[lkml]   [2006]   [Nov]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
Subject[PATCH 0/4] WorkStruct: Shrink work_struct by two thirds

The workqueue struct is huge, and this limits it's usefulness. On a 64-bit
architecture it's nearly 100 bytes in size, of which the timer_list is half.
These patches shrink work_struct by 8 of the 12 words it ordinarily consumes.
This is done by:

(1) Splitting the timer out so that delayable work items are defined by a
separate structure which incorporates a basic work_struct and a timer.

(2) Folding the pending bit and wq_data data together

(3) Removing the private data. This can almost always be derived from the
address of the work_struct using container_of() and the selection of the
work function. For the cases where the container of the work_struct may
go away the moment the pending bit is cleared, it is made possible to
defer the release of the structure by deferring the clearing of the
pending bit.

These patches reduce the size of the work_struct thusly:

#WORDS 32-bit arch 64-bit arch
=============== =============== ===============
As is 12 48 bytes 96 bytes
Non-delayable 4 16 bytes 32 bytes
Delayable 10 40 bytes 80 bytes

I've looked through most of the usages of work_structs, and I think that
probably fewer than half the work_structs used actually require delayability,
and I'm not sure that it's absolutely necessary in all cases.

With these patches applied, there are four classes of work item where
previously there was one. These are made up of a combination of the following

(*) Delayable vs Non-delayable.

Delayable work items have their execution deferred for at least a certain
amount of time; non-delayable items are executed as soon as possible.

(*) Auto-release vs Non-auto-release

Ordinarily, the work queue executor would release the work_struct for
further scheduling or deallocation by clearing the pending bit prior to
jumping to the work function. This means that, unless the driver makes
some guarantee itself that the work_struct won't go away, the work
function may not access anything else in the work_struct or its container
lest they be deallocated.. This is a problem if the auxiliary data is
taken away (as done by the last patch).

However, if the pending bit is *not* cleared before jumping to the work
function, then the work function *may* access the work_struct and its
container with no problems. But then the work function must itself
release the work_struct by calling work_release().

In most cases, automatic release is fine, so this is the default. Special
initiators exist for the non-auto-release case.

Note that this is a partial conversion. If these patches are generally
acceptable, then the rest of the kernel will also need modification. I've
tested these patches on my x86_64 testbox only thus far.

Furthermore, the timer_list struct could possibly be shrunk by 1 word if it
also lost its data member.

To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to
More majordomo info at
Please read the FAQ at

 \ /
  Last update: 2006-11-20 15:33    [W:0.141 / U:5.864 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site