PMAP(9) OpenBSD Kernel Manual PMAP(9)NAME
pmap - machine dependent interface to the MMU
The architecture-dependent pmap module describes how the physical mapping
is done between the user-processes and kernel virtual addresses and the
physical addresses of the main memory, providing machine-dependent
translation and access tables that are used directly or indirectly by the
memory-management hardware. The pmap layer can be viewed as a big array
of mapping entries that are indexed by virtual address to produce a
physical address and flags. These flags describe the page's protection,
whether the page has been referenced or modified and other
The pmap interface is consistent across all platforms and hides the way
page mappings are stored.
The pmap_init() function is called from the machine-independent uvm(9)
initialization code, when the MMU is enabled.
Modified/referenced information is only tracked for pages managed by
uvm(9) (pages for which a vm_page structure exists). Only managed
mappings of those pages have modified/referenced tracking. The use of
unmanaged mappings should be limited to code which may execute in
interrupt context (such as malloc(9)) or to enter mappings for physical
addresses which are not managed by uvm(9). This allows pmap modules to
avoid blocking interrupts when manipulating data structures or holding
locks. Unmanaged mappings may only be entered into the kernel's virtual
address space. The modified/referenced bits must be tracked on a per-
page basis, as they are not attributes of a mapping, but attributes of a
page. Therefore, even after all mappings for a given page have been
removed, the modified/referenced bits for that page must be preserved.
The only time the modified/referenced bits may be cleared is when uvm(9)
explicitly calls the pmap_clear_modify() and pmap_clear_reference()
functions. These functions must also change any internal state necessary
to detect the page being modified or referenced again after the
modified/referenced state is cleared.
Mappings entered by pmap_enter() are managed, mappings entered by
pmap_kenter_pa() are not.
pmap_enter(pmap_t pmap, vaddr_t va, paddr_t pa, vm_prot_t prot, int
pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot);
pmap_remove(pmap_t pmap, vaddr_t sva, vaddr_t eva);
pmap_kremove(vaddr_t va, vsize_t size);
The pmap_enter() function creates a managed mapping for physical page pa
at the specified virtual address va in the target physical map pmap with
protection specified by prot:
VM_PROT_READ The mapping must allow reading.
VM_PROT_WRITE The mapping must allow writing.
VM_PROT_EXECUTE The page mapped contains instructions that will be
executed by the processor.
The flags argument contains protection bits (the same bits used in the
prot argument) indicating the type of access that caused the mapping to
be created. This information may be used to seed modified/referenced
information for the page being mapped, possibly avoiding redundant faults
on platforms that track modified/referenced information in software.
Other information provided by flags:
PMAP_WIRED The mapping being created is a wired mapping.
PMAP_CANFAIL The call to pmap_enter() is allowed to fail. If this flag
is not set, and the pmap_enter() call is unable to create
the mapping, perhaps due to insufficient resources, the
pmap module must panic.
The access type provided in the flags argument will never exceed the
protection specified by prot.
The pmap_enter() function is called by the fault routine to establish a
mapping for the page being faulted in. If pmap_enter() is called to
enter a mapping at a virtual address for which a mapping already exists,
the previous mapping must be invalidated. pmap_enter() is sometimes
called to change the protection for a pre-existing mapping, or to change
the ``wired'' attribute for a pre-existing mapping.
The pmap_kenter_pa() function creates an unmanaged mapping of physical
address pa at the specified virtual address va with the protection
specified by prot.
The pmap_remove() function removes the range of virtual addresses sva to
eva from pmap, assuming proper alignment. pmap_remove() is called during
an unmap operation to remove low-level machine dependent mappings.
The pmap_kremove() function removes an unmanaged mapping at virtual
address va of size size.
A call to pmap_update() must be made after pmap_kenter_pa() or
pmap_kremove() to notify the pmap layer that the mappings need to be made
pmap_unwire(pmap_t pmap, vaddr_t va);
pmap_protect(pmap_t pmap, vaddr_t sva, vaddr_t eva, vm_prot_t prot);
pmap_page_protect(struct vm_page *pg, vm_prot_t prot);
The pmap_unwire() function clears the wired attribute for a map/virtual-
address pair. The mapping must already exist in pmap.
The pmap_protect() function sets the physical protection on range sva to
eva, in pmap.
The pmap_protect() function is called during a copy-on-write operation to
write protect copy-on-write memory, and when paging out a page to remove
all mappings of a page. The pmap_page_protect() function sets the
permission for all mapping to page pg. The pmap_page_protect() function
is called before a pageout operation to ensure that all pmap references
to a page are removed.
PHYSICAL PAGE-USAGE INFORMATION
pmap_is_modified(struct vm_page *pg);
pmap_clear_modify(struct vm_page *pg);
pmap_is_referenced(struct vm_page *pg);
pmap_clear_reference(struct vm_page *pg);
The pmap_is_modified() and pmap_clear_modify() functions read/set the
modify bits on the specified physical page pg. The pmap_is_referenced()
and pmap_clear_reference() functions read/set the reference bits on the
specified physical page pg.
The pmap_is_referenced() and pmap_is_modified() functions are called by
the pagedaemon when looking for pages to free. The
pmap_clear_referenced() and pmap_clear_modify() functions are called by
the pagedaemon to help identification of pages that are no longer in
PHYSICAL PAGE INITIALIZATION
pmap_copy_page(struct vm_page *src, struct vm_page *dst);
pmap_zero_page(struct vm_page *page);
The pmap_copy_page() function copies the content of the physical page src
to physical page dst.
The pmap_zero_page() function fills page with zeroes.
INTERNAL DATA STRUCTURE MANAGEMENT
The pmap_create() function creates an instance of the pmap structure.
The pmap_reference() function increments the reference count on pmap.
The pmap_destroy() function decrements the reference count on physical
map pmap and retires it from service if the count drops to zero, assuming
it contains no valid mappings.
pmap_steal_memory(vsize_t size, vaddr_t *vstartp, vaddr_t *vendp);
pmap_virtual_space(vaddr_t *vstartp, vaddr_t *vendp);
pmap_copy(pmap_t dst_pmap, pmap_t src_pmap, vaddr_t dst_addr, vsize_t
len, vaddr_t src_addr);
Wired memory allocation before the virtual memory system is bootstrapped
is accomplished by the pmap_steal_memory() function. After that point,
the kernel memory allocation routines should be used.
The pmap_growkernel() function can preallocate kernel page tables to a
specified virtual address.
The pmap_update() function notifies the pmap module to force processing
of all delayed actions for all pmaps.
The pmap_collect() function informs the pmap module that the given pmap
is not expected to be used for some time, giving the pmap module a chance
to prioritize. The initial bounds of the kernel virtual address space
are returned by pmap_virtual_space().
The pmap_copy() function copies the range specified by src_addr and
src_len from src_pmap to the range described by dst_addr and dst_len in
dst_map. pmap_copy() is called during a fork(2) operation to give the
child process an initial set of low-level mappings.
SEE ALSOfork(2), uvm(9)HISTORY
The 4.4BSD pmap module is based on Mach 3.0. The introduction of uvm(9)
left the pmap interface unchanged for the most part.
Ifdefs must be documented.
pmap_update() should be mandatory.
OpenBSD 4.9 December 31, 2008 OpenBSD 4.9