]> git.baikalelectronics.ru Git - kernel.git/commitdiff
sparc: Use generic pci_mmap_resource_range()
authorArnd Bergmann <arnd@arndb.de>
Fri, 15 Jul 2022 15:36:17 +0000 (17:36 +0200)
committerBjorn Helgaas <bhelgaas@google.com>
Fri, 29 Jul 2022 17:09:32 +0000 (12:09 -0500)
The main feature of the sparc-specific implementation of
pci_mmap_resource_range() is that it allows mapping the entire PCI I/O
space for a PCI host bridge using the /proc/bus/pci interface on a bridge
device.

The generic implementation cannot do this, but it also appears that this
got broken for sparc by commit 003a00ffabd9 ("PCI: check mmap range of
/proc/bus/pci files too"), which enforces that each address is part of a
BAR for kernels after 2.6.28.

Remove it all, assuming that the corresponding user space code has already
been changed to access /dev/ioport instead a long time ago.  Add
pci_iobar_pfn() to make it possible to map I/O resources. This is adapted
from the powerpc version.

Link: https://lore.kernel.org/lkml/1519887203.622.3.camel@infradead.org/t/
Link: https://lore.kernel.org/r/20220715153617.3393420-2-arnd@kernel.org
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
arch/sparc/include/asm/pci.h
arch/sparc/kernel/pci.c

index 4deddf430e5d18cae2710cf1d74ebdb2aaf170a6..dff90dce6cb71f51b05520742cece4d8684edb9b 100644 (file)
@@ -37,6 +37,7 @@ static inline int pci_proc_domain(struct pci_bus *bus)
 #define HAVE_PCI_MMAP
 #define arch_can_pci_mmap_io() 1
 #define HAVE_ARCH_PCI_GET_UNMAPPED_AREA
+#define ARCH_GENERIC_PCI_MMAP_RESOURCE
 #define get_pci_unmapped_area get_fb_unmapped_area
 #endif /* CONFIG_SPARC64 */
 
index f580db840bf704ed8a08ce8f520fa9402849d65a..cb1ef25116e943436c83d68ae033dbb7d7161b25 100644 (file)
@@ -751,161 +751,15 @@ 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)
+int pci_iobar_pfn(struct pci_dev *pdev, int bar, struct vm_area_struct *vma)
 {
        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;
+       resource_size_t ioaddr = pci_resource_start(pdev, bar);
 
-       if (user_offset >= space_size ||
-           (user_offset + user_size) > space_size)
+       if (!pbm)
                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.
- */
-static void __pci_mmap_set_pgprot(struct pci_dev *dev, struct vm_area_struct *vma,
-                                            enum pci_mmap_state mmap_state)
-{
-       /* Our io_remap_pfn_range takes care of this, do nothing.  */
-}
-
-/* Perform the actual remap of the pages for a PCI device mapping, as appropriate
- * for this architecture.  The region in the process to map is described by vm_start
- * and vm_end members of VMA, the BAR relative address is found in vm_pgoff.
- * The pci device structure is provided so that architectures may make mapping
- * decisions on a per-device or per-bus basis.
- *
- * Returns a negative error code on failure, zero on success.
- */
-int pci_mmap_resource_range(struct pci_dev *dev, int bar,
-                           struct vm_area_struct *vma,
-                           enum pci_mmap_state mmap_state, int write_combine)
-{
-       int ret;
-       resource_size_t start, end;
-
-       /* convert per-BAR address to PCI bus address */
-       pci_resource_to_user(dev, bar, &dev->resource[bar], &start, &end);
-       vma->vm_pgoff += start >> PAGE_SHIFT;
-
-       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);
-       ret = io_remap_pfn_range(vma, vma->vm_start,
-                                vma->vm_pgoff,
-                                vma->vm_end - vma->vm_start,
-                                vma->vm_page_prot);
-       if (ret)
-               return ret;
+       vma->vm_pgoff += (ioaddr + pbm->io_space.start) >> PAGE_SHIFT;
 
        return 0;
 }