[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20190824035100.7969-1-sst2005@gmail.com>
Date: Sat, 24 Aug 2019 09:20:59 +0530
From: Satendra Singh Thakur <sst2005@...il.com>
To: unlisted-recipients:; (no To-header on input)
Cc: satendrasingh.thakur@....com,
Satendra Singh Thakur <sst2005@...il.com>,
Peter Zijlstra <peterz@...radead.org>,
Ingo Molnar <mingo@...hat.com>, Will Deacon <will@...nel.org>,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH v1] [semaphore] Removed redundant code from semaphore's down family of function
On Thu, 22 Aug 2019 17:51:12 +0200, Peter Zijlstra wrote:
> On Mon, Aug 12, 2019 at 07:18:59PM +0530, Satendra Singh Thakur wrote:
> > -The semaphore code has four funcs
> > down,
> > down_interruptible,
> > down_killable,
> > down_timeout
> > -These four funcs have almost similar code except that
> > they all call lower level function __down_xyz.
> > -This lower level func in-turn call inline func
> > __down_common with appropriate arguments.
> > -This patch creates a common macro for above family of funcs
> > so that duplicate code is eliminated.
> > -Also, __down_common has been made noinline so that code is
> > functionally similar to previous one
> > -For example, earlier down_killable would call __down_killable
> > , which in-turn would call inline func __down_common
> > Now, down_killable calls noinline __down_common directly
> > through a macro
> > -The funcs __down_interruptible, __down_killable etc have been
> > removed as they were just wrapper to __down_common
>
> The above is unreadable and seems to lack a reason for this change.
Hi Mr Peter,
Thanks for the comments.
I will try to explain it further:
The semaphore has four functions named down*.
The call flow of the functions is
down* ----> __down* ----> inline __down_common
The code of down* and __down* is redundant/duplicate except that
the __down_common is called with different arguments from __down*
functions.
This patch defines a macro down_common which contain this common
code of all down* functions.
new call flow is
down* ----> noinline __down_common (through a macro down_common).
> AFAICT from the actual patch, you're destroying the explicit
> instantiation of the __down*() functions
> through constant propagation into __down_common().
Intead of instantiation of __down* functions, we are instaintiating
__down_common, is it a problem ?
Thanks
Satendra
Powered by blists - more mailing lists