Date: Mon, 20 Jul 2015 10:56:00 +0200
From: Martin Schwidefsky <schwidefsky@de.ibm.com>
To: Ingo Molnar <mingo@kernel.org>
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>,
        Linus Torvalds <torvalds@linux-foundation.org>,
        linux-kernel@vger.kernel.org, Thomas Gleixner <tglx@linutronix.de>,
        "H. Peter Anvin" <hpa@zytor.com>, Andy Lutomirski <luto@kernel.org>,
        Dave Hansen <dave@sr71.net>, Andrew Morton <akpm@linux-foundation.org>,
        Oleg Nesterov <oleg@redhat.com>
Subject: Re: [PATCH] sched, s390: Fix the fallout of increasing the offset
 of 'thread_struct' within 'task_struct'
Message-ID: <20150720105600.79b10fe4@mschwide>
In-Reply-To: <20150720083847.GC12468@gmail.com>
References: <20150718031810.GA19818@gmail.com>
	<20150720072037.GA3607@osiris>
	<20150720080032.GA12468@gmail.com>
	<20150720101219.79fb06bd@mschwide>
	<20150720083847.GC12468@gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 3598
Lines: 88

On Mon, 20 Jul 2015 10:38:47 +0200
Ingo Molnar <mingo@kernel.org> wrote:

> 
> * Martin Schwidefsky <schwidefsky@de.ibm.com> wrote:
> 
> > diff --git a/arch/s390/kernel/entry.S b/arch/s390/kernel/entry.S
> > index 3238893..84062e7 100644
> > --- a/arch/s390/kernel/entry.S
> > +++ b/arch/s390/kernel/entry.S
> > @@ -178,17 +178,21 @@ _PIF_WORK	= (_PIF_PER_TRAP)
> >   */
> >  ENTRY(__switch_to)
> >  	stmg	%r6,%r15,__SF_GPRS(%r15)	# store gprs of prev task
> > -	stg	%r15,__THREAD_ksp(%r2)		# store kernel stack of prev
> > -	lg	%r4,__THREAD_info(%r2)		# get thread_info of prev
> > -	lg	%r5,__THREAD_info(%r3)		# get thread_info of next
> > +	lgr	%r1,%r2
> > +	aghi	%r1,__TASK_thread		# thread_struct of prev task
> > +	lg	%r4,__TASK_thread_info(%r2)	# get thread_info of prev
> > +	lg	%r5,__TASK_thread_info(%r3)	# get thread_info of next
> > +	stg	%r15,__THREAD_ksp(%r1)		# store kernel stack of prev
> > +	lgr	%r1,%r3
> > +	aghi	%r1,__TASK_thread		# thread_struct of next task
> >  	lgr	%r15,%r5
> >  	aghi	%r15,STACK_INIT			# end of kernel stack of next
> >  	stg	%r3,__LC_CURRENT		# store task struct of next
> >  	stg	%r5,__LC_THREAD_INFO		# store thread info of next
> >  	stg	%r15,__LC_KERNEL_STACK		# store end of kernel stack
> > +	lg	%r15,__THREAD_ksp(%r1)		# load kernel stack of next
> >  	lctl	%c4,%c4,__TASK_pid(%r3)		# load pid to control reg. 4
> >  	mvc	__LC_CURRENT_PID+4(4,%r0),__TASK_pid(%r3) # store pid of next
> > -	lg	%r15,__THREAD_ksp(%r3)		# load kernel stack of next
> >  	lmg	%r6,%r15,__SF_GPRS(%r15)	# load gprs of next task
> >  	br	%r14
> 
> Btw., I think we'd be slightly better off with the variant I sent: that way the 
> offset arithmetics are done in C code and are ready in the relevant registers by 
> the time they are used in __switch_to().

If we save any instruction then not many, we are trading additional parameter
passing vs. four additional instructions in __switch_to. With this variant
the fallout is kept at a minimum which I would prefer at this point.
 
> >  
> > @@ -417,6 +421,7 @@ ENTRY(pgm_check_handler)
> >  	LAST_BREAK %r14
> >  	lg	%r15,__LC_KERNEL_STACK
> >  	lg	%r14,__TI_task(%r12)
> > +	aghi	%r14,__TASK_thread	# pointer to thread_struct
> >  	lghi	%r13,__LC_PGM_TDB
> >  	tm	__LC_PGM_ILC+2,0x02	# check for transaction abort
> >  	jz	2f
> 
> Don't we also need the chunk I have:
> 
> @@ -448,6 +449,7 @@ ENTRY(pgm_check_handler)
>  	nill	%r10,0x007f
>  	sll	%r10,2
>  	je	.Lsysc_return
> +	ahi     %r14,-__TASK_thread_struct # r14 now points to 'current'
>  	lgf	%r1,0(%r10,%r1)		# load address of handler routine
>  	lgr	%r2,%r11		# pass pointer to pt_regs
>  	basr	%r14,%r1		# branch to interrupt-handler
> 
> Because the 'basr' line relies on 'r14' having task_struct, I think?
> 
> But I don't really know what I'm talking about here ...

The basr instruction is one of the function calling instructions and
%r14 is used for the return address. Between function calls the register
is basically free. So far it is used for the task_struct pointer
(the code following the "lg %r14,__TI_task(%r12)"). What I noticed is
that the pointer is only used to access the embedded thread_struct.
To get from the task_struct to a thread_struct a single aghi is
needed. The rest is moving stuff around in asm-offsets.c

-- 
blue skies,
   Martin.

"Reality continues to ruin my life." - Calvin.

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