_to(struct task_struct *prev, 
		set_dabr(new->thread.dabr);

		set_dabr(new->thread.dabr);

		__get_cpu_var(current_dabr) = new->thread.dabr;

		__get_cpu_var(current_dabr) = new->thread.dabr;

	}

	}



#endif /* CONFIG_PPC64 */

	flush_tlb_pending();

#endif





	new_thread = &new->thread;

	new_thread = &new->thread;

	old_thread = &current->thread;

	old_thread = &current->thread;

arch/powerpc/kernel/smp.c

+0 −4
Original line number Original line Diff line number Diff line
@@ -428,10 +428,6 @@ void generic_mach_cpu_die(void) 
	smp_wmb();

	smp_wmb();

	while (__get_cpu_var(cpu_state) != CPU_UP_PREPARE)

	while (__get_cpu_var(cpu_state) != CPU_UP_PREPARE)

		cpu_relax();

		cpu_relax();



#ifdef CONFIG_PPC64

	flush_tlb_pending();

#endif

	cpu_set(cpu, cpu_online_map);

	cpu_set(cpu, cpu_online_map);

	local_irq_enable();

	local_irq_enable();

}

}

arch/powerpc/mm/hugetlbpage.c

+5 −11
Original line number Original line Diff line number Diff line
@@ -316,12 +316,11 @@ void set_huge_pte_at(struct mm_struct *mm, unsigned long addr, 
{

{

	if (pte_present(*ptep)) {

	if (pte_present(*ptep)) {

		/* We open-code pte_clear because we need to pass the right

		/* We open-code pte_clear because we need to pass the right

		 * argument to hpte_update (huge / !huge)

		 * argument to hpte_need_flush (huge / !huge). Might not be

		 * necessary anymore if we make hpte_need_flush() get the

		 * page size from the slices

		 */

		 */

		unsigned long old = pte_update(ptep, ~0UL);

		pte_update(mm, addr & HPAGE_MASK, ptep, ~0UL, 1);

		if (old & _PAGE_HASHPTE)

			hpte_update(mm, addr & HPAGE_MASK, ptep, old, 1);

		flush_tlb_pending();

	}

	}

	*ptep = __pte(pte_val(pte) & ~_PAGE_HPTEFLAGS);

	*ptep = __pte(pte_val(pte) & ~_PAGE_HPTEFLAGS);

}

}
@@ -329,12 +328,7 @@ void set_huge_pte_at(struct mm_struct *mm, unsigned long addr, 
pte_t huge_ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,

pte_t huge_ptep_get_and_clear(struct mm_struct *mm, unsigned long addr,

			      pte_t *ptep)

			      pte_t *ptep)

{

{

	unsigned long old = pte_update(ptep, ~0UL);

	unsigned long old = pte_update(mm, addr, ptep, ~0UL, 1);



	if (old & _PAGE_HASHPTE)

		hpte_update(mm, addr & HPAGE_MASK, ptep, old, 1);

	*ptep = __pte(0);



	return __pte(old);

	return __pte(old);

}

}

arch/powerpc/mm/tlb_64.c

+43 −25
Original line number Original line Diff line number Diff line
@@ -120,17 +120,20 @@ void pgtable_free_tlb(struct mmu_gather *tlb, pgtable_free_t pgf) 
}

}





/*

/*

 * Update the MMU hash table to correspond with a change to

 * A linux PTE was changed and the corresponding hash table entry

 * a Linux PTE.  If wrprot is true, it is permissible to

 * neesd to be flushed. This function will either perform the flush

 * change the existing HPTE to read-only rather than removing it

 * immediately or will batch it up if the current CPU has an active

 * (if we remove it we should clear the _PTE_HPTEFLAGS bits).

 * batch on it.

 *

 * Must be called from within some kind of spinlock/non-preempt region...

 */

 */

void hpte_update(struct mm_struct *mm, unsigned long addr,

void hpte_need_flush(struct mm_struct *mm, unsigned long addr,

		     pte_t *ptep, unsigned long pte, int huge)

		     pte_t *ptep, unsigned long pte, int huge)

{

{

	struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);

	struct ppc64_tlb_batch *batch = &__get_cpu_var(ppc64_tlb_batch);

	unsigned long vsid;

	unsigned long vsid, vaddr;

	unsigned int psize;

	unsigned int psize;

	real_pte_t rpte;

	int i;

	int i;





	i = batch->index;

	i = batch->index;
@@ -151,6 +154,26 @@ void hpte_update(struct mm_struct *mm, unsigned long addr, 
	} else

	} else

		psize = pte_pagesize_index(pte);

		psize = pte_pagesize_index(pte);





	/* Build full vaddr */

	if (!is_kernel_addr(addr)) {

		vsid = get_vsid(mm->context.id, addr);

		WARN_ON(vsid == 0);

	} else

		vsid = get_kernel_vsid(addr);

	vaddr = (vsid << 28 ) | (addr & 0x0fffffff);

	rpte = __real_pte(__pte(pte), ptep);



	/*

	 * Check if we have an active batch on this CPU. If not, just

	 * flush now and return. For now, we don global invalidates

	 * in that case, might be worth testing the mm cpu mask though

	 * and decide to use local invalidates instead...

	 */

	if (!batch->active) {

		flush_hash_page(vaddr, rpte, psize, 0);

		return;

	}



	/*

	/*

	 * This can happen when we are in the middle of a TLB batch and

	 * This can happen when we are in the middle of a TLB batch and

	 * we encounter memory pressure (eg copy_page_range when it tries

	 * we encounter memory pressure (eg copy_page_range when it tries
@@ -162,47 +185,42 @@ void hpte_update(struct mm_struct *mm, unsigned long addr, 
	 * batch

	 * batch

	 */

	 */

	if (i != 0 && (mm != batch->mm || batch->psize != psize)) {

	if (i != 0 && (mm != batch->mm || batch->psize != psize)) {

		flush_tlb_pending();

		__flush_tlb_pending(batch);

		i = 0;

		i = 0;

	}

	}

	if (i == 0) {

	if (i == 0) {

		batch->mm = mm;

		batch->mm = mm;

		batch->psize = psize;

		batch->psize = psize;

	}

	}

	if (!is_kernel_addr(addr)) {

	batch->pte[i] = rpte;

		vsid = get_vsid(mm->context.id, addr);

	batch->vaddr[i] = vaddr;

		WARN_ON(vsid == 0);

	} else

		vsid = get_kernel_vsid(addr);

	batch->vaddr[i] = (vsid << 28 ) | (addr & 0x0fffffff);

	batch->pte[i] = __real_pte(__pte(pte), ptep);

	batch->index = ++i;

	batch->index = ++i;

	if (i >= PPC64_TLB_BATCH_NR)

	if (i >= PPC64_TLB_BATCH_NR)

		flush_tlb_pending();

		__flush_tlb_pending(batch);

}

}





/*

 * This function is called when terminating an mmu batch or when a batch

 * is full. It will perform the flush of all the entries currently stored

 * in a batch.

 *

 * Must be called from within some kind of spinlock/non-preempt region...

 */

void __flush_tlb_pending(struct ppc64_tlb_batch *batch)

void __flush_tlb_pending(struct ppc64_tlb_batch *batch)

{

{

	int i;

	int cpu;

	cpumask_t tmp;

	cpumask_t tmp;

	int local = 0;

	int i, local = 0;





	BUG_ON(in_interrupt());



	cpu = get_cpu();

	i = batch->index;

	i = batch->index;

	tmp = cpumask_of_cpu(cpu);

	tmp = cpumask_of_cpu(smp_processor_id());

	if (cpus_equal(batch->mm->cpu_vm_mask, tmp))

	if (cpus_equal(batch->mm->cpu_vm_mask, tmp))

		local = 1;

		local = 1;



	if (i == 1)

	if (i == 1)

		flush_hash_page(batch->vaddr[0], batch->pte[0],

		flush_hash_page(batch->vaddr[0], batch->pte[0],

				batch->psize, local);

				batch->psize, local);

	else

	else

		flush_hash_range(i, local);

		flush_hash_range(i, local);

	batch->index = 0;

	batch->index = 0;

	put_cpu();

}

}





void pte_free_finish(void)

void pte_free_finish(void)

include/asm-powerpc/pgtable.h

+16 −34
Original line number Original line Diff line number Diff line
@@ -272,7 +272,10 @@ static inline pte_t pte_mkhuge(pte_t pte) { 
	return pte; }

	return pte; }





/* Atomic PTE updates */

/* Atomic PTE updates */

static inline unsigned long pte_update(pte_t *p, unsigned long clr)

static inline unsigned long pte_update(struct mm_struct *mm,

				       unsigned long addr,

				       pte_t *ptep, unsigned long clr,

				       int huge)

{

{

	unsigned long old, tmp;

	unsigned long old, tmp;
@@ -283,20 +286,15 @@ static inline unsigned long pte_update(pte_t *p, unsigned long clr) 
	andc	%1,%0,%4 \n\

	andc	%1,%0,%4 \n\

	stdcx.	%1,0,%3 \n\

	stdcx.	%1,0,%3 \n\

	bne-	1b"

	bne-	1b"

	: "=&r" (old), "=&r" (tmp), "=m" (*p)

	: "=&r" (old), "=&r" (tmp), "=m" (*ptep)

	: "r" (p), "r" (clr), "m" (*p), "i" (_PAGE_BUSY)

	: "r" (ptep), "r" (clr), "m" (*ptep), "i" (_PAGE_BUSY)

	: "cc" );

	: "cc" );



	if (old & _PAGE_HASHPTE)

		hpte_need_flush(mm, addr, ptep, old, huge);

	return old;

	return old;

}

}





/* PTE updating functions, this function puts the PTE in the

 * batch, doesn't actually triggers the hash flush immediately,

 * you need to call flush_tlb_pending() to do that.

 * Pass -1 for "normal" size (4K or 64K)

 */

extern void hpte_update(struct mm_struct *mm, unsigned long addr,

			pte_t *ptep, unsigned long pte, int huge);



static inline int __ptep_test_and_clear_young(struct mm_struct *mm,

static inline int __ptep_test_and_clear_young(struct mm_struct *mm,

					      unsigned long addr, pte_t *ptep)

					      unsigned long addr, pte_t *ptep)

{

{
@@ -304,11 +302,7 @@ static inline int __ptep_test_and_clear_young(struct mm_struct *mm, 




       	if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)

       	if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)

		return 0;

		return 0;

	old = pte_update(ptep, _PAGE_ACCESSED);

	old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0);

	if (old & _PAGE_HASHPTE) {

		hpte_update(mm, addr, ptep, old, 0);

		flush_tlb_pending();

	}

	return (old & _PAGE_ACCESSED) != 0;

	return (old & _PAGE_ACCESSED) != 0;

}

}

#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG

#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
@@ -331,9 +325,7 @@ static inline int __ptep_test_and_clear_dirty(struct mm_struct *mm, 




       	if ((pte_val(*ptep) & _PAGE_DIRTY) == 0)

       	if ((pte_val(*ptep) & _PAGE_DIRTY) == 0)

		return 0;

		return 0;

	old = pte_update(ptep, _PAGE_DIRTY);

	old = pte_update(mm, addr, ptep, _PAGE_DIRTY, 0);

	if (old & _PAGE_HASHPTE)

		hpte_update(mm, addr, ptep, old, 0);

	return (old & _PAGE_DIRTY) != 0;

	return (old & _PAGE_DIRTY) != 0;

}

}

#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY

#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_DIRTY
@@ -352,9 +344,7 @@ static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, 




       	if ((pte_val(*ptep) & _PAGE_RW) == 0)

       	if ((pte_val(*ptep) & _PAGE_RW) == 0)

       		return;

       		return;

	old = pte_update(ptep, _PAGE_RW);

	old = pte_update(mm, addr, ptep, _PAGE_RW, 0);

	if (old & _PAGE_HASHPTE)

		hpte_update(mm, addr, ptep, old, 0);

}

}





/*

/*
@@ -378,7 +368,6 @@ static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, 
({									\

({									\

	int __dirty = __ptep_test_and_clear_dirty((__vma)->vm_mm, __address, \

	int __dirty = __ptep_test_and_clear_dirty((__vma)->vm_mm, __address, \

						  __ptep); 		\

						  __ptep); 		\

	flush_tlb_page(__vma, __address);				\

	__dirty;							\

	__dirty;							\

})

})
@@ -386,20 +375,14 @@ static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, 
static inline pte_t ptep_get_and_clear(struct mm_struct *mm,

static inline pte_t ptep_get_and_clear(struct mm_struct *mm,

				       unsigned long addr, pte_t *ptep)

				       unsigned long addr, pte_t *ptep)

{

{

	unsigned long old = pte_update(ptep, ~0UL);

	unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0);



	if (old & _PAGE_HASHPTE)

		hpte_update(mm, addr, ptep, old, 0);

	return __pte(old);

	return __pte(old);

}

}





static inline void pte_clear(struct mm_struct *mm, unsigned long addr,

static inline void pte_clear(struct mm_struct *mm, unsigned long addr,

			     pte_t * ptep)

			     pte_t * ptep)

{

{

	unsigned long old = pte_update(ptep, ~0UL);

	pte_update(mm, addr, ptep, ~0UL, 0);



	if (old & _PAGE_HASHPTE)

		hpte_update(mm, addr, ptep, old, 0);

}

}





/*

/*
@@ -408,10 +391,8 @@ static inline void pte_clear(struct mm_struct *mm, unsigned long addr, 
static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,

static inline void set_pte_at(struct mm_struct *mm, unsigned long addr,

			      pte_t *ptep, pte_t pte)

			      pte_t *ptep, pte_t pte)

{

{

	if (pte_present(*ptep)) {

	if (pte_present(*ptep))

		pte_clear(mm, addr, ptep);

		pte_clear(mm, addr, ptep);

		flush_tlb_pending();

	}

	pte = __pte(pte_val(pte) & ~_PAGE_HPTEFLAGS);

	pte = __pte(pte_val(pte) & ~_PAGE_HPTEFLAGS);

	*ptep = pte;

	*ptep = pte;

}

}
@@ -522,6 +503,7 @@ void pgtable_cache_init(void); 
	return pt;

	return pt;

}

}







#include <asm-generic/pgtable.h>

#include <asm-generic/pgtable.h>





#endif /* __ASSEMBLY__ */

#endif /* __ASSEMBLY__ */