mmaping /dev/mem

mmaping /dev/mem

[sridhar vaidyanathan]

I have a problem mmaping /dev/mem on some address in RAM. I am aware of 
caveats,but I am trying to mmap a region which I am sure is not in use by the 
kernel(some additional code does this and returns a physical address which is 
what I use for mmap). The mmap call itself succeeds and /proc/pid/maps also 
shows that region, but I am unable to see what I write in target memory.I also 
tried with the O_SYNC flag as I was wondering is caching had anything to do 
with the results that I was seeing.No effect.
This however works with a mem= option and when the mmap region falls out of the 
mem= boundary.
any clues?
Please cc as I am not subscribed
sridhar

Re: mmaping /dev/mem

[Alex Tomas]

>>>>> sridhar vaidyanathan (sv) writes:

 sv> I have a problem mmaping /dev/mem on some address in RAM. I am
 sv> aware of caveats,but I am trying to mmap a region which I am sure
 sv> is not in use by the kernel(some additional code does this and
 sv> returns a physical address which is what I use for mmap). The
 sv> mmap call itself succeeds and /proc/pid/maps also shows that
 sv> region, but I am unable to see what I write in target memory.I
 sv> also tried with the O_SYNC flag as I was wondering is caching had
 sv> anything to do with the results that I was seeing.No effect.
 sv> This however works with a mem= option and when the mmap region
 sv> falls out of the mem= boundary.  any clues?  Please cc as I am
 sv> not subscribed sridhar

look at mm/memory.c:remap_pte_range():

		if ((!VALID_PAGE(page)) || PageReserved(page))
 			set_pte(pte, mk_pte_phys(phys_addr, prot));

so, your pages aren't mapped at all and pte'es contain zero ...