illumos: manual page: mmap.9e

MMAP(9E) Driver Entry Points MMAP(9E)

NAME

mmap - check virtual mapping for memory mapped device

SYNOPSIS

#include <sys/types.h>
#include <sys/cred.h>
#include <sys/mman.h>
#include <sys/ddi.h>

int prefixmmap(dev_t dev, off_t off, int prot);

INTERFACE LEVEL

This interface is obsolete. devmap(9E) should be used instead.

PARAMETERS

dev
Device whose memory is to be mapped.

off
Offset within device memory at which mapping begins.

prot
A bit field that specifies the protections this page of memory
will receive. Possible settings are:

PROT_READ
Read access will be granted.

PROT_WRITE
Write access will be granted.

PROT_EXEC
Execute access will be granted.

PROT_USER
User-level access will be granted.

PROT_ALL
All access will be granted.

DESCRIPTION

Future releases of Solaris will provide this function for binary and
source compatibility. However, for increased functionality, use
devmap(9E) instead. See devmap(9E) for details.

The mmap() entry point is a required entry point for character drivers
supporting memory-mapped devices. A memory mapped device has memory that
can be mapped into a process's address space. The mmap(2) system call,
when applied to a character special file, allows this device memory to be
mapped into user space for direct access by the user application.

The mmap() entry point is called as a result of an mmap(2) system call,
and also as a result of a page fault. mmap() is called to translate the
offset off in device memory to the corresponding physical page frame
number.

The mmap() entry point checks if the offset off is within the range of
pages exported by the device. For example, a device that has 512 bytes of
memory that can be mapped into user space should not support offsets
greater than 512. If the offset does not exist, then -1 is returned. If
the offset does exist, mmap() returns the value returned by
hat_getkpfnum(9F) for the physical page in device memory containing the
offset off.

hat_getkpfnum(9F) accepts a kernel virtual address as an argument. A
kernel virtual address can be obtained by calling ddi_regs_map_setup(9F)
in the driver's attach(9E) routine. The corresponding
ddi_regs_map_free(9F) call can be made in the driver's detach(9E)
routine. Refer to the example below mmap Entry Point for more
information.

mmap() should only be supported for memory-mapped devices. See segmap(9E)
for further information on memory-mapped device drivers.

If a device driver shares data structures with the application, for
example through exported kernel memory, and the driver gets recompiled
for a 64-bit kernel but the application remains 32-bit, the binary layout
of any data structures will be incompatible if they contain longs or
pointers. The driver needs to know whether there is a model mismatch
between the current thread and the kernel and take necessary action.
ddi_mmap_get_model(9F) can be use to get the C Language Type Model which
the current thread expects. In combination with
ddi_model_convert_from(9F) the driver can determine whether there is a
data model mismatch between the current thread and the device driver. The
device driver might have to adjust the shape of data structures before
exporting them to a user thread which supports a different data model.
See ddi_mmap_get_model(9F) for an example.

RETURN VALUES

If the protection and offset are valid for the device, the driver should
return the value returned by hat_getkpfnum(9F), for the page at offset
off in the device's memory. If not, -1 should be returned.

EXAMPLES

Example 1 mmap() Entry Point

The following is an example of the mmap() entry point. If offset off is
valid, hat_getkpfnum(9F) is called to obtain the page frame number
corresponding to this offset in the device's memory. In this example,
xsp->regp->csr is a kernel virtual address which maps to device memory.
ddi_regs_map_setup(9F) can be used to obtain this address. For example,
ddi_regs_map_setup(9F) can be called in the driver's attach(9E) routine.
The resulting kernel virtual address is stored in the xxstate structure,
which is accessible from the driver's mmap() entry point. See
ddi_soft_state(9F). The corresponding ddi_regs_map_free(9F) call can be
made in the driver's detach(9E) routine.

struct reg {
uint8_t csr;
uint8_t data;
};
struct xxstate {
...
struct reg *regp
...
};

struct xxstate *xsp;
...

static int
xxmmap(dev_t dev, off_t off, int prot)
{
int instance;
struct xxstate *xsp;

/* No write access */
if (prot & PROT_WRITE)
return (-1);

instance = getminor(dev);
xsp = ddi_get_soft_state(statep, instance);
if (xsp == NULL)
return (-1);

/* check for a valid offset */
if ( off is invalid )
return (-1);
return (hat_getkpfnum (xsp->regp->csr + off));
}

ATTRIBUTES

NOTES

For some devices, mapping device memory in the driver's attach(9E)
routine and unmapping device memory in the driver's detach(9E) routine is
a sizeable drain on system resources. This is especially true for devices
with a large amount of physical address space.

One alternative is to create a mapping for only the first page of device
memory in attach(9E). If the device memory is contiguous, a kernel page
frame number may be obtained by calling hat_getkpfnum(9F) with the kernel
virtual address of the first page of device memory and adding the desired
page offset to the result. The page offset may be obtained by converting
the byte offset off to pages. See ddi_btop(9F).

Another alternative is to call ddi_regs_map_setup(9F) and
ddi_regs_map_free(9F) in mmap(). These function calls would bracket the
call to hat_getkpfnum(9F).

However, note that the above alternatives may not work in all cases. The
existence of intermediate nexus devices with memory management unit
translation resources that are not locked down may cause unexpected and
undefined behavior.

September 27, 2002 MMAP(9E)