Hi
I am getting stack overflows on my Sparc64 station. They happen when I
copy to device-mapper snapshot origin device using small IO size (512
bytes) and simultaneously execute "lvs" command. The kernel is compiled
with most debugging functions enabled. The stack trace is this:
__ide_end_request
__blk_end_request
__end_that_request_first
req_bio_endio
bio_endio
clone_endio
dec_pending
bio_endio
clone_endio
dec_pending
bio_endio
clone_endio
dec_pending
bio_endio
end_bio_bh_io_sync
end_buffer_read_sync
__end_buffer_read_notouch
unlock_buffer
wake_up_bit
__wake_up_bit
__wake_up
__wake_up_common
wake_bit_function
autoremove_wake_function
default_wake_function
try_to_wake_up
task_rq_lock
__spin_lock
lock_acquire
__lock_acquire
*** crash, stack overflow
--- observations:
That loop bio_endio->clone_endio->dec_pending is repeating for each level
of nested devices --- so for any architecture there exists a level at
which it causes trouble. We need something to prevent recursion, maybe the
similar trick that was done with avoing bio request function recursion
(i.e. if bio_endio is called recursively, it just adds the bio to queue
and lets the top level to call endio method).
Wait queue waking looks like being written by a high-level maniac --- it
contains 8 levels of calls (none of them inlined). 7 of these calls (until
try_to_wake_up) do nothing but pass arguments to lower level call. And
each of these calls allocate at least 192 bytes of stack space. All these
7 useless calls consume 1360 bytes of stack (and cause windows traps that
needlessly damage performance). Would you agree to inline most of the
calls to save stack? Or do you see another solution?
Long-term consideration: Is it possible to implement interrupt stacks on
sparc64? Functions on sparc eat stack much more aggressively than on other
architectures (minimum stack size for a function is 192 bytes).
Mikulas
--
