buggy GFP_KERNEL allocators

[Rik van Riel]

Hi Andrea, Alan,

there have been a number of memory-management related hangs
in 2.2.14 and 2.2.15pre*. Because of the differences in the
error messages (if at all) spit out by both kernels I have
the idea that there must be one or more broken drivers or
subsystems out there which hang if their GFP_KERNEL allocation
fails.

In 2.2.14 these allocations were allowed to eat into the space
reserved for GFP_ATOMIC and PF_MEMALLOC allocations, leading
to the situation that atomic allocations failed and ``funny''
things happened.

Since my VM changes in 2.2.15pre4 the atomic allocation errors
no longer happen. GFP_KERNEL allocations are no longer allowed
to eat into the last free pages, _however_ now these allocations
are failing earlier than they used to, leading to other kinds
of kernel hangs.

What we need to do now is identify the places where it can go
wrong and fix them. Andrea, you seemed to know some of these
places judging what you told me in our MM discussion. Do you 
happen to know what we should fix for 2.2.15?

regards,

Rik
--
The Internet is not a network of computers. It is a network
of people. That is its real strength.


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo en vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/