Re: [PATCH 0/2] Per-task wait_queue_t

[Rasmus Villemoes]

Peter Zijlstra <peterz@xxxxxxxxxxxxx> writes:

> On Tue, Jun 10, 2014 at 02:29:17PM +0200, Rasmus Villemoes wrote:
>> This is an attempt to reduce the stack footprint of various functions
>> (those using any of the wait_event_* macros), by removing the need to
>> allocate a wait_queue_t on the stack and instead use a single instance
>> embedded in task_struct. I'm not really sure where the best place to
>> put it is; I just placed it next to other list bookkeeping fields.
>> 
>> For now, there is a little unconditional debugging. This could later
>> be removed or maybe be made dependent on some CONFIG_* variable. The
>> idea of using ->flags is taken from Pavel [1] (I originally stored
>> (void*)1 into ->private).
>> 
>> Compiles, but not actually tested.
>> 
>
> Doesn't look too bad, would be good to be tested and have some numbers
> on the amount of stack saved etc..

Here are some numbers, and the fact that it only has a positive effect
on 28 functions, together with Oleg's concerns, makes me think that it's
probably not worth it.

(defconfig on x86_64, based on 3.15)

file                 function                       old  new  delta
vmlinux              i915_pipe_crc_read              120  136  +16
vmlinux              try_to_free_pages               144  136   -8
vmlinux              mousedev_read                   112   96  -16
vmlinux              sky2_probe                      192  176  -16
vmlinux              gss_cred_init                   136  120  -16
vmlinux              do_coredump                     296  280  -16
vmlinux              md_do_sync                      360  344  -16
vmlinux              save_image_lzo                  200  168  -32
vmlinux              read_events                     184  152  -32
vmlinux              loop_make_request                96   64  -32
vmlinux              i801_access                     104   72  -32
vmlinux              tty_port_block_til_ready        104   72  -32
vmlinux              loop_thread                     152  120  -32
vmlinux              evdev_read                      136  104  -32
vmlinux              __sb_start_write                 96   64  -32
vmlinux              rfkill_fop_read                 104   72  -32
vmlinux              intel_dp_aux_ch                 152  120  -32
vmlinux              i801_transaction                 96   64  -32
vmlinux              blk_mq_queue_enter               96   64  -32
vmlinux              cypress_send_ext_cmd            152  104  -48
vmlinux              rcu_gp_kthread                  120   72  -48
vmlinux              locks_mandatory_area            264  216  -48
vmlinux              hub_thread                      232  184  -48
vmlinux              start_this_handle               120   72  -48
vmlinux              autofs4_wait                    136   88  -48
vmlinux              fcntl_setlk                     136   88  -48
vmlinux              load_module                     264  216  -48
vmlinux              serport_ldisc_read              144   96  -48
vmlinux              sg_read                         160  112  -48

Rasmus
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/