Skip to contentSkip to conversation

remove nopfn

PATCH In review Nick Piggin 3 participants 14 messages May 2, 2008 – May 6, 2008

Subsystems (6)
lore
Share & tools
lore
#0 lore raw

cover remove nopfn

Hi,

Can we pretty please get this into the current merge window? All it takes is a quick review and test ;) I think mspec was already tested, but it wouldn't hurt to verify again...

It shaves about .5K off mm/memory.o so it is pretty significant. Mispredicted branches are also actually a significant cost in the fault path which I'm trying to reduce (merging fault with page_mkwrite should help with this further).

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#1 lore raw

PATCH 1/4 mm: allow pfnmap ->fault()s

Take out an assertion to allow ->fault handlers to service PFNMAP regions. This is required to reimplement .nopfn handlers with .fault handlers and subsequently remove nopfn.

---

 mm/memory.c |    2 --
 1 file changed, 2 deletions(-)
mm/memory.c +0 -2
Index: linux-2.6/mm/memory.c
===================================================================
@@ -2275,8 +2275,6 @@ static int __do_fault(struct mm_struct *
 	vmf.flags = flags;
 	vmf.page = NULL;
 
-	BUG_ON(vma->vm_flags & VM_PFNMAP);
-
 	ret = vma->vm_ops->fault(vma, &vmf);
 	if (unlikely(ret & (VM_FAULT_ERROR | VM_FAULT_NOPAGE)))
 		return ret;

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#2 lore raw

PATCH 2/4 mspec: convert nopfn to fault

In reply to Nick Piggin

---

 drivers/char/mspec.c |   22 ++++++++++++++--------
 1 file changed, 14 insertions(+), 8 deletions(-)
drivers/char/mspec.c +14 -8
Index: linux-2.6/drivers/char/mspec.c
===================================================================
@@ -193,25 +193,24 @@ mspec_close(struct vm_area_struct *vma)
 }
 
 /*
- * mspec_nopfn
+ * mspec_fault
  *
  * Creates a mspec page and maps it to user space.
  */
-static unsigned long
-mspec_nopfn(struct vm_area_struct *vma, unsigned long address)
+static int
+mspec_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
 	unsigned long paddr, maddr;
 	unsigned long pfn;
+	pgoff_t index = vmf->pgoff;
 	int index;
 	struct vma_data *vdata = vma->vm_private_data;
 
-	BUG_ON(address < vdata->vm_start || address >= vdata->vm_end);
-	index = (address - vdata->vm_start) >> PAGE_SHIFT;
 	maddr = (volatile unsigned long) vdata->maddr[index];
 	if (maddr == 0) {
 		maddr = uncached_alloc_page(numa_node_id(), 1);
 		if (maddr == 0)
-			return NOPFN_OOM;
+			return VM_FAULT_OOM;
 
 		spin_lock(&vdata->lock);
 		if (vdata->maddr[index] == 0) {
@@ -231,13 +230,20 @@ mspec_nopfn(struct vm_area_struct *vma, 
 
 	pfn = paddr >> PAGE_SHIFT;
 
-	return pfn;
+	/*
+	 * vm_insert_pfn can fail with -EBUSY, but in that case it will
+	 * be because another thread has installed the pte first, so it
+	 * is no problem.
+	 */
+	vm_insert_pfn(vma, (unsigned long)vmf->virtual_address, pfn);
+
+	return VM_FAULT_NOPAGE;
 }
 
 static struct vm_operations_struct mspec_vm_ops = {
 	.open = mspec_open,
 	.close = mspec_close,
-	.nopfn = mspec_nopfn
+	.fault = mspec_fault,
 };
 
 /*

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#3 lore raw

Re: [patch 2/4] mspec: convert nopfn to fault

Sorry, I missed this original post. Saw the add this morning and went back to the archives.

Review

Quoted 9 lines
(quoted diff) +3 -2
-static unsigned long
-mspec_nopfn(struct vm_area_struct *vma, unsigned long address)
+static int
+mspec_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
 	unsigned long paddr, maddr;
 	unsigned long pfn;
+	pgoff_t index = vmf->pgoff;
 	int index;

I think this will cause problems. Two definitions of index. I removed the int index and tested. This appears to work fine. Sorry for the delay.

Thanks,
Robin

drivers/char/mspec.c +0 -1
Index: remove_nopfn/drivers/char/mspec.c
===================================================================
@@ -203,7 +203,6 @@ mspec_fault(struct vm_area_struct *vma, 
 	unsigned long paddr, maddr;
 	unsigned long pfn;
 	pgoff_t index = vmf->pgoff;
-	int index;
 	struct vma_data *vdata = vma->vm_private_data;
 
 	maddr = (volatile unsigned long) vdata->maddr[index];

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#4 lore raw

PATCH 3/4 spufs: convert nopfn to fault

In reply to Nick Piggin

---

 arch/powerpc/platforms/cell/spufs/file.c     |   91 ++++++++++++---------------
 arch/powerpc/platforms/cell/spufs/sputrace.c |    8 +-
 2 files changed, 46 insertions(+), 53 deletions(-)
arch/powerpc/platforms/cell/spufs/file.c +42 -49
Index: linux-2.6/arch/powerpc/platforms/cell/spufs/file.c
===================================================================
@@ -237,11 +237,10 @@ spufs_mem_write(struct file *file, const
 	return size;
 }
 
-static unsigned long spufs_mem_mmap_nopfn(struct vm_area_struct *vma,
-					  unsigned long address)
+static int spufs_mem_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
 	struct spu_context *ctx	= vma->vm_file->private_data;
-	unsigned long pfn, offset, addr0 = address;
+	unsigned long pfn, offset, address = (unsigned long)vmf->virtual_address;
 #ifdef CONFIG_SPU_FS_64K_LS
 	struct spu_state *csa = &ctx->csa;
 	int psize;
@@ -259,15 +258,14 @@ static unsigned long spufs_mem_mmap_nopf
 	}
 #endif /* CONFIG_SPU_FS_64K_LS */
 
-	offset = (address - vma->vm_start) + (vma->vm_pgoff << PAGE_SHIFT);
+	offset = vmf->pgoff << PAGE_SHIFT;
 	if (offset >= LS_SIZE)
-		return NOPFN_SIGBUS;
+		return VM_FAULT_SIGBUS;
 
-	pr_debug("spufs_mem_mmap_nopfn address=0x%lx -> 0x%lx, offset=0x%lx\n",
-		 addr0, address, offset);
+	pr_debug("spufs_mem_mmap_fault address=0x%lx, offset=0x%lx\n", address, offset);
 
 	if (spu_acquire(ctx))
-		return NOPFN_REFAULT;
+		return VM_FAULT_NOPAGE;
 
 	if (ctx->state == SPU_STATE_SAVED) {
 		vma->vm_page_prot = __pgprot(pgprot_val(vma->vm_page_prot)
@@ -278,16 +276,16 @@ static unsigned long spufs_mem_mmap_nopf
 					     | _PAGE_NO_CACHE);
 		pfn = (ctx->spu->local_store_phys + offset) >> PAGE_SHIFT;
 	}
-	vm_insert_pfn(vma, address, pfn);
+	vm_insert_pfn(vma, (unsigned long)vmf->virtual_address, pfn);
 
 	spu_release(ctx);
 
-	return NOPFN_REFAULT;
+	return VM_FAULT_NOPAGE;
 }
 
 
 static struct vm_operations_struct spufs_mem_mmap_vmops = {
-	.nopfn = spufs_mem_mmap_nopfn,
+	.fault = spufs_mem_mmap_fault,
 };
 
 static int spufs_mem_mmap(struct file *file, struct vm_area_struct *vma)
@@ -350,20 +348,19 @@ static const struct file_operations spuf
 #endif
 };
 
-static unsigned long spufs_ps_nopfn(struct vm_area_struct *vma,
-				    unsigned long address,
+static int spufs_ps_fault(struct vm_area_struct *vma,
+				    struct vm_fault *vmf,
 				    unsigned long ps_offs,
 				    unsigned long ps_size)
 {
 	struct spu_context *ctx = vma->vm_file->private_data;
-	unsigned long area, offset = address - vma->vm_start;
+	unsigned long area, offset = vmf->pgoff << PAGE_SHIFT;
 	int ret = 0;
 
-	spu_context_nospu_trace(spufs_ps_nopfn__enter, ctx);
+	spu_context_nospu_trace(spufs_ps_fault__enter, ctx);
 
-	offset += vma->vm_pgoff << PAGE_SHIFT;
 	if (offset >= ps_size)
-		return NOPFN_SIGBUS;
+		return VM_FAULT_SIGBUS;
 
 	/*
 	 * Because we release the mmap_sem, the context may be destroyed while
@@ -377,7 +374,7 @@ static unsigned long spufs_ps_nopfn(stru
 	 * pages to hand out to the user, but we don't want to wait
 	 * with the mmap_sem held.
 	 * It is possible to drop the mmap_sem here, but then we need
-	 * to return NOPFN_REFAULT because the mappings may have
+	 * to return VM_FAULT_NOPAGE because the mappings may have
 	 * hanged.
 	 */
 	if (spu_acquire(ctx))
@@ -385,14 +382,15 @@ static unsigned long spufs_ps_nopfn(stru
 
 	if (ctx->state == SPU_STATE_SAVED) {
 		up_read(&current->mm->mmap_sem);
-		spu_context_nospu_trace(spufs_ps_nopfn__sleep, ctx);
+		spu_context_nospu_trace(spufs_ps_fault__sleep, ctx);
 		ret = spufs_wait(ctx->run_wq, ctx->state == SPU_STATE_RUNNABLE);
-		spu_context_trace(spufs_ps_nopfn__wake, ctx, ctx->spu);
+		spu_context_trace(spufs_ps_fault__wake, ctx, ctx->spu);
 		down_read(&current->mm->mmap_sem);
 	} else {
 		area = ctx->spu->problem_phys + ps_offs;
-		vm_insert_pfn(vma, address, (area + offset) >> PAGE_SHIFT);
-		spu_context_trace(spufs_ps_nopfn__insert, ctx, ctx->spu);
+		vm_insert_pfn(vma, (unsigned long)vmf->virtual_address,
+					(area + offset) >> PAGE_SHIFT);
+		spu_context_trace(spufs_ps_fault__insert, ctx, ctx->spu);
 	}
 
 	if (!ret)
@@ -400,18 +398,18 @@ static unsigned long spufs_ps_nopfn(stru
 
 refault:
 	put_spu_context(ctx);
-	return NOPFN_REFAULT;
+	return VM_FAULT_NOPAGE;
 }
 
 #if SPUFS_MMAP_4K
-static unsigned long spufs_cntl_mmap_nopfn(struct vm_area_struct *vma,
-					   unsigned long address)
+static int spufs_cntl_mmap_fault(struct vm_area_struct *vma,
+					   struct vm_fault *vmf)
 {
-	return spufs_ps_nopfn(vma, address, 0x4000, 0x1000);
+	return spufs_ps_fault(vma, vmf, 0x4000, 0x1000);
 }
 
 static struct vm_operations_struct spufs_cntl_mmap_vmops = {
-	.nopfn = spufs_cntl_mmap_nopfn,
+	.fault = spufs_cntl_mmap_fault,
 };
 
 /*
@@ -1096,23 +1094,22 @@ static ssize_t spufs_signal1_write(struc
 	return 4;
 }
 
-static unsigned long spufs_signal1_mmap_nopfn(struct vm_area_struct *vma,
-					      unsigned long address)
+static int spufs_signal1_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
 #if PAGE_SIZE == 0x1000
-	return spufs_ps_nopfn(vma, address, 0x14000, 0x1000);
+	return spufs_ps_fault(vma, vmf, 0x14000, 0x1000);
 #elif PAGE_SIZE == 0x10000
 	/* For 64k pages, both signal1 and signal2 can be used to mmap the whole
 	 * signal 1 and 2 area
 	 */
-	return spufs_ps_nopfn(vma, address, 0x10000, 0x10000);
+	return spufs_ps_fault(vma, vmf, 0x10000, 0x10000);
 #else
 #error unsupported page size
 #endif
 }
 
 static struct vm_operations_struct spufs_signal1_mmap_vmops = {
-	.nopfn = spufs_signal1_mmap_nopfn,
+	.fault = spufs_signal1_mmap_fault,
 };
 
 static int spufs_signal1_mmap(struct file *file, struct vm_area_struct *vma)
@@ -1233,23 +1230,22 @@ static ssize_t spufs_signal2_write(struc
 }
 
 #if SPUFS_MMAP_4K
-static unsigned long spufs_signal2_mmap_nopfn(struct vm_area_struct *vma,
-					      unsigned long address)
+static int spufs_signal2_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
 #if PAGE_SIZE == 0x1000
-	return spufs_ps_nopfn(vma, address, 0x1c000, 0x1000);
+	return spufs_ps_fault(vma, vmf, 0x1c000, 0x1000);
 #elif PAGE_SIZE == 0x10000
 	/* For 64k pages, both signal1 and signal2 can be used to mmap the whole
 	 * signal 1 and 2 area
 	 */
-	return spufs_ps_nopfn(vma, address, 0x10000, 0x10000);
+	return spufs_ps_fault(vma, vmf, 0x10000, 0x10000);
 #else
 #error unsupported page size
 #endif
 }
 
 static struct vm_operations_struct spufs_signal2_mmap_vmops = {
-	.nopfn = spufs_signal2_mmap_nopfn,
+	.fault = spufs_signal2_mmap_fault,
 };
 
 static int spufs_signal2_mmap(struct file *file, struct vm_area_struct *vma)
@@ -1361,14 +1357,13 @@ DEFINE_SPUFS_ATTRIBUTE(spufs_signal2_typ
 		       spufs_signal2_type_set, "%llu\n", SPU_ATTR_ACQUIRE);
 
 #if SPUFS_MMAP_4K
-static unsigned long spufs_mss_mmap_nopfn(struct vm_area_struct *vma,
-					  unsigned long address)
+static int spufs_mss_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
-	return spufs_ps_nopfn(vma, address, 0x0000, 0x1000);
+	return spufs_ps_fault(vma, vmf, 0x0000, 0x1000);
 }
 
 static struct vm_operations_struct spufs_mss_mmap_vmops = {
-	.nopfn = spufs_mss_mmap_nopfn,
+	.fault = spufs_mss_mmap_fault,
 };
 
 /*
@@ -1423,14 +1418,13 @@ static const struct file_operations spuf
 	.mmap	 = spufs_mss_mmap,
 };
 
-static unsigned long spufs_psmap_mmap_nopfn(struct vm_area_struct *vma,
-					    unsigned long address)
+static int spufs_psmap_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
-	return spufs_ps_nopfn(vma, address, 0x0000, 0x20000);
+	return spufs_ps_fault(vma, vmf, 0x0000, 0x20000);
 }
 
 static struct vm_operations_struct spufs_psmap_mmap_vmops = {
-	.nopfn = spufs_psmap_mmap_nopfn,
+	.fault = spufs_psmap_mmap_fault,
 };
 
 /*
@@ -1483,14 +1477,13 @@ static const struct file_operations spuf
 
 
 #if SPUFS_MMAP_4K
-static unsigned long spufs_mfc_mmap_nopfn(struct vm_area_struct *vma,
-					  unsigned long address)
+static int spufs_mfc_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
-	return spufs_ps_nopfn(vma, address, 0x3000, 0x1000);
+	return spufs_ps_fault(vma, vmf, 0x3000, 0x1000);
 }
 
 static struct vm_operations_struct spufs_mfc_mmap_vmops = {
-	.nopfn = spufs_mfc_mmap_nopfn,
+	.fault = spufs_mfc_mmap_fault,
 };
 
 /*
arch/powerpc/platforms/cell/spufs/sputrace.c +4 -4
Index: linux-2.6/arch/powerpc/platforms/cell/spufs/sputrace.c
===================================================================
@@ -182,10 +182,10 @@ struct spu_probe spu_probes[] = {
 	{ "spu_yield__enter", "ctx %p", spu_context_nospu_event },
 	{ "spu_deactivate__enter", "ctx %p", spu_context_nospu_event },
 	{ "__spu_deactivate__unload", "ctx %p spu %p", spu_context_event },
-	{ "spufs_ps_nopfn__enter", "ctx %p", spu_context_nospu_event },
-	{ "spufs_ps_nopfn__sleep", "ctx %p", spu_context_nospu_event },
-	{ "spufs_ps_nopfn__wake", "ctx %p spu %p", spu_context_event },
-	{ "spufs_ps_nopfn__insert", "ctx %p spu %p", spu_context_event },
+	{ "spufs_ps_fault__enter", "ctx %p", spu_context_nospu_event },
+	{ "spufs_ps_fault__sleep", "ctx %p", spu_context_nospu_event },
+	{ "spufs_ps_fault__wake", "ctx %p spu %p", spu_context_event },
+	{ "spufs_ps_fault__insert", "ctx %p spu %p", spu_context_event },
 	{ "spu_acquire_saved__enter", "ctx %p", spu_context_nospu_event },
 	{ "destroy_spu_context__enter", "ctx %p", spu_context_nospu_event },
 	{ "spufs_stop_callback__enter", "ctx %p spu %p", spu_context_event },

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

Acked-by on v1 3/4
#5 lore raw

Re: [patch 3/4] spufs: convert nopfn to fault

Hi Nick,

Quoted 2 lines

-static unsigned long spufs_mem_mmap_nopfn(struct vm_area_struct
*vma, - unsigned long address)

Aside from the > 80 character lines, all is OK here.

Cheers,

Jeremy

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#6 lore raw

Re: [patch 3/4] spufs: convert nopfn to fault

Quoted 8 lines · Jeremy Kerr

Hi Nick,

Quoted 2 lines

-static unsigned long spufs_mem_mmap_nopfn(struct vm_area_struct
*vma, - unsigned long address)

Aside from the > 80 character lines, all is OK here.

Great, thanks very much!

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#7 lore raw

Re: [patch 3/4] spufs: convert nopfn to fault

Hi Nick,

Quoted 3 lines
Quoted 1 line

Great, thanks very much!

After more testing, it looks like these patches cause a huge increase in load (ie, system is unresponsive for large amounts of time) for various tests which depend on the fault path.

I need to get some quantitative numbers, but it looks like oprofile is broken at the moment. More debugging coming..

Cheers,

Jeremy

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#8 lore raw

Re: [patch 3/4] spufs: convert nopfn to fault

Quoted 4 lines · Jeremy Kerr

Hi Nick,

Quoted 2 lines
Quoted 1 line
Quoted 8 lines
Quoted 1 line

Great, thanks very much!

After more testing, it looks like these patches cause a huge increase in load (ie, system is unresponsive for large amounts of time) for various tests which depend on the fault path.

I need to get some quantitative numbers, but it looks like oprofile is broken at the moment. More debugging coming..

OK, thanks for testing that... It _should_ be 100% equivalent really, so it must be some problem in the conversion. Don't worry too much about getting exact numbers because any noticable difference would be a bug.

Hmm, in spufs_mem_mmap_fault, vm_insert_pfn should just take address (corrected for 64K), rather than the uncorrected address I gave it...

Can't see any other problems though. Is it getting stuck looping in faults somehow?

Thanks,
Nick

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#9 lore raw

Re: [patch 3/4] spufs: convert nopfn to fault

Hi Nick,

Quoted 3 lines

Hmm, in spufs_mem_mmap_fault, vm_insert_pfn should just take address (corrected for 64K), rather than the uncorrected address I gave it...

Yep, using the 'address' var for vm_insert_pfn fixes the problem for me.

Cheers,

Jeremy

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#10 lore raw

Re: [patch 3/4] spufs: convert nopfn to fault

Quoted 7 lines · Jeremy Kerr

Hi Nick,

Quoted 3 lines

Hmm, in spufs_mem_mmap_fault, vm_insert_pfn should just take address (corrected for 64K), rather than the uncorrected address I gave it...

Yep, using the 'address' var for vm_insert_pfn fixes the problem for me.

Ah, thanks for testing. Will send an updated patch also with the warning you noticed fixed.

Thanks,
Nick

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

author Acked-by Jeremy Kerr
#11 lore raw

Re: [patch 3/4] spufs: convert nopfn to fault

In reply to Jeremy Kerr
Quoted 6 lines · Jeremy Kerr

Hi Nick,

Quoted 2 lines

-static unsigned long spufs_mem_mmap_nopfn(struct vm_area_struct
*vma, - unsigned long address)

Aside from the > 80 character lines, all is OK here.

And here is an update with <= 80 column lines...

spufs: convert nopfn to fault

From: Nick Piggin <[email protected]>

---

 arch/powerpc/platforms/cell/spufs/file.c     |   91 ++++++++++++---------------
 arch/powerpc/platforms/cell/spufs/sputrace.c |    8 +-
 2 files changed, 46 insertions(+), 53 deletions(-)
arch/powerpc/platforms/cell/spufs/file.c +52 -49
Index: linux-2.6/arch/powerpc/platforms/cell/spufs/file.c
===================================================================
@@ -237,11 +237,14 @@ spufs_mem_write(struct file *file, const
 	return size;
 }
 
-static unsigned long spufs_mem_mmap_nopfn(struct vm_area_struct *vma,
-					  unsigned long address)
+static int
+spufs_mem_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
 	struct spu_context *ctx	= vma->vm_file->private_data;
-	unsigned long pfn, offset, addr0 = address;
+	unsigned long pfn, offset, address;
+
+	address = (unsigned long)vmf->virtual_address;
+
 #ifdef CONFIG_SPU_FS_64K_LS
 	struct spu_state *csa = &ctx->csa;
 	int psize;
@@ -259,15 +262,15 @@ static unsigned long spufs_mem_mmap_nopf
 	}
 #endif /* CONFIG_SPU_FS_64K_LS */
 
-	offset = (address - vma->vm_start) + (vma->vm_pgoff << PAGE_SHIFT);
+	offset = vmf->pgoff << PAGE_SHIFT;
 	if (offset >= LS_SIZE)
-		return NOPFN_SIGBUS;
+		return VM_FAULT_SIGBUS;
 
-	pr_debug("spufs_mem_mmap_nopfn address=0x%lx -> 0x%lx, offset=0x%lx\n",
-		 addr0, address, offset);
+	pr_debug("spufs_mem_mmap_fault address=0x%lx, offset=0x%lx\n",
+			address, offset);
 
 	if (spu_acquire(ctx))
-		return NOPFN_REFAULT;
+		return VM_FAULT_NOPAGE;
 
 	if (ctx->state == SPU_STATE_SAVED) {
 		vma->vm_page_prot = __pgprot(pgprot_val(vma->vm_page_prot)
@@ -278,16 +281,16 @@ static unsigned long spufs_mem_mmap_nopf
 					     | _PAGE_NO_CACHE);
 		pfn = (ctx->spu->local_store_phys + offset) >> PAGE_SHIFT;
 	}
-	vm_insert_pfn(vma, address, pfn);
+	vm_insert_pfn(vma, (unsigned long)vmf->virtual_address, pfn);
 
 	spu_release(ctx);
 
-	return NOPFN_REFAULT;
+	return VM_FAULT_NOPAGE;
 }
 
 
 static struct vm_operations_struct spufs_mem_mmap_vmops = {
-	.nopfn = spufs_mem_mmap_nopfn,
+	.fault = spufs_mem_mmap_fault,
 };
 
 static int spufs_mem_mmap(struct file *file, struct vm_area_struct *vma)
@@ -350,20 +353,19 @@ static const struct file_operations spuf
 #endif
 };
 
-static unsigned long spufs_ps_nopfn(struct vm_area_struct *vma,
-				    unsigned long address,
+static int spufs_ps_fault(struct vm_area_struct *vma,
+				    struct vm_fault *vmf,
 				    unsigned long ps_offs,
 				    unsigned long ps_size)
 {
 	struct spu_context *ctx = vma->vm_file->private_data;
-	unsigned long area, offset = address - vma->vm_start;
+	unsigned long area, offset = vmf->pgoff << PAGE_SHIFT;
 	int ret = 0;
 
-	spu_context_nospu_trace(spufs_ps_nopfn__enter, ctx);
+	spu_context_nospu_trace(spufs_ps_fault__enter, ctx);
 
-	offset += vma->vm_pgoff << PAGE_SHIFT;
 	if (offset >= ps_size)
-		return NOPFN_SIGBUS;
+		return VM_FAULT_SIGBUS;
 
 	/*
 	 * Because we release the mmap_sem, the context may be destroyed while
@@ -377,7 +379,7 @@ static unsigned long spufs_ps_nopfn(stru
 	 * pages to hand out to the user, but we don't want to wait
 	 * with the mmap_sem held.
 	 * It is possible to drop the mmap_sem here, but then we need
-	 * to return NOPFN_REFAULT because the mappings may have
+	 * to return VM_FAULT_NOPAGE because the mappings may have
 	 * hanged.
 	 */
 	if (spu_acquire(ctx))
@@ -385,14 +387,15 @@ static unsigned long spufs_ps_nopfn(stru
 
 	if (ctx->state == SPU_STATE_SAVED) {
 		up_read(&current->mm->mmap_sem);
-		spu_context_nospu_trace(spufs_ps_nopfn__sleep, ctx);
+		spu_context_nospu_trace(spufs_ps_fault__sleep, ctx);
 		ret = spufs_wait(ctx->run_wq, ctx->state == SPU_STATE_RUNNABLE);
-		spu_context_trace(spufs_ps_nopfn__wake, ctx, ctx->spu);
+		spu_context_trace(spufs_ps_fault__wake, ctx, ctx->spu);
 		down_read(&current->mm->mmap_sem);
 	} else {
 		area = ctx->spu->problem_phys + ps_offs;
-		vm_insert_pfn(vma, address, (area + offset) >> PAGE_SHIFT);
-		spu_context_trace(spufs_ps_nopfn__insert, ctx, ctx->spu);
+		vm_insert_pfn(vma, (unsigned long)vmf->virtual_address,
+					(area + offset) >> PAGE_SHIFT);
+		spu_context_trace(spufs_ps_fault__insert, ctx, ctx->spu);
 	}
 
 	if (!ret)
@@ -400,18 +403,18 @@ static unsigned long spufs_ps_nopfn(stru
 
 refault:
 	put_spu_context(ctx);
-	return NOPFN_REFAULT;
+	return VM_FAULT_NOPAGE;
 }
 
 #if SPUFS_MMAP_4K
-static unsigned long spufs_cntl_mmap_nopfn(struct vm_area_struct *vma,
-					   unsigned long address)
+static int spufs_cntl_mmap_fault(struct vm_area_struct *vma,
+					   struct vm_fault *vmf)
 {
-	return spufs_ps_nopfn(vma, address, 0x4000, 0x1000);
+	return spufs_ps_fault(vma, vmf, 0x4000, 0x1000);
 }
 
 static struct vm_operations_struct spufs_cntl_mmap_vmops = {
-	.nopfn = spufs_cntl_mmap_nopfn,
+	.fault = spufs_cntl_mmap_fault,
 };
 
 /*
@@ -1096,23 +1099,23 @@ static ssize_t spufs_signal1_write(struc
 	return 4;
 }
 
-static unsigned long spufs_signal1_mmap_nopfn(struct vm_area_struct *vma,
-					      unsigned long address)
+static int
+spufs_signal1_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
 #if PAGE_SIZE == 0x1000
-	return spufs_ps_nopfn(vma, address, 0x14000, 0x1000);
+	return spufs_ps_fault(vma, vmf, 0x14000, 0x1000);
 #elif PAGE_SIZE == 0x10000
 	/* For 64k pages, both signal1 and signal2 can be used to mmap the whole
 	 * signal 1 and 2 area
 	 */
-	return spufs_ps_nopfn(vma, address, 0x10000, 0x10000);
+	return spufs_ps_fault(vma, vmf, 0x10000, 0x10000);
 #else
 #error unsupported page size
 #endif
 }
 
 static struct vm_operations_struct spufs_signal1_mmap_vmops = {
-	.nopfn = spufs_signal1_mmap_nopfn,
+	.fault = spufs_signal1_mmap_fault,
 };
 
 static int spufs_signal1_mmap(struct file *file, struct vm_area_struct *vma)
@@ -1233,23 +1236,23 @@ static ssize_t spufs_signal2_write(struc
 }
 
 #if SPUFS_MMAP_4K
-static unsigned long spufs_signal2_mmap_nopfn(struct vm_area_struct *vma,
-					      unsigned long address)
+static int
+spufs_signal2_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
 #if PAGE_SIZE == 0x1000
-	return spufs_ps_nopfn(vma, address, 0x1c000, 0x1000);
+	return spufs_ps_fault(vma, vmf, 0x1c000, 0x1000);
 #elif PAGE_SIZE == 0x10000
 	/* For 64k pages, both signal1 and signal2 can be used to mmap the whole
 	 * signal 1 and 2 area
 	 */
-	return spufs_ps_nopfn(vma, address, 0x10000, 0x10000);
+	return spufs_ps_fault(vma, vmf, 0x10000, 0x10000);
 #else
 #error unsupported page size
 #endif
 }
 
 static struct vm_operations_struct spufs_signal2_mmap_vmops = {
-	.nopfn = spufs_signal2_mmap_nopfn,
+	.fault = spufs_signal2_mmap_fault,
 };
 
 static int spufs_signal2_mmap(struct file *file, struct vm_area_struct *vma)
@@ -1361,14 +1364,14 @@ DEFINE_SPUFS_ATTRIBUTE(spufs_signal2_typ
 		       spufs_signal2_type_set, "%llu\n", SPU_ATTR_ACQUIRE);
 
 #if SPUFS_MMAP_4K
-static unsigned long spufs_mss_mmap_nopfn(struct vm_area_struct *vma,
-					  unsigned long address)
+static int
+spufs_mss_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
-	return spufs_ps_nopfn(vma, address, 0x0000, 0x1000);
+	return spufs_ps_fault(vma, vmf, 0x0000, 0x1000);
 }
 
 static struct vm_operations_struct spufs_mss_mmap_vmops = {
-	.nopfn = spufs_mss_mmap_nopfn,
+	.fault = spufs_mss_mmap_fault,
 };
 
 /*
@@ -1423,14 +1426,14 @@ static const struct file_operations spuf
 	.mmap	 = spufs_mss_mmap,
 };
 
-static unsigned long spufs_psmap_mmap_nopfn(struct vm_area_struct *vma,
-					    unsigned long address)
+static int
+spufs_psmap_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
-	return spufs_ps_nopfn(vma, address, 0x0000, 0x20000);
+	return spufs_ps_fault(vma, vmf, 0x0000, 0x20000);
 }
 
 static struct vm_operations_struct spufs_psmap_mmap_vmops = {
-	.nopfn = spufs_psmap_mmap_nopfn,
+	.fault = spufs_psmap_mmap_fault,
 };
 
 /*
@@ -1483,14 +1486,14 @@ static const struct file_operations spuf
 
 
 #if SPUFS_MMAP_4K
-static unsigned long spufs_mfc_mmap_nopfn(struct vm_area_struct *vma,
-					  unsigned long address)
+static int
+spufs_mfc_mmap_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 {
-	return spufs_ps_nopfn(vma, address, 0x3000, 0x1000);
+	return spufs_ps_fault(vma, vmf, 0x3000, 0x1000);
 }
 
 static struct vm_operations_struct spufs_mfc_mmap_vmops = {
-	.nopfn = spufs_mfc_mmap_nopfn,
+	.fault = spufs_mfc_mmap_fault,
 };
 
 /*
arch/powerpc/platforms/cell/spufs/sputrace.c +4 -4
Index: linux-2.6/arch/powerpc/platforms/cell/spufs/sputrace.c
===================================================================
@@ -182,10 +182,10 @@ struct spu_probe spu_probes[] = {
 	{ "spu_yield__enter", "ctx %p", spu_context_nospu_event },
 	{ "spu_deactivate__enter", "ctx %p", spu_context_nospu_event },
 	{ "__spu_deactivate__unload", "ctx %p spu %p", spu_context_event },
-	{ "spufs_ps_nopfn__enter", "ctx %p", spu_context_nospu_event },
-	{ "spufs_ps_nopfn__sleep", "ctx %p", spu_context_nospu_event },
-	{ "spufs_ps_nopfn__wake", "ctx %p spu %p", spu_context_event },
-	{ "spufs_ps_nopfn__insert", "ctx %p spu %p", spu_context_event },
+	{ "spufs_ps_fault__enter", "ctx %p", spu_context_nospu_event },
+	{ "spufs_ps_fault__sleep", "ctx %p", spu_context_nospu_event },
+	{ "spufs_ps_fault__wake", "ctx %p spu %p", spu_context_event },
+	{ "spufs_ps_fault__insert", "ctx %p spu %p", spu_context_event },
 	{ "spu_acquire_saved__enter", "ctx %p", spu_context_nospu_event },
 	{ "destroy_spu_context__enter", "ctx %p", spu_context_nospu_event },
 	{ "spufs_stop_callback__enter", "ctx %p spu %p", spu_context_event },

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#12 lore raw

Re: [patch 3/4] spufs: convert nopfn to fault

Hi Nick,

Quoted 5 lines

-static unsigned long spufs_mem_mmap_nopfn(struct vm_area_struct
*vma, - unsigned long address)

(quoted diff) +2 -0
+static int
+spufs_mem_mmap_fault(struct vm_area_struct *vma, struct vm_fault

*vmf) {

arch/powerpc/platforms/cell/spufs/file.c spufs_mem_write(struct file *file, const · line 250

Quoted 9 lines
arch/powerpc/platforms/cell/spufs/file.c +4 -1
 	struct spu_context *ctx	= vma->vm_file->private_data;
-	unsigned long pfn, offset, addr0 = address;
+	unsigned long pfn, offset, address;
+
+	address = (unsigned long)vmf->virtual_address;
+
 #ifdef CONFIG_SPU_FS_64K_LS
 	struct spu_state *csa = &ctx->csa;
 	int psize;

This will add a warning (you're "mixing declarations and code") if CONFIG_SPU_FS_64K_LS.

Cheers,

Jeremy

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#13 lore raw

PATCH 4/4 mm: remove nopfn

In reply to Nick Piggin

There are no users of nopfn in the tree. Remove it.

---

 include/linux/mm.h |    9 -------
 mm/memory.c        |   61 ++++-------------------------------------------------
 2 files changed, 5 insertions(+), 65 deletions(-)
include/linux/mm.h +0 -9
Index: linux-2.6/include/linux/mm.h
===================================================================
@@ -165,8 +165,6 @@ struct vm_operations_struct {
 	void (*open)(struct vm_area_struct * area);
 	void (*close)(struct vm_area_struct * area);
 	int (*fault)(struct vm_area_struct *vma, struct vm_fault *vmf);
-	unsigned long (*nopfn)(struct vm_area_struct *area,
-			unsigned long address);
 
 	/* notification that a previously read-only page is about to become
 	 * writable, if an error is returned it will cause a SIGBUS */
@@ -674,13 +672,6 @@ static inline int page_mapped(struct pag
 }
 
 /*
- * Error return values for the *_nopfn functions
- */
-#define NOPFN_SIGBUS	((unsigned long) -1)
-#define NOPFN_OOM	((unsigned long) -2)
-#define NOPFN_REFAULT	((unsigned long) -3)
-
-/*
  * Different kinds of faults, as returned by handle_mm_fault().
  * Used to decide whether a process gets delivered SIGBUS or
  * just gets major/minor fault counters bumped up.
mm/memory.c +5 -56

79 lines not on this page · raw message

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>