PFN to Base Address

[Bithika Khargharia]

Hi,

1. Given the PFN, how do I obtain the base physical address for the PFN.

I use
             'pfn <<12'.

Is this correct?

2. Does the kernel generate PFN serailly from 0 to TOTAL_PHYS or are there breaks in between?

Thanks
Bithika

[Fawad Lateef]

On 11/15/06, Bithika Khargharia <bit...@gmail.com> wrote:
> Hi,
>
> 1. Given the PFN, how do I obtain the base physical address for the PFN.
>
> I use
> 'pfn <<12'.
>
> Is this correct?

Yes 'pfn << PAGE_SHIFT' gives physical-base-address although its
better to use page_to_phys macro

>
> 2. Does the kernel generate PFN serailly from 0 to TOTAL_PHYS or are there
> breaks in between?
>

Yes serially for x86 from 0UL but might contain some starting offset
according to some architectures. Take a look at page_to_phys macro
(http://lxr.free-electrons.com/source/include/asm-i386/io.h#101) and
__page_to_pfn macro
(http://lxr.free-electrons.com/source/include/asm-generic/memory_model.h#032)

--
Fawad Lateef

--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive: http://mail.nl.linux.org/kernelnewbies/
FAQ: http://kernelnewbies.org/faq/

[Bithika Khargharia]

Hi,

1. I checked the ARCH_PFN_OFFSET variable for my kernel and it was set to 0. In that case the kernel should generate PFN serially from 0 to TOTAL_PHYS. Am I right?

2. What confuses me is the following. I know that I have 8 GB of memory physically installed on my System. 8 GB should translate to a PFN of 2097152
(8 * 2^30)/4 where 4KB is the PAGE_SIZE in my system. However the kernel gives a TOTAL_PHYS value of 2411724 which is almost equal to 9.2 GB of memory. If PFNs are generated serially why do I have this problem of higher PFNs that 8 GB?

Thanks
Bithika

On 11/15/06, Fawad Lateef <fawad...@gmail.com> wrote:

On 11/15/06, Bithika Khargharia <bit...@gmail.com> wrote:
> Hi,
>
> 1. Given the PFN, how do I obtain the base physical address for the PFN.
>
> I use
>              'pfn <<12'.
>
> Is this correct?

Yes 'pfn << PAGE_SHIFT' gives physical-base-address although its
better to use page_to_phys macro

>
> 2. Does the kernel generate PFN serailly from 0 to TOTAL_PHYS or are there
> breaks in between?
>

Yes serially for x86 from 0UL but might contain some starting offset
according to some architectures. Take a look at page_to_phys macro

( http://lxr.free-electrons.com/source/include/asm-i386/io.h#101) and

__page_to_pfn macro
(http://lxr.free-electrons.com/source/include/asm-generic/memory_model.h#032 )


--
Fawad Lateef

--
O for a Life of Sensations rather than of Thoughts!-- Keats

[Fawad Lateef]

On 11/17/06, Bithika Khargharia <bit...@gmail.com> wrote:
> Hi,
>

> 1. I checked the ARCH_PFN_OFFSET variable for my kernel and it was set to 0.
> In that case the kernel should generate PFN serially from 0 to TOTAL_PHYS.
> Am I right?
>

Yes, Right. In kernel/protected_mode the memory is referenced/accessed
through pages not from physical addresses and pages have pfn which is
directly associated with physical address by PAGE_SHIFT bits

> 2. What confuses me is the following. I know that I have 8 GB of memory
> physically installed on my System. 8 GB should translate to a PFN of 2097152
> (8 * 2^30)/4 where 4KB is the PAGE_SIZE in my system. However the kernel
> gives a TOTAL_PHYS value of 2411724 which is almost equal to 9.2 GB of
> memory. If PFNs are generated serially why do I have this problem of higher
> PFNs that 8 GB?
>

The PFNs AFAIK must be 2097152 for 8GB RAM, have you tried to do
read/kmap physical-memory representing by PFN greater-than 2097152 ?
TOTAL_PHYS is the term you are referring for total-physical-memory or
its used in the kernel too ? as I not able to find-out that TOTAL_PHYS
in kernel sources. AFAIR max_mapnr variable (if CONFIG_DISCONTIGMEM
isn't defined) can tell maximum number of pages (of physical-memory)
available in system/kernel which actually calculated during memory
initialization at boot-time.

If still not able to get things correctly then please provide you
kernel version number and config file.

[Bithika Khargharia]

Thanks Fawad,

On 11/18/06, Fawad Lateef <fawad...@gmail.com> wrote:

Your complete system information to know will be good for better
understanding of your ptoblem. Means output of 'uname -a' and
processor/motherboard model.

uname -a gives Linux dhcp-233.ece.arizona.edu 2.6.18 #131 SMP Fri Nov 17 16:38:34 MST 2006 i686 i686 i386 GNU/Linux.I have dual core Intel xeon processor and the motherboard is Intel® Server Board S5000PAL

I can't say exactly what is happening on your side, might be some
in-compatiblity of your motherboard (bios) with Linux Kernel. But the
thing which you can see is: point 8. Total usable pfns =  2070237 =
~8GB memory, so might possible your bios is telling IO-Devices mapped
region as some more memory available (but I am not sure about this).

 
That's a good point. Also with total usable pfns = 2070237 and last usable pfn 2490368

there must be some PFNs in between that are IO-Devices mapped as you mentioned perhaps.

Have you tried to read/write/kmap on pages whose pfns are giving
physical address after 8GB ? Is it working ? and what memory size 'cat
/proc/meminfo' is mentioning ?

No I have not. I will try that and let you know what happens. cat /proc/meminfo actually gives  MemTotal: 8280948 kB !!
 

Its better is you also do CC to kernelnewbies, so that some other
person might give you some better idea too. And please don't do
top-posting, rather do bottom-posting (search google for it) on
mailing-list while replying.

Thanks Fawad or introducing me to the lists'  netiquettes

Bithika

[Fawad Lateef]

On 11/21/06, Bithika Khargharia <bit...@gmail.com> wrote:
> Thanks Fawad,
>
> On 11/18/06, Fawad Lateef <fawad...@gmail.com> wrote:
>
>
> > Your complete system information to know will be good for better
> > understanding of your ptoblem. Means output of 'uname -a' and
> > processor/motherboard model.
> >
>
> uname -a gives Linux dhcp-233.ece.arizona.edu 2.6.18 #131 SMP Fri Nov 17
> 16:38:34 MST 2006 i686 i686 i386 GNU/Linux.I have dual core Intel xeon

> processor and the motherboard is Intel(r) Server Board S5000PAL

>
> > I can't say exactly what is happening on your side, might be some
> > in-compatiblity of your motherboard (bios) with Linux Kernel. But the
> > thing which you can see is: point 8. Total usable pfns = 2070237 =
> > ~8GB memory, so might possible your bios is telling IO-Devices mapped
> > region as some more memory available (but I am not sure about this).
>
>
> That's a good point. Also with total usable pfns = 2070237 and last usable
> pfn 2490368
> there must be some PFNs in between that are IO-Devices mapped as you
> mentioned perhaps.
>

Yes, this might be the case. Try google for your possible bios related issues.

> > Have you tried to read/write/kmap on pages whose pfns are giving
> > physical address after 8GB ? Is it working ? and what memory size 'cat
> > /proc/meminfo' is mentioning ?
>
> No I have not. I will try that and let you know what happens. cat
> /proc/meminfo actually gives MemTotal: 8280948 kB !!
>

This means Linux/Kernel actually knows ~8GB memory so will use only
8GB, now as I said might be your memory mapped IO devices is also
counted in your total_pfns. Someone might give more/correct details

> > Its better is you also do CC to kernelnewbies, so that some other
> > person might give you some better idea too. And please don't do
> > top-posting, rather do bottom-posting (search google for it) on
> > mailing-list while replying.
>
> Thanks Fawad or introducing me to the lists' netiquettes
>

No problem :)