Very Computer

Hi:

I've tried to write a solaris 2.1 X86 mmap driver that allows me to mmap my Mach32
ATI's framebuffer region into user space. The framebuffer region is 4megs in
size. Not being the Solaris kernel guru, I have the following questions:

1. I've noticed that when I try to mmap(2) the 4Meg region, my device driver xxmmap(9E)
gets called for each 4k page of the 4Meg region.  My xxmmap(9E) routine then calls
ddi_mmap_regs(9F). Each ddi_map_reg() call has the "size" parameter set to the total
register length I've specified in my driver.conf(4).  Why is the kernel calling my xxmmap(9E)
routine that many times?  Should I be calling ddi_map_regs(9F) on each invocation of
xxmmap(9E)?

2. When my test program does a munmap(2) call, how do I relate that to a driver function
that should do a ddi_unmap_regs(9F)? I'd rather not put the ddi_unmap_reg(9F) in my
xxclose(9E) routine as many legitimate uses of mmaping involve closing the device
after the mmap(2) is established. If there is no way to relate munmap(2) to any driver
routine, then how does the kernel clean up the region?  What is ddi_unmap_reg(9F) then
used for?

Any help/comments/suggestions are greatly appreciated.

Doug Anson
--
-------------------------------------------
Doug Anson                                      

Phone:    713.560.1274
FAX:      713.560.1277
SNAIL:    Landmark Graphics Corporation LGC    
          15150 Memorial Drive
          Houston, TX 77079                    
-------------------------------------------

The manual page mmap(9E) states:      
     An mmap() routine checks if the offset is within  the  range
     of pages supported 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 page at offset off in the device's
     memory.

So it calls you once for each page by design.  

Quote:>  Should I be calling ddi_map_regs(9F) on each invocation of
>xxmmap(9E)?

Probably not; I suspect what the above implies is that you call
ddi_map_regs(9F) for the whole card in your attach(9E) entry point, and then
use hat_getkpfnum(9F) to return the physical page.  I won't guarantee
this will work, but it's what I'd try first.

You might want to get the DDK; it talks about memory-mapping drivers.

Quote:

>2. When my test program does a munmap(2) call, how do I relate that to a driver function
>that should do a ddi_unmap_regs(9F)? I'd rather not put the ddi_unmap_reg(9F) in my
>xxclose(9E) routine as many legitimate uses of mmaping involve closing the device
>after the mmap(2) is established. If there is no way to relate munmap(2) to any driver
>routine, then how does the kernel clean up the region?  

The user mapping disappears; your driver's kernel mapping remains until
the driver is unloaded.  Several users may map your device at the same
time.

Quote:> What is ddi_unmap_reg(9F) then
>used for?

When you detach(9E) and undo everything done in attach(9E).

Quote:>Any help/comments/suggestions are greatly appreciated.

Hope it helped.