Re: [PATCH v12 01/15] PCI: Let pci_mmap_page_range() take extra resource pointer

[Bjorn Helgaas <helgaas@kernel.org>]

On Fri, Jun 03, 2016 at 05:06:28PM -0700, Yinghai Lu wrote:
> This one is preparing patch for next one:
>   PCI: Let pci_mmap_page_range() take resource addr
> 
> We need to pass extra resource pointer to avoid searching that again
> for powerpc and microblaze prot set operation.

I'm not convinced yet that the extra resource pointer is necessary.

Microblaze does look up the resource in pci_mmap_page_range(), but it
never actually uses it.  It *looks* like it uses it, but that code is
actually dead and I think we should apply the first patch below.

That leaves powerpc as the only arch that would use this extra
resource pointer.  It uses it in __pci_mmap_set_pgprot() to help
decide whether to make a normal uncacheable mapping or a write-
combining one.  There's nothing here that's specific to the powerpc
architecture, and I don't think we should add this parameter just to
cater to powerpc.

There are two cases where __pci_mmap_set_pgprot() on powerpc does
something based on the resource:

  1) We're using procfs to mmap I/O port space after we requested
     write-combining, e.g., we did this:

       ioctl(fd, PCIIOC_MMAP_IS_IO);           # request I/O port space
       ioctl(fd, PCIIOC_WRITE_COMBINE, 1);     # request write-combining
       mmap(fd, ...)

     On powerpc, we ignore the write-combining request in this case.

     I think we can handle this case by applying the second patch
     below to ignore write-combining on I/O space for all arches, not
     just powerpc.

  2) We're using sysfs to mmap resourceN (not resourceN_wc), and
     the resource is prefetchable.  On powerpc, we turn *on*
     write-combining, even though the user didn't ask for it.

     I'm not sure this case is actually safe, because it changes the
     ordering properties.  If it *is* safe, we could enable write-
     combining in pci_mmap_resource(), where we already have the
     resource and it could be done for all arches.

     This case is not strictly necessary, except to avoid a
     performance regression, because the user could have mapped
     resourceN_wc to explicitly request write-combining.

> diff --git a/drivers/pci/pci-sysfs.c b/drivers/pci/pci-sysfs.c
> index d319a9c..5bbe20c 100644
> --- a/drivers/pci/pci-sysfs.c
> +++ b/drivers/pci/pci-sysfs.c
> @@ -1027,7 +1027,7 @@ static int pci_mmap_resource(struct kobject *kobj, struct bin_attribute *attr,
>  	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;
> -	return pci_mmap_page_range(pdev, vma, mmap_type, write_combine);
> +	return pci_mmap_page_range(pdev, res, vma, mmap_type, write_combine);
>  }
>  
>  static int pci_mmap_resource_uc(struct file *filp, struct kobject *kobj,
> diff --git a/drivers/pci/proc.c b/drivers/pci/proc.c
> index 3f155e7..f19ee2a 100644
> --- a/drivers/pci/proc.c
> +++ b/drivers/pci/proc.c
> @@ -245,7 +245,7 @@ static int proc_bus_pci_mmap(struct file *file, struct vm_area_struct *vma)
>  	if (i >= PCI_ROM_RESOURCE)
>  		return -ENODEV;
>  
> -	ret = pci_mmap_page_range(dev, vma,
> +	ret = pci_mmap_page_range(dev, &dev->resource[i], vma,
>  				  fpriv->mmap_state,
>  				  fpriv->write_combine);
>  	if (ret < 0)
> diff --git a/include/linux/pci.h b/include/linux/pci.h
> index b67e4df..3c1a0f4 100644
> --- a/include/linux/pci.h
> +++ b/include/linux/pci.h
> @@ -70,6 +70,12 @@ enum pci_mmap_state {
>  	pci_mmap_mem
>  };
>  
> +struct vm_area_struct;
> +/* Map a range of PCI memory or I/O space for a device into user space */
> +int pci_mmap_page_range(struct pci_dev *dev, struct resource *res,
> +			struct vm_area_struct *vma,
> +			enum pci_mmap_state mmap_state, int write_combine);
> +
>  /*
>   *  For PCI devices, the region numbers are assigned this way:
>   */


commit 4e712b691abc5b579e3e4327f56b0b7988bdd1cb
Author: Bjorn Helgaas <bhelgaas@google.com>
Date:   Wed Jun 8 14:00:14 2016 -0500

    microblaze/PCI: Remove useless __pci_mmap_set_pgprot()
    
    The microblaze __pci_mmap_set_pgprot() was apparently copied from powerpc,
    where it computes either an uncacheable pgprot_t or a write-combining one.
    But on microblaze, we always use the regular uncacheable pgprot_t.
    
    Remove the useless code in __pci_mmap_set_pgprot() and inline the
    pgprot_noncached() at the only caller.
    
    Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>

diff --git a/arch/microblaze/pci/pci-common.c b/arch/microblaze/pci/pci-common.c
index 14cba60..1974567 100644
--- a/arch/microblaze/pci/pci-common.c
+++ b/arch/microblaze/pci/pci-common.c
@@ -219,33 +219,6 @@ static struct resource *__pci_mmap_make_offset(struct pci_dev *dev,
 }
 
 /*
- * Set vm_page_prot of VMA, as appropriate for this architecture, for a pci
- * device mapping.
- */
-static pgprot_t __pci_mmap_set_pgprot(struct pci_dev *dev, struct resource *rp,
-				      pgprot_t protection,
-				      enum pci_mmap_state mmap_state,
-				      int write_combine)
-{
-	pgprot_t prot = protection;
-
-	/* Write combine is always 0 on non-memory space mappings. On
-	 * memory space, if the user didn't pass 1, we check for a
-	 * "prefetchable" resource. This is a bit hackish, but we use
-	 * this to workaround the inability of /sysfs to provide a write
-	 * combine bit
-	 */
-	if (mmap_state != pci_mmap_mem)
-		write_combine = 0;
-	else if (write_combine = 0) {
-		if (rp->flags & IORESOURCE_PREFETCH)
-			write_combine = 1;
-	}
-
-	return pgprot_noncached(prot);
-}
-
-/*
  * 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
@@ -317,9 +290,7 @@ int pci_mmap_page_range(struct pci_dev *dev, struct vm_area_struct *vma,
 		return -EINVAL;
 
 	vma->vm_pgoff = offset >> PAGE_SHIFT;
-	vma->vm_page_prot = __pci_mmap_set_pgprot(dev, rp,
-						  vma->vm_page_prot,
-						  mmap_state, write_combine);
+	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
 
 	ret = remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
 			       vma->vm_end - vma->vm_start, vma->vm_page_prot);



commit 962972ee5e0ba6ceb680cb182bad65f8886586a6
Author: Bjorn Helgaas <bhelgaas@google.com>
Date:   Wed Jun 8 14:46:54 2016 -0500

    PCI: Ignore write-combining when mapping I/O port space
    
    PCI exposes files like /proc/bus/pci/00/00.0 in procfs.  These files
    support operations like this:
    
      ioctl(fd, PCIIOC_MMAP_IS_IO);           # request I/O port space
      ioctl(fd, PCIIOC_WRITE_COMBINE, 1);     # request write-combining
      mmap(fd, ...)
    
    Many architectures don't allow mmap of I/O port space at all, but I don't
    think it makes sense to do a write-combining mapping on the ones that do.
    We could change proc_bus_pci_ioctl() so the user could never enable write-
    combining for I/O port space, but that would break the following sequence,
    which is currently legal:
    
      mmap(fd, ...)                           # default is I/O, non-combining
      ioctl(fd, PCIIOC_WRITE_COMBINE, 1);     # request write-combining
      ioctl(fd, PCIIOC_MMAP_IS_MEM);          # request memory space
      mmap(fd, ...)
    
    Ignore the write-combining flag when mapping I/O port space.
    
    Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>

diff --git a/drivers/pci/proc.c b/drivers/pci/proc.c
index 3f155e7..21f8d613 100644
--- a/drivers/pci/proc.c
+++ b/drivers/pci/proc.c
@@ -247,7 +247,8 @@ static int proc_bus_pci_mmap(struct file *file, struct vm_area_struct *vma)
 
 	ret = pci_mmap_page_range(dev, vma,
 				  fpriv->mmap_state,
-				  fpriv->write_combine);
+				  (fpriv->mmap_state = pci_mmap_mem) ?
+					fpriv->write_combine : 0);
 	if (ret < 0)
 		return ret;