[PATCH v13 01/16] PCI: Let pci_mmap_page_range() take resource address

[PATCH v13 01/16] PCI: Let pci_mmap_page_range() take resource address

[Yinghai Lu]

In 8c05cd08a7 ("PCI: fix offset check for sysfs mmapped files"), try
to check exposed value with resource start/end in proc mmap path.

|        start = vma->vm_pgoff;
|        size = ((pci_resource_len(pdev, resno) - 1) >> PAGE_SHIFT) + 1;
|        pci_start = (mmap_api == PCI_MMAP_PROCFS) ?
|                        pci_resource_start(pdev, resno) >> PAGE_SHIFT : 0;
|        if (start >= pci_start && start < pci_start + size &&
|                        start + nr <= pci_start + size)

That breaks sparc that exposed value is BAR value, and need to be offseted
to resource address.

Original pci_mmap_page_range() is taking PCI BAR value aka usr_address.

Bjorn found out that it would be much simple to pass resource address
directly and avoid extra those __pci_mmap_make_offset.

In this patch:
1. in proc path: proc_bus_pci_mmap, try convert back to resource
   before calling pci_mmap_page_range
2. in sysfs path: pci_mmap_resource will just offset with resource start.
3. all pci_mmap_page_range will have vma->vm_pgoff with in resource
   range instead of BAR value.
4. skip calling __pci_mmap_make_offset, as the checking is done
   in pci_mmap_fits().

-v2: add pci_user_to_resource and remove __pci_mmap_make_offset
-v3: pass resource pointer with pci_mmap_page_range()
-v4: put __pci_mmap_make_offset() removing to following patch
     seperate /sys io access alignment checking to another patch
     updated after Bjorn's pci_resource_to_user() changes.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: sparclinux@vger.kernel.org
Cc: linux-xtensa@linux-xtensa.org
---
 arch/microblaze/pci/pci-common.c | 11 +++++---
 arch/powerpc/kernel/pci-common.c | 11 +++++---
 arch/sparc/kernel/pci.c          |  4 ---
 arch/xtensa/kernel/pci.c         | 13 +++++++--
 drivers/pci/pci-sysfs.c          | 23 +++++++++------
 drivers/pci/pci.h                |  2 +-
 drivers/pci/proc.c               | 60 ++++++++++++++++++++++++++++++++++------
 7 files changed, 91 insertions(+), 33 deletions(-)

diff --git a/arch/microblaze/pci/pci-common.c b/arch/microblaze/pci/pci-common.c
index 81556b8..9e3bc05 100644
--- a/arch/microblaze/pci/pci-common.c
+++ b/arch/microblaze/pci/pci-common.c
@@ -282,12 +282,15 @@ int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
 {
 	resource_size_t offset =
 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
-	struct resource *rp;
 	int ret;
 
-	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
-	if (rp == NULL)
-		return -EINVAL;
+	if (mmap_state == pci_mmap_io) {
+		struct pci_controller *hose = pci_bus_to_host(dev->bus);
+
+		/* hose should never be NULL */
+		offset += hose->io_base_phys -
+			 ((unsigned long)hose->io_base_virt - _IO_BASE);
+	}
 
 	vma->vm_pgoff = offset >> PAGE_SHIFT;
 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
diff --git a/arch/powerpc/kernel/pci-common.c b/arch/powerpc/kernel/pci-common.c
index 6de6e0e..53ba098 100644
--- a/arch/powerpc/kernel/pci-common.c
+++ b/arch/powerpc/kernel/pci-common.c
@@ -420,12 +420,15 @@ int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
 {
 	resource_size_t offset =
 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
-	struct resource *rp;
 	int ret;
 
-	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
-	if (rp == NULL)
-		return -EINVAL;
+	if (mmap_state == pci_mmap_io) {
+		struct pci_controller *hose = pci_bus_to_host(dev->bus);
+
+		/* hose should never be NULL */
+		offset += hose->io_base_phys -
+			  ((unsigned long)hose->io_base_virt - _IO_BASE);
+	}
 
 	vma->vm_pgoff = offset >> PAGE_SHIFT;
 	if (write_combine)
diff --git a/arch/sparc/kernel/pci.c b/arch/sparc/kernel/pci.c
index 9c1878f..5f2d78e 100644
--- a/arch/sparc/kernel/pci.c
+++ b/arch/sparc/kernel/pci.c
@@ -868,10 +868,6 @@ int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
 {
 	int ret;
 
-	ret = __pci_mmap_make_offset(dev, vma, mmap_state);
-	if (ret < 0)
-		return ret;
-
 	__pci_mmap_set_pgprot(dev, vma, mmap_state);
 
 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
diff --git a/arch/xtensa/kernel/pci.c b/arch/xtensa/kernel/pci.c
index b848cc3..4c5f1fa 100644
--- a/arch/xtensa/kernel/pci.c
+++ b/arch/xtensa/kernel/pci.c
@@ -366,11 +366,18 @@ int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
 			enum pci_mmap_state mmap_state,
 			int write_combine)
 {
+	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
 	int ret;
 
-	ret = __pci_mmap_make_offset(dev, vma, mmap_state);
-	if (ret < 0)
-		return ret;
+	if (mmap_state == pci_mmap_io) {
+		struct pci_controller *pci_ctrl =
+					 (struct pci_controller *)dev->sysdata;
+
+		/* pci_ctrl should never be NULL */
+		offset += pci_ctrl->io_space.start - pci_ctrl->io_space.base;
+	}
+
+	vma->vm_pgoff = offset >> PAGE_SHIFT;
 
 	__pci_mmap_set_pgprot(dev, vma, mmap_state, write_combine);
 
diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
index d319a9c..82b98dd 100644
--- a/drivers/pci/pci-sysfs.c
+++ b/drivers/pci/pci-sysfs.c
@@ -967,12 +967,23 @@ void pci_remove_legacy_files(struct pci_bus *b)
 #ifdef HAVE_PCI_MMAP
 
 int pci_mmap_fits(struct pci_dev *pdev, int resno, struct vm_area_struct *vma,
+		  enum pci_mmap_state mmap_type,
 		  enum pci_mmap_api mmap_api)
 {
 	unsigned long nr, start, size, pci_start;
+	int flags;
 
 	if (pci_resource_len(pdev, resno) == 0)
 		return 0;
+
+	if (mmap_type == pci_mmap_mem)
+		flags = IORESOURCE_MEM;
+	else
+		flags = IORESOURCE_IO;
+
+	if (!(pci_resource_flags(pdev, resno) & flags))
+		return 0;
+
 	nr = vma_pages(vma);
 	start = vma->vm_pgoff;
 	size = ((pci_resource_len(pdev, resno) - 1) >> PAGE_SHIFT) + 1;
@@ -999,7 +1010,6 @@ static int pci_mmap_resource(struct kobject *kobj, struct bin_attribute *attr,
 	struct pci_dev *pdev = to_pci_dev(kobj_to_dev(kobj));
 	struct resource *res = attr->private;
 	enum pci_mmap_state mmap_type;
-	resource_size_t start, end;
 	int i;
 
 	for (i = 0; i < PCI_ROM_RESOURCE; i++)
@@ -1011,7 +1021,8 @@ static int pci_mmap_resource(struct kobject *kobj, struct bin_attribute *attr,
 	if (res->flags & IORESOURCE_MEM && iomem_is_exclusive(res->start))
 		return -EINVAL;
 
-	if (!pci_mmap_fits(pdev, i, vma, PCI_MMAP_SYSFS)) {
+	mmap_type = res->flags & IORESOURCE_MEM ? pci_mmap_mem : pci_mmap_io;
+	if (!pci_mmap_fits(pdev, i, vma, mmap_type, PCI_MMAP_SYSFS)) {
 		WARN(1, "process \"%s\" tried to map 0x%08lx bytes at page 0x%08lx on %s BAR %d (start 0x%16Lx, size 0x%16Lx)\n",
 			current->comm, vma->vm_end-vma->vm_start, vma->vm_pgoff,
 			pci_name(pdev), i,
@@ -1020,13 +1031,7 @@ static int pci_mmap_resource(struct kobject *kobj, struct bin_attribute *attr,
 		return -EINVAL;
 	}
 
-	/* pci_mmap_page_range() expects the same kind of entry as coming
-	 * from /proc/bus/pci/ which is a "user visible" value. If this is
-	 * different from the resource itself, arch will do necessary fixup.
-	 */
-	pci_resource_to_user(pdev, i, res, &start, &end);
-	vma->vm_pgoff += start >> PAGE_SHIFT;
-	mmap_type = res->flags & IORESOURCE_MEM ? pci_mmap_mem : pci_mmap_io;
+	vma->vm_pgoff += res->start >> PAGE_SHIFT;
 	return pci_mmap_page_range(pdev, vma, mmap_type, write_combine);
 }
 
diff --git a/drivers/pci/pci.h b/drivers/pci/pci.h
index a814bbb..7d339c3 100644
--- a/drivers/pci/pci.h
+++ b/drivers/pci/pci.h
@@ -30,7 +30,7 @@ enum pci_mmap_api {
 	PCI_MMAP_PROCFS	/* mmap on /proc/bus/pci/<BDF> */
 };
 int pci_mmap_fits(struct pci_dev *pdev, int resno, struct vm_area_struct *vmai,
-		  enum pci_mmap_api mmap_api);
+		  enum pci_mmap_state mmap_type, enum pci_mmap_api mmap_api);
 #endif
 int pci_probe_reset_function(struct pci_dev *dev);
 
diff --git a/drivers/pci/proc.c b/drivers/pci/proc.c
index 2408abe..fadab8a 100644
--- a/drivers/pci/proc.c
+++ b/drivers/pci/proc.c
@@ -227,24 +227,68 @@ static long proc_bus_pci_ioctl(struct file *file, unsigned int cmd,
 }
 
 #ifdef HAVE_PCI_MMAP
+
+static int pci_user_to_resource(struct pci_dev *dev, resource_size_t *offset,
+				int flags)
+{
+	int i;
+
+	for (i = 0; i < PCI_ROM_RESOURCE; i++) {
+		resource_size_t start, end;
+		struct resource *res = &dev->resource[i];
+
+		if (!(res->flags & flags))
+			continue;
+
+		if (pci_resource_len(dev, i) == 0)
+			continue;
+
+		/*
+		 * here *offset is PAGE_SIZE aligned from caller.
+		 * need align start/end for io port resource that is
+		 * usually not PAGE_SIZE aligned.
+		 * that means we let it go if they falls in same page.
+		 */
+		pci_resource_to_user(dev, i, res, &start, &end);
+		if ((start & PAGE_MASK) <= *offset &&
+		     *offset <= (end & PAGE_MASK)) {
+			*offset = res->start + (*offset - start);
+			return i;
+		}
+	}
+
+	return -ENODEV;
+}
+
 static int proc_bus_pci_mmap(struct file *file, struct vm_area_struct *vma)
 {
 	struct pci_dev *dev = PDE_DATA(file_inode(file));
 	struct pci_filp_private *fpriv = file->private_data;
-	int i, ret, write_combine;
+	resource_size_t offset;
+	int i, ret, flags, write_combine;
 
 	if (!capable(CAP_SYS_RAWIO))
 		return -EPERM;
 
-	/* Make sure the caller is mapping a real resource for this device */
-	for (i = 0; i < PCI_ROM_RESOURCE; i++) {
-		if (pci_mmap_fits(dev, i, vma,  PCI_MMAP_PROCFS))
-			break;
-	}
-
-	if (i >= PCI_ROM_RESOURCE)
+	offset = vma->vm_pgoff << PAGE_SHIFT;
+	if (fpriv->mmap_state == pci_mmap_mem)
+		flags = IORESOURCE_MEM;
+	else
+		flags = IORESOURCE_IO;
+	i = pci_user_to_resource(dev, &offset, flags);
+	if (i < 0)
 		return -ENODEV;
 
+	vma->vm_pgoff = offset >> PAGE_SHIFT;
+	if (!pci_mmap_fits(dev, i, vma, fpriv->mmap_state, PCI_MMAP_PROCFS)) {
+		WARN(1, "process \"%s\" tried to map 0x%08lx bytes at page 0x%08lx on %s BAR %d (start 0x%16Lx, size 0x%16Lx)\n",
+			current->comm, vma->vm_end-vma->vm_start, vma->vm_pgoff,
+			pci_name(dev), i,
+			(u64)pci_resource_start(dev, i),
+			(u64)pci_resource_len(dev, i));
+		return -EINVAL;
+	}
+
 	if (fpriv->mmap_state == pci_mmap_mem)
 		write_combine = fpriv->write_combine;
 	else
-- 
2.8.3

[PATCH v13 02/16] PCI: Remove __pci_mmap_make_offset()

[Yinghai Lu]

After
  PCI: Let pci_mmap_page_range() take resource address
No user for __pci_mmap_make_offset in those arch.

Remove them.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: sparclinux@vger.kernel.org
Cc: linux-xtensa@linux-xtensa.org
---
 arch/microblaze/pci/pci-common.c |  63 ----------------------
 arch/powerpc/kernel/pci-common.c |  63 ----------------------
 arch/sparc/kernel/pci.c          | 113 ---------------------------------------
 arch/xtensa/kernel/pci.c         |  62 ---------------------
 4 files changed, 301 deletions(-)

diff --git a/arch/microblaze/pci/pci-common.c b/arch/microblaze/pci/pci-common.c
index 9e3bc05..e7cd0ab 100644
--- a/arch/microblaze/pci/pci-common.c
+++ b/arch/microblaze/pci/pci-common.c
@@ -156,69 +156,6 @@ void pcibios_set_master(struct pci_dev *dev)
  */
 
 /*
- * Adjust vm_pgoff of VMA such that it is the physical page offset
- * corresponding to the 32-bit pci bus offset for DEV requested by the user.
- *
- * Basically, the user finds the base address for his device which he wishes
- * to mmap.  They read the 32-bit value from the config space base register,
- * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
- * offset parameter of mmap on /proc/bus/pci/XXX for that device.
- *
- * Returns negative error code on failure, zero on success.
- */
-static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
-					       resource_size_t *offset,
-					       enum pci_mmap_state mmap_state)
-{
-	struct pci_controller *hose = pci_bus_to_host(dev->bus);
-	unsigned long io_offset = 0;
-	int i, res_bit;
-
-	if (!hose)
-		return NULL;		/* should never happen */
-
-	/* If memory, add on the PCI bridge address offset */
-	if (mmap_state == pci_mmap_mem) {
-#if 0 /* See comment in pci_resource_to_user() for why this is disabled */
-		*offset += hose->pci_mem_offset;
-#endif
-		res_bit = IORESOURCE_MEM;
-	} else {
-		io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
-		*offset += io_offset;
-		res_bit = IORESOURCE_IO;
-	}
-
-	/*
-	 * Check that the offset requested corresponds to one of the
-	 * resources of the device.
-	 */
-	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
-		struct resource *rp = &dev->resource[i];
-		int flags = rp->flags;
-
-		/* treat ROM as memory (should be already) */
-		if (i == PCI_ROM_RESOURCE)
-			flags |= IORESOURCE_MEM;
-
-		/* Active and same type? */
-		if ((flags & res_bit) == 0)
-			continue;
-
-		/* In the range of this resource? */
-		if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
-			continue;
-
-		/* found it! construct the final physical address */
-		if (mmap_state == pci_mmap_io)
-			*offset += hose->io_base_phys - io_offset;
-		return rp;
-	}
-
-	return NULL;
-}
-
-/*
  * This one is used by /dev/mem and fbdev who have no clue about the
  * PCI device, it tries to find the PCI device first and calls the
  * above routine
diff --git a/arch/powerpc/kernel/pci-common.c b/arch/powerpc/kernel/pci-common.c
index 53ba098..14c183d 100644
--- a/arch/powerpc/kernel/pci-common.c
+++ b/arch/powerpc/kernel/pci-common.c
@@ -293,69 +293,6 @@ static int pci_read_irq_line(struct pci_dev *pci_dev)
  */
 
 /*
- * Adjust vm_pgoff of VMA such that it is the physical page offset
- * corresponding to the 32-bit pci bus offset for DEV requested by the user.
- *
- * Basically, the user finds the base address for his device which he wishes
- * to mmap.  They read the 32-bit value from the config space base register,
- * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
- * offset parameter of mmap on /proc/bus/pci/XXX for that device.
- *
- * Returns negative error code on failure, zero on success.
- */
-static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
-					       resource_size_t *offset,
-					       enum pci_mmap_state mmap_state)
-{
-	struct pci_controller *hose = pci_bus_to_host(dev->bus);
-	unsigned long io_offset = 0;
-	int i, res_bit;
-
-	if (hose == NULL)
-		return NULL;		/* should never happen */
-
-	/* If memory, add on the PCI bridge address offset */
-	if (mmap_state == pci_mmap_mem) {
-#if 0 /* See comment in pci_resource_to_user() for why this is disabled */
-		*offset += hose->pci_mem_offset;
-#endif
-		res_bit = IORESOURCE_MEM;
-	} else {
-		io_offset = (unsigned long)hose->io_base_virt - _IO_BASE;
-		*offset += io_offset;
-		res_bit = IORESOURCE_IO;
-	}
-
-	/*
-	 * Check that the offset requested corresponds to one of the
-	 * resources of the device.
-	 */
-	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
-		struct resource *rp = &dev->resource[i];
-		int flags = rp->flags;
-
-		/* treat ROM as memory (should be already) */
-		if (i == PCI_ROM_RESOURCE)
-			flags |= IORESOURCE_MEM;
-
-		/* Active and same type? */
-		if ((flags & res_bit) == 0)
-			continue;
-
-		/* In the range of this resource? */
-		if (*offset < (rp->start & PAGE_MASK) || *offset > rp->end)
-			continue;
-
-		/* found it! construct the final physical address */
-		if (mmap_state == pci_mmap_io)
-			*offset += hose->io_base_phys - io_offset;
-		return rp;
-	}
-
-	return NULL;
-}
-
-/*
  * This one is used by /dev/mem and fbdev who have no clue about the
  * PCI device, it tries to find the PCI device first and calls the
  * above routine
diff --git a/arch/sparc/kernel/pci.c b/arch/sparc/kernel/pci.c
index 5f2d78e..c7d2ce6 100644
--- a/arch/sparc/kernel/pci.c
+++ b/arch/sparc/kernel/pci.c
@@ -732,119 +732,6 @@ int pcibios_enable_device(struct pci_dev *dev, int mask)
 
 /* Platform support for /proc/bus/pci/X/Y mmap()s. */
 
-/* If the user uses a host-bridge as the PCI device, he may use
- * this to perform a raw mmap() of the I/O or MEM space behind
- * that controller.
- *
- * This can be useful for execution of x86 PCI bios initialization code
- * on a PCI card, like the xfree86 int10 stuff does.
- */
-static int __pci_mmap_make_offset_bus(struct pci_dev *pdev, struct vm_area_struct *vma,
-				      enum pci_mmap_state mmap_state)
-{
-	struct pci_pbm_info *pbm = pdev->dev.archdata.host_controller;
-	unsigned long space_size, user_offset, user_size;
-
-	if (mmap_state == pci_mmap_io) {
-		space_size = resource_size(&pbm->io_space);
-	} else {
-		space_size = resource_size(&pbm->mem_space);
-	}
-
-	/* Make sure the request is in range. */
-	user_offset = vma->vm_pgoff << PAGE_SHIFT;
-	user_size = vma->vm_end - vma->vm_start;
-
-	if (user_offset >= space_size ||
-	    (user_offset + user_size) > space_size)
-		return -EINVAL;
-
-	if (mmap_state == pci_mmap_io) {
-		vma->vm_pgoff = (pbm->io_space.start +
-				 user_offset) >> PAGE_SHIFT;
-	} else {
-		vma->vm_pgoff = (pbm->mem_space.start +
-				 user_offset) >> PAGE_SHIFT;
-	}
-
-	return 0;
-}
-
-/* Adjust vm_pgoff of VMA such that it is the physical page offset
- * corresponding to the 32-bit pci bus offset for DEV requested by the user.
- *
- * Basically, the user finds the base address for his device which he wishes
- * to mmap.  They read the 32-bit value from the config space base register,
- * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
- * offset parameter of mmap on /proc/bus/pci/XXX for that device.
- *
- * Returns negative error code on failure, zero on success.
- */
-static int __pci_mmap_make_offset(struct pci_dev *pdev,
-				  struct vm_area_struct *vma,
-				  enum pci_mmap_state mmap_state)
-{
-	unsigned long user_paddr, user_size;
-	int i, err;
-
-	/* First compute the physical address in vma->vm_pgoff,
-	 * making sure the user offset is within range in the
-	 * appropriate PCI space.
-	 */
-	err = __pci_mmap_make_offset_bus(pdev, vma, mmap_state);
-	if (err)
-		return err;
-
-	/* If this is a mapping on a host bridge, any address
-	 * is OK.
-	 */
-	if ((pdev->class >> 8) == PCI_CLASS_BRIDGE_HOST)
-		return err;
-
-	/* Otherwise make sure it's in the range for one of the
-	 * device's resources.
-	 */
-	user_paddr = vma->vm_pgoff << PAGE_SHIFT;
-	user_size = vma->vm_end - vma->vm_start;
-
-	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
-		struct resource *rp = &pdev->resource[i];
-		resource_size_t aligned_end;
-
-		/* Active? */
-		if (!rp->flags)
-			continue;
-
-		/* Same type? */
-		if (i == PCI_ROM_RESOURCE) {
-			if (mmap_state != pci_mmap_mem)
-				continue;
-		} else {
-			if ((mmap_state == pci_mmap_io &&
-			     (rp->flags & IORESOURCE_IO) == 0) ||
-			    (mmap_state == pci_mmap_mem &&
-			     (rp->flags & IORESOURCE_MEM) == 0))
-				continue;
-		}
-
-		/* Align the resource end to the next page address.
-		 * PAGE_SIZE intentionally added instead of (PAGE_SIZE - 1),
-		 * because actually we need the address of the next byte
-		 * after rp->end.
-		 */
-		aligned_end = (rp->end + PAGE_SIZE) & PAGE_MASK;
-
-		if ((rp->start <= user_paddr) &&
-		    (user_paddr + user_size) <= aligned_end)
-			break;
-	}
-
-	if (i > PCI_ROM_RESOURCE)
-		return -EINVAL;
-
-	return 0;
-}
-
 /* Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
  * device mapping.
  */
diff --git a/arch/xtensa/kernel/pci.c b/arch/xtensa/kernel/pci.c
index 4c5f1fa..97ad3dd 100644
--- a/arch/xtensa/kernel/pci.c
+++ b/arch/xtensa/kernel/pci.c
@@ -272,68 +272,6 @@ pci_controller_num(struct pci_dev *dev)
  */
 
 /*
- * Adjust vm_pgoff of VMA such that it is the physical page offset
- * corresponding to the 32-bit pci bus offset for DEV requested by the user.
- *
- * Basically, the user finds the base address for his device which he wishes
- * to mmap.  They read the 32-bit value from the config space base register,
- * add whatever PAGE_SIZE multiple offset they wish, and feed this into the
- * offset parameter of mmap on /proc/bus/pci/XXX for that device.
- *
- * Returns negative error code on failure, zero on success.
- */
-static __inline__ int
-__pci_mmap_make_offset(struct pci_dev *dev, struct vm_area_struct *vma,
-		       enum pci_mmap_state mmap_state)
-{
-	struct pci_controller *pci_ctrl = (struct pci_controller*) dev->sysdata;
-	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
-	unsigned long io_offset = 0;
-	int i, res_bit;
-
-	if (pci_ctrl == 0)
-		return -EINVAL;		/* should never happen */
-
-	/* If memory, add on the PCI bridge address offset */
-	if (mmap_state == pci_mmap_mem) {
-		res_bit = IORESOURCE_MEM;
-	} else {
-		io_offset = (unsigned long)pci_ctrl->io_space.base;
-		offset += io_offset;
-		res_bit = IORESOURCE_IO;
-	}
-
-	/*
-	 * Check that the offset requested corresponds to one of the
-	 * resources of the device.
-	 */
-	for (i = 0; i <= PCI_ROM_RESOURCE; i++) {
-		struct resource *rp = &dev->resource[i];
-		int flags = rp->flags;
-
-		/* treat ROM as memory (should be already) */
-		if (i == PCI_ROM_RESOURCE)
-			flags |= IORESOURCE_MEM;
-
-		/* Active and same type? */
-		if ((flags & res_bit) == 0)
-			continue;
-
-		/* In the range of this resource? */
-		if (offset < (rp->start & PAGE_MASK) || offset > rp->end)
-			continue;
-
-		/* found it! construct the final physical address */
-		if (mmap_state == pci_mmap_io)
-			offset += pci_ctrl->io_space.start - io_offset;
-		vma->vm_pgoff = offset >> PAGE_SHIFT;
-		return 0;
-	}
-
-	return -EINVAL;
-}
-
-/*
  * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
  * device mapping.
  */
-- 
2.8.3

[PATCH v13 09/16] powerpc/PCI: Keep resource idx order with bridge register number

[Yinghai Lu]

Same as sparc version.

Make resource with consistent sequence
like other arch or directly from pci_read_bridge_bases(),
even when non-pref mmio is missing, or out of ordering in firmware reporting.

Just hold i = 1 for non pref mmio, and i = 2 for pref mmio.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: linuxppc-dev@lists.ozlabs.org
---
 arch/powerpc/kernel/pci_of_scan.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/kernel/pci_of_scan.c b/arch/powerpc/kernel/pci_of_scan.c
index 526ac67..719f225 100644
--- a/arch/powerpc/kernel/pci_of_scan.c
+++ b/arch/powerpc/kernel/pci_of_scan.c
@@ -252,7 +252,7 @@ void of_scan_pci_bridge(struct pci_dev *dev)
 		bus->resource[i] = res;
 		++res;
 	}
-	i = 1;
+	i = 3;
 	for (; len >= 32; len -= 32, ranges += 8) {
 		flags = pci_parse_of_flags(of_read_number(ranges, 1), 1);
 		size = of_read_number(&ranges[6], 2);
@@ -265,6 +265,12 @@ void of_scan_pci_bridge(struct pci_dev *dev)
 				       " for bridge %s\n", node->full_name);
 				continue;
 			}
+		} else if ((flags & IORESOURCE_PREFETCH) &&
+			   !bus->resource[2]->flags) {
+			res = bus->resource[2];
+		} else if (((flags & (IORESOURCE_MEM | IORESOURCE_PREFETCH)) ==
+			    IORESOURCE_MEM) && !bus->resource[1]->flags) {
+			res = bus->resource[1];
 		} else {
 			if (i >= PCI_NUM_RESOURCES - PCI_BRIDGE_RESOURCES) {
 				printk(KERN_ERR "PCI: too many memory ranges"
-- 
2.8.3

[PATCH v13 10/16] powerpc/PCI: Add IORESOURCE_MEM_64 for 64-bit resource in OF parsing

[Yinghai Lu]

For device resource PREF bit setting under bridge 64-bit pref resource,
we need to make sure only set PREF for 64bit resource.

This patch set IORESOUCE_MEM_64 for 64bit resource during OF device resource
flags parsing.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=96261
Link: https://bugzilla.kernel.org/show_bug.cgi?id=96241
Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Gavin Shan <gwshan@linux.vnet.ibm.com>
Cc: Yijing Wang <wangyijing@huawei.com>
Cc: Anton Blanchard <anton@samba.org>
Cc: linuxppc-dev@lists.ozlabs.org
---
 arch/powerpc/kernel/pci_of_scan.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/arch/powerpc/kernel/pci_of_scan.c b/arch/powerpc/kernel/pci_of_scan.c
index 719f225..476b8ac5 100644
--- a/arch/powerpc/kernel/pci_of_scan.c
+++ b/arch/powerpc/kernel/pci_of_scan.c
@@ -44,8 +44,10 @@ static unsigned int pci_parse_of_flags(u32 addr0, int bridge)
 
 	if (addr0 & 0x02000000) {
 		flags = IORESOURCE_MEM | PCI_BASE_ADDRESS_SPACE_MEMORY;
-		flags |= (addr0 >> 22) & PCI_BASE_ADDRESS_MEM_TYPE_64;
 		flags |= (addr0 >> 28) & PCI_BASE_ADDRESS_MEM_TYPE_1M;
+		if (addr0 & 0x01000000)
+			flags |= IORESOURCE_MEM_64
+				 | PCI_BASE_ADDRESS_MEM_TYPE_64;
 		if (addr0 & 0x40000000)
 			flags |= IORESOURCE_PREFETCH
 				 | PCI_BASE_ADDRESS_MEM_PREFETCH;
-- 
2.8.3

Re: [PATCH v13 01/16] PCI: Let pci_mmap_page_range() take resource address

[Bjorn Helgaas]

On Fri, Jun 17, 2016 at 07:24:46PM -0700, Yinghai Lu wrote:
> In 8c05cd08a7 ("PCI: fix offset check for sysfs mmapped files"), try
> to check exposed value with resource start/end in proc mmap path.
> 
> |        start = vma->vm_pgoff;
> |        size = ((pci_resource_len(pdev, resno) - 1) >> PAGE_SHIFT) + 1;
> |        pci_start = (mmap_api == PCI_MMAP_PROCFS) ?
> |                        pci_resource_start(pdev, resno) >> PAGE_SHIFT : 0;
> |        if (start >= pci_start && start < pci_start + size &&
> |                        start + nr <= pci_start + size)
> 
> That breaks sparc that exposed value is BAR value, and need to be offseted
> to resource address.

I asked this same question of the v12 patch, but I don't think you
answered it:

I'm not quite sure what you're saying here.  Are you saying that sparc
is currently broken, and this patch fixes it?  If so, what exactly is
broken?  Can you give a small example of an mmap that is currently
broken?

Re: [PATCH v13 01/16] PCI: Let pci_mmap_page_range() take resource address

[Yinghai Lu]

On Sat, Jun 18, 2016 at 5:17 AM, Bjorn Helgaas <helgaas@kernel.org> wrote:
> On Fri, Jun 17, 2016 at 07:24:46PM -0700, Yinghai Lu wrote:
>> In 8c05cd08a7 ("PCI: fix offset check for sysfs mmapped files"), try
>> to check exposed value with resource start/end in proc mmap path.
>>
>> |        start = vma->vm_pgoff;
>> |        size = ((pci_resource_len(pdev, resno) - 1) >> PAGE_SHIFT) + 1;
>> |        pci_start = (mmap_api == PCI_MMAP_PROCFS) ?
>> |                        pci_resource_start(pdev, resno) >> PAGE_SHIFT : 0;
>> |        if (start >= pci_start && start < pci_start + size &&
>> |                        start + nr <= pci_start + size)
>>
>> That breaks sparc that exposed value is BAR value, and need to be offseted
>> to resource address.
>
> I asked this same question of the v12 patch, but I don't think you
> answered it:
>
> I'm not quite sure what you're saying here.  Are you saying that sparc
> is currently broken, and this patch fixes it?  If so, what exactly is
> broken?  Can you give a small example of an mmap that is currently
> broken?

Yes, for sparc that path (proc mmap) is broken, but only according to
code checking.

The reason for the problem is not discovered is that seem all users
(other than x86) are not
use proc_mmap ?

vma->vm_pgoff is that code segment is User/BAR value >> PAGE_SHIFT.
pci_start is resource->start >> PAGE_SHIFT.

For sparc, resource start is different from BAR start aka pci bus address.
pci bus address add offset to be the resource start.

Thanks

Yinghai

Re: [PATCH v13 01/16] PCI: Let pci_mmap_page_range() take resource address

[Bjorn Helgaas]

On Tue, Jun 21, 2016 at 09:32:49PM -0700, Yinghai Lu wrote:
> On Sat, Jun 18, 2016 at 5:17 AM, Bjorn Helgaas <helgaas@kernel.org> wrote:
> > On Fri, Jun 17, 2016 at 07:24:46PM -0700, Yinghai Lu wrote:
> >> In 8c05cd08a7 ("PCI: fix offset check for sysfs mmapped files"), try
> >> to check exposed value with resource start/end in proc mmap path.
> >>
> >> |        start = vma->vm_pgoff;
> >> |        size = ((pci_resource_len(pdev, resno) - 1) >> PAGE_SHIFT) + 1;
> >> |        pci_start = (mmap_api == PCI_MMAP_PROCFS) ?
> >> |                        pci_resource_start(pdev, resno) >> PAGE_SHIFT : 0;
> >> |        if (start >= pci_start && start < pci_start + size &&
> >> |                        start + nr <= pci_start + size)
> >>
> >> That breaks sparc that exposed value is BAR value, and need to be offseted
> >> to resource address.
> >
> > I asked this same question of the v12 patch, but I don't think you
> > answered it:
> >
> > I'm not quite sure what you're saying here.  Are you saying that sparc
> > is currently broken, and this patch fixes it?  If so, what exactly is
> > broken?  Can you give a small example of an mmap that is currently
> > broken?
> 
> Yes, for sparc that path (proc mmap) is broken, but only according to
> code checking.
> 
> The reason for the problem is not discovered is that seem all users
> (other than x86) are not
> use proc_mmap ?
> 
> vma->vm_pgoff is that code segment is User/BAR value >> PAGE_SHIFT.
> pci_start is resource->start >> PAGE_SHIFT.
> 
> For sparc, resource start is different from BAR start aka pci bus address.
> pci bus address add offset to be the resource start.

If sparc is broken, let's make this a tiny sparc-only patch that fixes
only the breakage -- no cleanup or restructuring.  Then we can do the
more extensive work in a separate patch.

The example mmap() I keep asking for would be very helpful to me in
understanding the problem.  It would probably also help folks who
maintain user programs that use mmap.  They need to figure out whether
they have code that worked most places but has always been broken on
sparc, or code that depended on the previous sparc behavior and will
be broken by this change, or what.

Bjorn

Re: [PATCH v13 01/16] PCI: Let pci_mmap_page_range() take resource address

[Yinghai Lu]

[-- Attachment #1: Type: text/plain, Size: 2206 bytes --]

On Wed, Jun 22, 2016 at 8:22 AM, Bjorn Helgaas <helgaas@kernel.org> wrote:
> On Tue, Jun 21, 2016 at 09:32:49PM -0700, Yinghai Lu wrote:
>
> If sparc is broken, let's make this a tiny sparc-only patch that fixes
> only the breakage -- no cleanup or restructuring.  Then we can do the
> more extensive work in a separate patch.

ok. please check attached two patches that take position for
[PATCH v13 01/16]

>
> The example mmap() I keep asking for would be very helpful to me in
> understanding the problem.  It would probably also help folks who
> maintain user programs that use mmap.  They need to figure out whether
> they have code that worked most places but has always been broken on
> sparc, or code that depended on the previous sparc behavior and will
> be broken by this change, or what.

ok.

calling : ./test_mmap_proc /proc/bus/pci/0000:00/04.0 0x2000000

code : test_mmap_proc.c

#include <stdio.h>
#include <sys/mman.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>

#define PCIIOC_BASE             ('P' << 24 | 'C' << 16 | 'I' << 8)
#define PCIIOC_CONTROLLER       (PCIIOC_BASE | 0x00)    /* Get
controller for PCI device. */
#define PCIIOC_MMAP_IS_IO       (PCIIOC_BASE | 0x01)    /* Set mmap
state to I/O space. */
#define PCIIOC_MMAP_IS_MEM      (PCIIOC_BASE | 0x02)    /* Set mmap
state to MEM space. */
#define PCIIOC_WRITE_COMBINE    (PCIIOC_BASE | 0x03)    /*
Enable/disable write-combining. */

/* sparc64 */
#define PAGE_SHIFT 13
#define PAGE_SIZE (1UL << PAGE_SHIFT)
#define PAGE_MASK (~(PAGE_SIZE-1))

int main(int argc, char *argv[])
{
  int i;
  int fd;
  char *addr;
  unsigned long offset = 0;
  unsigned long left = 0;

  fd = open(argv[1], O_RDONLY);
  if (fd < 0) {
    perror("open");
    return 1;
  }

  if (argc > 2)
     sscanf(argv[2], "0x%lx", &offset);

  left = offset & (PAGE_SIZE - 1);
  offset &= PAGE_MASK;

  ioctl(fd, PCIIOC_MMAP_IS_MEM);
  addr = mmap(NULL, PAGE_SIZE, PROT_READ, MAP_SHARED, fd, offset);
  if (addr == MAP_FAILED) {
    perror("mmap");
    return 1;
  }

  printf("map finished\n");


  for (i = 0; i < 8; i++)
    printf("%x ", addr[i + left]);
  printf("\n");

  munmap(addr, PAGE_SIZE);

  close(fd);

  return 0;
}

[-- Attachment #2: new_pci_mmap_page_range_5_2_0.patch --]
[-- Type: text/x-patch, Size: 2530 bytes --]

From: Yinghai Lu <yinghai@kernel.org>
Subject: [PATCH v13.update.part1 01/16] PCI: Fix proc mmap on sparc

In 8c05cd08a7 ("PCI: fix offset check for sysfs mmapped files"), try
to check exposed value with resource start/end in proc mmap path.

|        start = vma->vm_pgoff;
|        size = ((pci_resource_len(pdev, resno) - 1) >> PAGE_SHIFT) + 1;
|        pci_start = (mmap_api == PCI_MMAP_PROCFS) ?
|                        pci_resource_start(pdev, resno) >> PAGE_SHIFT : 0;
|        if (start >= pci_start && start < pci_start + size &&
|                        start + nr <= pci_start + size)

vma->vm_pgoff is above code segment is user/BAR value >> PAGE_SHIFT.
pci_start is resource->start >> PAGE_SHIFT.

For sparc, resource start is different from BAR start aka pci bus address.
pci bus address need to add offset to be the resource start.

So that commit breaks all arch that exposed value is BAR/user value,
and need to be offseted to resource address.

test code using: ./test_mmap_proc /proc/bus/pci/0000:00/04.0 0x2000000
test code segment:
  fd = open(argv[1], O_RDONLY);
  ...
  sscanf(argv[2], "0x%lx", &offset);
  left = offset & (PAGE_SIZE - 1);
  offset &= PAGE_MASK;
  ioctl(fd, PCIIOC_MMAP_IS_MEM);
  addr = mmap(NULL, PAGE_SIZE, PROT_READ, MAP_SHARED, fd, offset);
  for (i = 0; i < 8; i++)
    printf("%x ", addr[i + left]);
  munmap(addr, PAGE_SIZE);
  close(fd);

Fixes: 8c05cd08a7 ("PCI: fix offset check for sysfs mmapped files")
Signed-off-by: Yinghai Lu <yinghai@kernel.org>

diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
index bcd10c7..e907154 100644
--- a/drivers/pci/pci-sysfs.c
+++ b/drivers/pci/pci-sysfs.c
@@ -974,15 +974,20 @@ void pci_remove_legacy_files(struct pci_bus *b)
 int pci_mmap_fits(struct pci_dev *pdev, int resno, struct vm_area_struct *vma,
 		  enum pci_mmap_api mmap_api)
 {
-	unsigned long nr, start, size, pci_start;
+	unsigned long nr, start, size, pci_start = 0;
 
 	if (pci_resource_len(pdev, resno) == 0)
 		return 0;
 	nr = vma_pages(vma);
 	start = vma->vm_pgoff;
 	size = ((pci_resource_len(pdev, resno) - 1) >> PAGE_SHIFT) + 1;
-	pci_start = (mmap_api == PCI_MMAP_PROCFS) ?
-			pci_resource_start(pdev, resno) >> PAGE_SHIFT : 0;
+	if (mmap_api == PCI_MMAP_PROCFS) {
+		resource_size_t user_start, user_end;
+
+		pci_resource_to_user(pdev, resno, &pdev->resource[resno],
+					&user_start, &user_end);
+		pci_start = user_start >> PAGE_SHIFT;
+	}
 	if (start >= pci_start && start < pci_start + size &&
 			start + nr <= pci_start + size)
 		return 1;

[-- Attachment #3: new_pci_mmap_page_range_5_2_1.patch --]
[-- Type: text/x-patch, Size: 9998 bytes --]

From: Yinghai Lu <yinghai@kernel.org>
Subject: [PATCH v13.update.part2 01/16] PCI: Let pci_mmap_page_range() take resource address

Original pci_mmap_page_range() is taking PCI BAR value aka usr_address.

Bjorn found out that it would be much simple to pass resource address
directly and avoid extra those __pci_mmap_make_offset.

In this patch:
1. in proc path: proc_bus_pci_mmap, try convert back to resource
   before calling pci_mmap_page_range
2. in sysfs path: pci_mmap_resource will just offset with resource start.
3. all pci_mmap_page_range will have vma->vm_pgoff with in resource
   range instead of BAR value.
4. skip calling __pci_mmap_make_offset, as the checking is done
   in pci_mmap_fits().

-v2: add pci_user_to_resource and remove __pci_mmap_make_offset
-v3: pass resource pointer with pci_mmap_page_range()
-v4: put __pci_mmap_make_offset() removing to following patch
     seperate /sys io access alignment checking to another patch
     updated after Bjorn's pci_resource_to_user() changes.
-v5: update after fix for pci_mmap with proc path accoring to
     Bjorn.

Signed-off-by: Yinghai Lu <yinghai@kernel.org>
Cc: linuxppc-dev@lists.ozlabs.org
Cc: sparclinux@vger.kernel.org
Cc: linux-xtensa@linux-xtensa.org

---
 arch/microblaze/pci/pci-common.c |   11 ++++---
 arch/powerpc/kernel/pci-common.c |   11 ++++---
 arch/sparc/kernel/pci.c          |    4 --
 arch/xtensa/kernel/pci.c         |   13 ++++++--
 drivers/pci/pci-sysfs.c          |   32 ++++++++++----------
 drivers/pci/pci.h                |    2 -
 drivers/pci/proc.c               |   60 +++++++++++++++++++++++++++++++++------
 7 files changed, 93 insertions(+), 40 deletions(-)

Index: linux-2.6/arch/microblaze/pci/pci-common.c
===================================================================
--- linux-2.6.orig/arch/microblaze/pci/pci-common.c
+++ linux-2.6/arch/microblaze/pci/pci-common.c
@@ -282,12 +282,15 @@ int pci_mmap_page_range(struct pci_dev *
 {
 	resource_size_t offset =
 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
-	struct resource *rp;
 	int ret;
 
-	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
-	if (rp == NULL)
-		return -EINVAL;
+	if (mmap_state == pci_mmap_io) {
+		struct pci_controller *hose = pci_bus_to_host(dev->bus);
+
+		/* hose should never be NULL */
+		offset += hose->io_base_phys -
+			 ((unsigned long)hose->io_base_virt - _IO_BASE);
+	}
 
 	vma->vm_pgoff = offset >> PAGE_SHIFT;
 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
Index: linux-2.6/arch/powerpc/kernel/pci-common.c
===================================================================
--- linux-2.6.orig/arch/powerpc/kernel/pci-common.c
+++ linux-2.6/arch/powerpc/kernel/pci-common.c
@@ -420,12 +420,15 @@ int pci_mmap_page_range(struct pci_dev *
 {
 	resource_size_t offset =
 		((resource_size_t)vma->vm_pgoff) << PAGE_SHIFT;
-	struct resource *rp;
 	int ret;
 
-	rp = __pci_mmap_make_offset(dev, &offset, mmap_state);
-	if (rp == NULL)
-		return -EINVAL;
+	if (mmap_state == pci_mmap_io) {
+		struct pci_controller *hose = pci_bus_to_host(dev->bus);
+
+		/* hose should never be NULL */
+		offset += hose->io_base_phys -
+			  ((unsigned long)hose->io_base_virt - _IO_BASE);
+	}
 
 	vma->vm_pgoff = offset >> PAGE_SHIFT;
 	if (write_combine)
Index: linux-2.6/arch/sparc/kernel/pci.c
===================================================================
--- linux-2.6.orig/arch/sparc/kernel/pci.c
+++ linux-2.6/arch/sparc/kernel/pci.c
@@ -868,10 +868,6 @@ int pci_mmap_page_range(struct pci_dev *
 {
 	int ret;
 
-	ret = __pci_mmap_make_offset(dev, vma, mmap_state);
-	if (ret < 0)
-		return ret;
-
 	__pci_mmap_set_pgprot(dev, vma, mmap_state);
 
 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
Index: linux-2.6/arch/xtensa/kernel/pci.c
===================================================================
--- linux-2.6.orig/arch/xtensa/kernel/pci.c
+++ linux-2.6/arch/xtensa/kernel/pci.c
@@ -366,11 +366,18 @@ int pci_mmap_page_range(struct pci_dev *
 			enum pci_mmap_state mmap_state,
 			int write_combine)
 {
+	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
 	int ret;
 
-	ret = __pci_mmap_make_offset(dev, vma, mmap_state);
-	if (ret < 0)
-		return ret;
+	if (mmap_state == pci_mmap_io) {
+		struct pci_controller *pci_ctrl =
+					 (struct pci_controller *)dev->sysdata;
+
+		/* pci_ctrl should never be NULL */
+		offset += pci_ctrl->io_space.start - pci_ctrl->io_space.base;
+	}
+
+	vma->vm_pgoff = offset >> PAGE_SHIFT;
 
 	__pci_mmap_set_pgprot(dev, vma, mmap_state, write_combine);
 
Index: linux-2.6/drivers/pci/pci-sysfs.c
===================================================================
--- linux-2.6.orig/drivers/pci/pci-sysfs.c
+++ linux-2.6/drivers/pci/pci-sysfs.c
@@ -972,22 +972,28 @@ void pci_remove_legacy_files(struct pci_
 #ifdef HAVE_PCI_MMAP
 
 int pci_mmap_fits(struct pci_dev *pdev, int resno, struct vm_area_struct *vma,
+		  enum pci_mmap_state mmap_type,
 		  enum pci_mmap_api mmap_api)
 {
 	unsigned long nr, start, size, pci_start = 0;
+	int flags;
 
 	if (pci_resource_len(pdev, resno) == 0)
 		return 0;
+
+	if (mmap_type == pci_mmap_mem)
+		flags = IORESOURCE_MEM;
+	else
+		flags = IORESOURCE_IO;
+
+	if (!(pci_resource_flags(pdev, resno) & flags))
+		return 0;
+
 	nr = vma_pages(vma);
 	start = vma->vm_pgoff;
 	size = ((pci_resource_len(pdev, resno) - 1) >> PAGE_SHIFT) + 1;
-	if (mmap_api == PCI_MMAP_PROCFS) {
-		resource_size_t user_start, user_end;
-
-		pci_resource_to_user(pdev, resno, &pdev->resource[resno],
-					&user_start, &user_end);
-		pci_start = user_start >> PAGE_SHIFT;
-	}
+	pci_start = (mmap_api == PCI_MMAP_PROCFS) ?
+			pci_resource_start(pdev, resno) >> PAGE_SHIFT : 0;
 	if (start >= pci_start && start < pci_start + size &&
 			start + nr <= pci_start + size)
 		return 1;
@@ -1009,7 +1015,6 @@ static int pci_mmap_resource(struct kobj
 	struct pci_dev *pdev = to_pci_dev(kobj_to_dev(kobj));
 	struct resource *res = attr->private;
 	enum pci_mmap_state mmap_type;
-	resource_size_t start, end;
 	int i;
 
 	for (i = 0; i < PCI_ROM_RESOURCE; i++)
@@ -1021,7 +1026,8 @@ static int pci_mmap_resource(struct kobj
 	if (res->flags & IORESOURCE_MEM && iomem_is_exclusive(res->start))
 		return -EINVAL;
 
-	if (!pci_mmap_fits(pdev, i, vma, PCI_MMAP_SYSFS)) {
+	mmap_type = res->flags & IORESOURCE_MEM ? pci_mmap_mem : pci_mmap_io;
+	if (!pci_mmap_fits(pdev, i, vma, mmap_type, PCI_MMAP_SYSFS)) {
 		WARN(1, "process \"%s\" tried to map 0x%08lx bytes at page 0x%08lx on %s BAR %d (start 0x%16Lx, size 0x%16Lx)\n",
 			current->comm, vma->vm_end-vma->vm_start, vma->vm_pgoff,
 			pci_name(pdev), i,
@@ -1030,13 +1036,7 @@ static int pci_mmap_resource(struct kobj
 		return -EINVAL;
 	}
 
-	/* pci_mmap_page_range() expects the same kind of entry as coming
-	 * from /proc/bus/pci/ which is a "user visible" value. If this is
-	 * different from the resource itself, arch will do necessary fixup.
-	 */
-	pci_resource_to_user(pdev, i, res, &start, &end);
-	vma->vm_pgoff += start >> PAGE_SHIFT;
-	mmap_type = res->flags & IORESOURCE_MEM ? pci_mmap_mem : pci_mmap_io;
+	vma->vm_pgoff += res->start >> PAGE_SHIFT;
 	return pci_mmap_page_range(pdev, vma, mmap_type, write_combine);
 }
 
Index: linux-2.6/drivers/pci/proc.c
===================================================================
--- linux-2.6.orig/drivers/pci/proc.c
+++ linux-2.6/drivers/pci/proc.c
@@ -227,24 +227,68 @@ static long proc_bus_pci_ioctl(struct fi
 }
 
 #ifdef HAVE_PCI_MMAP
+
+static int pci_user_to_resource(struct pci_dev *dev, resource_size_t *offset,
+				int flags)
+{
+	int i;
+
+	for (i = 0; i < PCI_ROM_RESOURCE; i++) {
+		resource_size_t start, end;
+		struct resource *res = &dev->resource[i];
+
+		if (!(res->flags & flags))
+			continue;
+
+		if (pci_resource_len(dev, i) == 0)
+			continue;
+
+		/*
+		 * here *offset is PAGE_SIZE aligned from caller.
+		 * need align start/end for io port resource that is
+		 * usually not PAGE_SIZE aligned.
+		 * that means we let it go if they falls in same page.
+		 */
+		pci_resource_to_user(dev, i, res, &start, &end);
+		if ((start & PAGE_MASK) <= *offset &&
+		     *offset <= (end & PAGE_MASK)) {
+			*offset = res->start + (*offset - start);
+			return i;
+		}
+	}
+
+	return -ENODEV;
+}
+
 static int proc_bus_pci_mmap(struct file *file, struct vm_area_struct *vma)
 {
 	struct pci_dev *dev = PDE_DATA(file_inode(file));
 	struct pci_filp_private *fpriv = file->private_data;
-	int i, ret, write_combine;
+	resource_size_t offset;
+	int i, ret, flags, write_combine;
 
 	if (!capable(CAP_SYS_RAWIO))
 		return -EPERM;
 
-	/* Make sure the caller is mapping a real resource for this device */
-	for (i = 0; i < PCI_ROM_RESOURCE; i++) {
-		if (pci_mmap_fits(dev, i, vma,  PCI_MMAP_PROCFS))
-			break;
-	}
-
-	if (i >= PCI_ROM_RESOURCE)
+	offset = vma->vm_pgoff << PAGE_SHIFT;
+	if (fpriv->mmap_state == pci_mmap_mem)
+		flags = IORESOURCE_MEM;
+	else
+		flags = IORESOURCE_IO;
+	i = pci_user_to_resource(dev, &offset, flags);
+	if (i < 0)
 		return -ENODEV;
 
+	vma->vm_pgoff = offset >> PAGE_SHIFT;
+	if (!pci_mmap_fits(dev, i, vma, fpriv->mmap_state, PCI_MMAP_PROCFS)) {
+		WARN(1, "process \"%s\" tried to map 0x%08lx bytes at page 0x%08lx on %s BAR %d (start 0x%16Lx, size 0x%16Lx)\n",
+			current->comm, vma->vm_end-vma->vm_start, vma->vm_pgoff,
+			pci_name(dev), i,
+			(u64)pci_resource_start(dev, i),
+			(u64)pci_resource_len(dev, i));
+		return -EINVAL;
+	}
+
 	if (fpriv->mmap_state == pci_mmap_mem)
 		write_combine = fpriv->write_combine;
 	else
Index: linux-2.6/drivers/pci/pci.h
===================================================================
--- linux-2.6.orig/drivers/pci/pci.h
+++ linux-2.6/drivers/pci/pci.h
@@ -30,7 +30,7 @@ enum pci_mmap_api {
 	PCI_MMAP_PROCFS	/* mmap on /proc/bus/pci/<BDF> */
 };
 int pci_mmap_fits(struct pci_dev *pdev, int resno, struct vm_area_struct *vmai,
-		  enum pci_mmap_api mmap_api);
+		  enum pci_mmap_state mmap_type, enum pci_mmap_api mmap_api);
 #endif
 int pci_probe_reset_function(struct pci_dev *dev);