Hi all, first post here so I'll try to be as clear as I can

Essentially I need to read an int that is stored in memory, and I have it's physical address. This address was obtained from a running kernel module, and I can confirm 100% that it is a physical address, not a virtual one.

I need to either create a kernel module, or a userspace app that will allow me to read the int from the physical address. Read only access is required, I don't need/want to write anything to memory at this time.

From what I've been reading, I think it might be possible to do what I want using the mmap function, which should (I might be wrong here) map a physical memory address into a process space so that it can be read from.

If anyone could provide some information on accessing physical memory, or mapping a physical address to a virtual address I would greatly appreciate it.

Thanks in advance!

Nope.

Kernel physical addresses are useless outside the kernel. mmap will not map kernel memory.

This was why the sysfs virtual filesystem was created. Kernel modules that need to import/export data in odd units define entries that allow the sysfs filesystem to have the appropriate directory tree, and can import/export whatever they need.

The problem with physical addresses is that they can change... And updating a physical address does not guarantee that what is there is always there. It depends on how the module is loaded, what other modules may be present. So "peeking and poking" the kernel is not supported.

as a quickie followup, what are you trying to do?

Thanks jpollard.

Basically, I've got a loaded kernel module that as part of its init routine creates a 4KB struct in memory.

What I need to do is share this struct so that at a later date, a user space process can have read only access to it.

I've done a bit more reading and I think I need to either use mmap, or remap_pfn_range/remap_pfn_pages.

If you could provide any further information, that would be great.

Thanks

As I said, mmap cannot map kernel memory. NONE of the user mode mappings can do so.

What you have to do is add a sysfs entry for your module. The user process can then read the entry in the normal way.

Anything else can crash your system.

There used to be a device, /dev/kmem, for accessing kernel virtual memory; but it's been eliminated in recent kernels as a security hole. Privileged processes can access physical addresses from user space using /dev/mem. However using this to look at a kernel struct is like using a cannon to turn off your lights. Use a sysfs entry as jpollard suggests.

And kmem is kernel memory... not necessarily the same as physical memory. It does get remapped as pages move around.