Re: [V9fs-developer] finit_module broken on 9p because kernel_read doesn't work?

[Richard Yao <ryao@gentoo.org>]

[-- Attachment #1: Type: text/plain, Size: 1085 bytes --]

On 02/08/2014 02:54 PM, Andy Lutomirski wrote:
> On Sat, Feb 8, 2014 at 11:27 AM, Richard Yao <ryao@gentoo.org> wrote:
>> On 02/08/2014 02:20 PM, Andy Lutomirski wrote:
>>> Are we looking at the same patch?
>>>
>>> + if (is_vmalloc_or_module_addr(data))
>>> + pages[index++] = vmalloc_to_page(data);
>>>
>>> if (is_vmalloc_or_module_addr(data) && !is_vmalloc_addr(data)), the
>>> vmalloc_to_page(data) sounds unhealthy.
>>>
>>> --Andy
>>>
>>
>> Mainline loads all Linux kernel modules into virtual memory. No
>> architecture is known to me where this is not the case.
>>
> 
> Hmm.  I stand corrected.  vmalloc_to_page is safe on module addresses.
> 
> --Andy
> 

I also stand corrected. After you poked me on this, I sent this patch
with a second patch to export is_vmalloc_or_module_addr() to Linus
Torvalds, who wrote is_vmalloc_or_module_addr(). He provided a very
concise explanation why is_vmalloc_addr() is not only safe, but preferable:

https://lkml.org/lkml/2014/2/8/272

I have resubmitted it with that change. I expect it to be merged soon.


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 901 bytes --]