Re: unable to mmap /dev/kmem

[Nadia Derbey <Nadia.Derbey@bull.net>]

Hugh Dickins wrote:
> On Thu, 18 Jan 2007, Nadia Derbey wrote:
> 
>>Trying to mmap /dev/kmem with an offset I take from /boot/System.map,
>>I get an EIO error on a 2.6.20-rc4.
>>This is something that used to work on older kernels.
>>
>>Had a look at mmap_kmem() in drivers/char/mem.c, and I'm wondering whether
>>pfn is correctly computed there: shouldn't we have something like
>>
>>pfn = PFN_DOWN(virt_to_phys((void *)PAGE_OFFSET)) +
>>      __pa(vma->vm_pgoff << PAGE_SHIFT);
>>
>>instead of
>>
>>pfn = PFN_DOWN(virt_to_phys((void *)PAGE_OFFSET)) + vma->vm_pgoff;
>>
>>Or may be should I substract PAGE_OFFSET from the value I get from System.map
>>before mmapping /dev/kmem?
> 
> 
> Sigh, you're right, 2.6.19 messed that up completely.
> No, you never had to subtract PAGE_OFFSET from that address
> in the past, and you shouldn't have to do so now.
> 
> Please revert the offending patch below, and then maybe Franck
> can come up with a patch which preserves the original behaviour
> on architectures which used to work (e.g. i386), while fixing
> it for those architectures (which are they?) that did not.
> 
> I guess it's reassuring to know that not many are actually
> using mmap of /dev/kmem, so you're the first to notice: thanks.
> 
I find this feature very interesting from a testing perspective. Now, 
since I don't like the idea of being the only one that uses a feature 
(not maintained - may be even to be removed?) may be you could advice me 
on a more broadly used way to get the value of a non exported kernel 
variable from inside a test running in user mode? should I use /dev/mem 
instead? But in that case, I should do the virtual to physical 
conversion myself, right?

Regards,
Nadia