Messages in this thread | | | Date | Thu, 31 Jan 2008 00:39:01 +0200 | From | Pekka Paalanen <> | Subject | Re: [RFC PATCH] x86: mmiotrace - trace memory mapped IO |
| |
On Sun, 27 Jan 2008 19:55:36 +0200 Pekka Paalanen <pq@iki.fi> wrote:
> The following is a copy of the patch description: ... > Mmiotrace works by wrapping the ioremap family of kernel functions and > marking the returned pages as not present. Access to the IO memory > triggers a page fault, which will be handled by mmiotrace's custom page > fault handler. This will single-step the faulted instruction with the > MMIO page marked as present. Access logs are directed to user space via > relay and debug_fs.
Hi,
a test user found a problem with mmiotrace (ported to x86/mm git tree). The symbol 'init_mm' is undefined, but required when building the 32-bit mmiotrace kernel module.
commit 3abf024d2abb79614d8c4cb25a70d5596f77d0ad Author: Thomas Gleixner <tglx@linutronix.de> Date: Wed Jan 30 13:30:28 2008 +0100
x86: nuke a ton of unused exports
This commit removes EXPORT_SYMBOL(init_mm). Apparently mmiotrace requires this symbol, but I do not know why. I do not use init_mm myself, so apparently some function I call is inlined and requires it. Page table or attribute handling stuff maybe?
Is it possible to simply put back this export or should I dig deeper into why it seems to be required? I can provide the simple patch to put it back in my next patch set, if this is the right solution.
I will post the patch set after this issue is solved. In the mean time, for the interested, the current patches are here: http://jumi.lut.fi/~paalanen/scratch/mmio25/
Ps. I tested x86/mm version of mmiotrace on my 64-bit system and it seems to work ok. I had to write a simple test module because I did not have any drivers that a) called ioremap, b) were not broken on x86/mm, and c) I had the hardware to actually use it. So this testing is very light.
-- Pekka Paalanen http://www.iki.fi/pq/
| |