public inbox for linuxppc-dev@ozlabs.org 
 help / color / mirror / Atom feed
From: Segher Boessenkool <segher@kernel•crashing.org>
To: Benjamin Herrenschmidt <benh@kernel•crashing.org>
Cc: Samuel Mendoza-Jonas <sam.mj@au1•ibm.com>, linuxppc-dev@ozlabs•org
Subject: Re: [PATCH V3 2/2] powerpc/kexec: Reset HILE before entering target kernel
Date: Fri, 17 Jul 2015 04:59:57 -0500	[thread overview]
Message-ID: <20150717095957.GB15502@gate.crashing.org> (raw)
In-Reply-To: <1437098018.28088.62.camel@kernel.crashing.org>

On Fri, Jul 17, 2015 at 11:53:38AM +1000, Benjamin Herrenschmidt wrote:
> On Fri, 2015-07-10 at 15:19 +1000, Samuel Mendoza-Jonas wrote:
> > +#if defined(CONFIG_PPC_BOOK3S_64) && defined(CONFIG_PPC_POWERNV)
> > +       li      r3,(FW_FEATURE_OPAL >> 16)
> > +       rldicr  r3,r3,16,63
> > +       and.    r3,r3,r26
> > +       cmpwi   r3,0
> > +       beq     99f
> 
> If FW_FEATRURE_OPAL is 0x80000000 then the li will sign extend.
> 
> The rldicr has a mask of all F's so it will keep all the bits you
> don't care about.
> 
> So together, you'll get compares happening on bits above the 16 you care
> about that might change the result of your comparison incorrectly.
> 
> Since FW_FEATURE_* bits aren't ABI, they can change, so we don't want
> to impose a constraint on them.
> 
> Thus I would recommend using an rdlicl r3,r3,16,48 (aka srdi r3,r3,48)
> instead which is going to clear all bits above 0xffff.

Or the other way around: instead of loading and masking, load zero and
OR the bits in:

+       li      r3,0
+       ori     r3,(FW_FEATURE_OPAL >> 16)
+       rldicr  r3,r3,16,63
+       and.    r3,r3,r26
+       cmpwi   r3,0
+       beq     99f

which of course is better written as

+       li      r3,0
+       oris    r3,(FW_FEATURE_OPAL >> 16)
+       and.    r3,r3,r26
+       cmpwi   r3,0
+       beq     99f

which is

+       andis.  r3,r26,(FW_FEATURE_OPAL >> 16)
+       cmpwi   r3,0
+       beq     99f

which is

+       andis.  r3,r26,(FW_FEATURE_OPAL >> 16)
+       beq     99f

> Now, that being said, FW_FEATURE_* can be 64-bit

All of the code above only works for single-bit constants inside
0xffff0000 though (or 0x7fff0000 for the original).

> and this isn't perf
> critical so why not just load the full 64-bit constant into r3 and
> be done with it ? There's a macro to do that:
> 
> 	LOAD_REG_IMMEDIATE(r3,FW_FEATURE_OPAL)

And as you say later, use C.  Yeah.


Segher

      parent reply	other threads:[~2015-07-17 10:00 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2015-07-10  5:19 [PATCH V3 0/2] powerpc/kexec: Reset endianess before kexec Samuel Mendoza-Jonas
2015-07-10  5:19 ` [PATCH V3 1/2] powerpc/kexec: Reset secondary cpu " Samuel Mendoza-Jonas
2015-07-10  5:19 ` [PATCH V3 2/2] powerpc/kexec: Reset HILE before entering target kernel Samuel Mendoza-Jonas
2015-07-17  1:53   ` Benjamin Herrenschmidt
2015-07-17  3:34     ` Benjamin Herrenschmidt
2015-07-17  9:59     ` Segher Boessenkool [this message]

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20150717095957.GB15502@gate.crashing.org \
    --to=segher@kernel$(echo .)crashing.org \
    --cc=benh@kernel$(echo .)crashing.org \
    --cc=linuxppc-dev@ozlabs$(echo .)org \
    --cc=sam.mj@au1$(echo .)ibm.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox