2002-10-02 15:22:36

by Andrea Arcangeli

[permalink] [raw]
Subject: 2.4.20pre8aa2

URL:

http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.20pre8aa2.gz
http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.20pre8aa2/

Changelog between 2.4.20pre8aa1 and 2.4.20pre8aa2:

Only in 2.4.20pre8aa1: 00_extraversion-8
Only in 2.4.20pre8aa2: 00_extraversion-9

Rediffed.

Only in 2.4.20pre8aa2: 00_proc-cmdline-1

Read within the page.

Only in 2.4.20pre8aa2: 00_thinkpad-1

Thinkpad support at http://tpctl.sourceforge.net/ integrated
by Chip Salzenberg.

Only in 2.4.20pre8aa2: 00_usb_get_string-len-1

Don't read behind the end of the string
to avoid timeouts with some hardware. From Andreas Klein.

Only in 2.4.20pre8aa1: 08_qlogicfc-template-aa-3
Only in 2.4.20pre8aa2: 08_qlogicfc-template-aa-4

Allow qlogicfc to compile.

Only in 2.4.20pre8aa1: 20_sched-o1-fixes-1
Only in 2.4.20pre8aa2: 20_sched-o1-fixes-2

Resurrect the starvation logic, fix the idle->prio value, and further
fixes to the run-child-first logic too.

Only in 2.4.20pre8aa2: 84_x86-64-arch-1
Only in 2.4.20pre8aa2: 85_x86-64-includes-1

Synchronize x86-64 with all the additional features
(aio, cpu affinity etc..)

Only in 2.4.20pre8aa1: 9900_aio-8.gz
Only in 2.4.20pre8aa2: 9900_aio-9.gz

Add reiserfs support and fix one bug with timeout
null to getevents. From Chris Mason.

Andrea


2002-10-02 21:40:51

by J.A. Magallon

[permalink] [raw]
Subject: Re: 2.4.20pre8aa2


On 2002.10.02 Andrea Arcangeli wrote:
>URL:
>
> http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.20pre8aa2.gz
> http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.20pre8aa2/
>
>Changelog between 2.4.20pre8aa1 and 2.4.20pre8aa2:
>

I was rediffing the task_cpu patch, when reached a new hunk in -aa:

kernel/sched.c::sched_init(void):

+ current->cpu = smp_processor_id();

As include/asm-i386/smp.h says,

#define smp_processor_id() (current->cpu)

So you have a harmless and useless assignment...
unless you really wanted to do any other thing, or
smp_processor_id() != current->cpu in some arch.

???

TIA

--
J.A. Magallon <[email protected]> \ Software is like sex:
werewolf.able.es \ It's better when it's free
Mandrake Linux release 9.0 (dolphin) for i586
Linux 2.4.20-pre8-jam1 (gcc 3.2 (Mandrake Linux 9.0 3.2-1mdk))

2002-10-02 22:06:10

by Andrea Arcangeli

[permalink] [raw]
Subject: Re: 2.4.20pre8aa2

On Wed, Oct 02, 2002 at 11:45:40PM +0200, J.A. Magallon wrote:
>
> On 2002.10.02 Andrea Arcangeli wrote:
> >URL:
> >
> > http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.20pre8aa2.gz
> > http://www.us.kernel.org/pub/linux/kernel/people/andrea/kernels/v2.4/2.4.20pre8aa2/
> >
> >Changelog between 2.4.20pre8aa1 and 2.4.20pre8aa2:
> >
>
> I was rediffing the task_cpu patch, when reached a new hunk in -aa:
>
> kernel/sched.c::sched_init(void):
>
> + current->cpu = smp_processor_id();
>
> As include/asm-i386/smp.h says,
>
> #define smp_processor_id() (current->cpu)
>
> So you have a harmless and useless assignment...
> unless you really wanted to do any other thing, or
> smp_processor_id() != current->cpu in some arch.

yes, it's a superflous line, it's the equivalent of the 2.5.40 line
here:

rq->idle = current;
set_task_cpu(current, smp_processor_id());
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
wake_up_process(current);

it is a superflous bit in the o1 scheduler changes. If the boot cpu
isn't id 0 it must be initialized by the architectural code in arch/.
Either that or sched_init must use hard_smp_processor_id(). On most
archs the boot cpu is id 0 so it probably doesn't trigger if needed.

Andrea

2002-10-02 22:13:38

by Robert Love

[permalink] [raw]
Subject: Re: 2.4.20pre8aa2

On Wed, 2002-10-02 at 17:45, J.A. Magallon wrote:

> I was rediffing the task_cpu patch, when reached a new hunk in -aa:
>
> kernel/sched.c::sched_init(void):
>
> + current->cpu = smp_processor_id();

Ew we do that in 2.5, too.

I think this is a mistake; we don't need to do this here. But if I am
wrong, we need to get the current processor number elsewhere.

Robert Love