Very Computer

In Linux I see they can use phys_to_virt() and virt_to_phys() to map
physical memory to/from virtual memory.

I have dug into source and found some BIOS conversions, but not generic
conversions.

How do I do these conversions in FreeBSD?

thanks

--

-- Steve --

Quote:> I am writing KLD driver and want to access physical memory. I assume I will
> need to convert to/from virtual memory from looking at some Linux code.

> In Linux I see they can use phys_to_virt() and virt_to_phys() to map
> physical memory to/from virtual memory.

> I have dug into source and found some BIOS conversions, but not generic
> conversions.

> How do I do these conversions in FreeBSD?

I suggest looking through the pmap module, and hunt for symbols
like pmap_extract or vtophys.  It might have changed.

IN a general case, phys_to_virt makes less sense, because a signal page
can be mapped to multiple virtual addresses, but the specific case of an
individual processes mapping or kernel mapping is possible, and there has
been such a call in the code.

Actually, the pmap code isn't that complicated, and much of it in the area
that you are talking about isn't going to change alot...  As in Linux, the source
code is your friend :-).

I am giving you this answer now, just in case someone else dosesn't give you
more accurate or complete info!!!   If someone else gives you better info,
then listen to them, not me!!!  :-).

John

thanks again, nothing is easy, but anymore help would be great. If not I
will plod along...

steve

Quote:> > I am writing KLD driver and want to access physical memory. I assume I
will
> > need to convert to/from virtual memory from looking at some Linux code.

> > In Linux I see they can use phys_to_virt() and virt_to_phys() to map
> > physical memory to/from virtual memory.

> > I have dug into source and found some BIOS conversions, but not generic
> > conversions.

> > How do I do these conversions in FreeBSD?

> Even though I wrote alot of the code, I haven't looked at it for a few
years.

> I suggest looking through the pmap module, and hunt for symbols
> like pmap_extract or vtophys.  It might have changed.

> IN a general case, phys_to_virt makes less sense, because a signal page
> can be mapped to multiple virtual addresses, but the specific case of an
> individual processes mapping or kernel mapping is possible, and there has
> been such a call in the code.

> Actually, the pmap code isn't that complicated, and much of it in the area
> that you are talking about isn't going to change alot...  As in Linux, the
source
> code is your friend :-).

> I am giving you this answer now, just in case someone else dosesn't give
you
> more accurate or complete info!!!   If someone else gives you better info,
> then listen to them, not me!!!  :-).

> John

was the physical memory range mapped into the virtual address space by a
driver? It sounds like you are doing the begining of a device driver
and you have not mapped the physical memory into kernel virtual space
yet. when you do, this is the best time to remember the physical
and virtual addresses.

For the next step of using the physical address found in the first step,
the same question is asked has this physical address within the range of
the above mapping, if not then that has to be mapped to a virtual address.

Once the Virtual <> Physical memory has been mapped and you know the
PTE/PDE and pmap, you can go back and forth between physical and virtual
memory addresses, but it is much easier to remember in a variable.

There are also issues of the new fancy I/O mapping code, that others
know more about that I knew at all (or remember.)

So, when normally allocating memory, the seperation between virtual
and physical allocation is somewhat hidden.   When allocating physical
space, then allocating the viritual memory is done, then the mappings
explicitly forced.

It is (or used to be) okay to do a hard kernel mapping, since kernel
mappings mostly don't change from process to process (at least,
normal kernel mappings.)   pmap_enter is more useful when pages
should be managed for paging and things like that.   pmap_kenter
is kind of a hard-mapping.

If I get around to it, I'll look at the code to see what is going on, but
again, there are people who are MUCH MORE familiar with current
practices!!!

John

Quote:> physical addresses are not automatically mapped into kernel virtual space.
> on startup, the kernel maps space for itself and some data structures,
> but individual device drivers should map the appropriate BIOS/DMA/bus
> register areas.

> was the physical memory range mapped into the virtual address space by a
> driver? It sounds like you are doing the begining of a device driver
> and you have not mapped the physical memory into kernel virtual space
> yet. when you do, this is the best time to remember the physical
> and virtual addresses.

thanks
Steve

...frag...

vm_offset_t SMAPI_virtual;

/* get 128K allocated by the kernel */
SMAPI_virtual = kmem_alloc_nofault(kernel_map, 0x20000);

/* now map the physical onto this allocation */
pmap_map(SMAPI_virtual, 0xE0000, 0xFFFFF, 0);

/* now SMAPI_virtual is my virtual offset in VM which is mapped <> with phys
*/

--

-- Steve --

Quote:>I hate to give up and I like the challenge.. so .. here is what I have
>cobbled together.. any comments?

>...frag...

>vm_offset_t SMAPI_virtual;

>/* get 128K allocated by the kernel */
>SMAPI_virtual = kmem_alloc_nofault(kernel_map, 0x20000);

>/* now map the physical onto this allocation */
>pmap_map(SMAPI_virtual, 0xE0000, 0xFFFFF, 0);

>/* now SMAPI_virtual is my virtual offset in VM which is mapped <> with phys
>*/

        va = pmap_mapdev(0xe0000, 0x2000);
                ...
        pmap_unmapdev(va, 0x2000);

--mark tinguely.