[PATCH] x86: mm: Remove unnecessary assignment for max_pfn_mapped:q

[PATCH] x86: mm: Remove unnecessary assignment for max_pfn_mapped:q

[Zhang Yanfei]

init_memory_mapping will set max_pfn_mapped:
int_memory_mapping
  --> add_pfn_range_mapped
    --> max_pfn_mapped = max(max_pfn_mapped, end_pfn)

In init_mem_mapping, before we set max_pfn_mapped to 0, we
have already called init_memory_mapping to setup pagetable
for [0, ISA_END_ADDRESS], and that sets max_pfn_mapped. So
the assignment to 0 is not necessary, remove it.

Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Yinghai Lu <yinghai@kernel.org>
---
 arch/x86/mm/init.c |    1 -
 1 files changed, 0 insertions(+), 1 deletions(-)

diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
index fdc5dca..404c889 100644
--- a/arch/x86/mm/init.c
+++ b/arch/x86/mm/init.c
@@ -416,7 +416,6 @@ void __init init_mem_mapping(void)
 
 	/* step_size need to be small so pgt_buf from BRK could cover it */
 	step_size = PMD_SIZE;
-	max_pfn_mapped = 0; /* will get exact value next */
 	min_pfn_mapped = real_end >> PAGE_SHIFT;
 	last_start = start = real_end;
 	while (last_start > ISA_END_ADDRESS) {
-- 
1.7.1

Re: [PATCH] x86: mm: Remove unnecessary assignment for max_pfn_mapped:q

[Yinghai Lu]

On Fri, May 10, 2013 at 2:01 AM, Zhang Yanfei
<zhangyanfei@cn.fujitsu.com> wrote:
> init_memory_mapping will set max_pfn_mapped:
> int_memory_mapping
>   --> add_pfn_range_mapped
>     --> max_pfn_mapped = max(max_pfn_mapped, end_pfn)
>
> In init_mem_mapping, before we set max_pfn_mapped to 0, we
> have already called init_memory_mapping to setup pagetable
> for [0, ISA_END_ADDRESS], and that sets max_pfn_mapped. So
> the assignment to 0 is not necessary, remove it.

NAK.

for 32bit or Xen, max_pfn_mapped is set way before in head_32.S and
xen-enlighen.

Yinghai

Re: [PATCH] x86: mm: Remove unnecessary assignment for max_pfn_mapped:q

[Zhang Yanfei]

于 2013年05月10日 17:27, Yinghai Lu 写道:
> On Fri, May 10, 2013 at 2:01 AM, Zhang Yanfei
> <zhangyanfei@cn.fujitsu.com> wrote:
>> init_memory_mapping will set max_pfn_mapped:
>> int_memory_mapping
>>   --> add_pfn_range_mapped
>>     --> max_pfn_mapped = max(max_pfn_mapped, end_pfn)
>>
>> In init_mem_mapping, before we set max_pfn_mapped to 0, we
>> have already called init_memory_mapping to setup pagetable
>> for [0, ISA_END_ADDRESS], and that sets max_pfn_mapped. So
>> the assignment to 0 is not necessary, remove it.
> 
> NAK.
> 
> for 32bit or Xen, max_pfn_mapped is set way before in head_32.S and
> xen-enlighen.

Hi Yinghai

I might be wrong, but just from the code in init_mem_mapping only:

 410         /* the ISA range is always mapped regardless of memory holes */
 411         init_memory_mapping(0, ISA_END_ADDRESS);
 412 
 413         /* xen has big range in reserved near end of ram, skip it at first.*/
 414         addr = memblock_find_in_range(ISA_END_ADDRESS, end, PMD_SIZE, PMD_SIZE);
 415         real_end = addr + PMD_SIZE;
 416 
 417         /* step_size need to be small so pgt_buf from BRK could cover it */
 418         step_size = PMD_SIZE;
 419         max_pfn_mapped = 0; /* will get exact value next */

Line 411 set max_pfn_mapped, and then line 419 set it to zero again, so
why keep the later assignment?

Thanks
Zhang

Re: [PATCH] x86: mm: Remove unnecessary assignment for max_pfn_mapped:q

[Yinghai Lu]

On Fri, May 10, 2013 at 3:28 AM, Zhang Yanfei
<zhangyanfei@cn.fujitsu.com> wrote:
> 于 2013年05月10日 17:27, Yinghai Lu 写道:
>> On Fri, May 10, 2013 at 2:01 AM, Zhang Yanfei
>> <zhangyanfei@cn.fujitsu.com> wrote:
>>> init_memory_mapping will set max_pfn_mapped:
>>> int_memory_mapping
>>>   --> add_pfn_range_mapped
>>>     --> max_pfn_mapped = max(max_pfn_mapped, end_pfn)
>>>
>>> In init_mem_mapping, before we set max_pfn_mapped to 0, we
>>> have already called init_memory_mapping to setup pagetable
>>> for [0, ISA_END_ADDRESS], and that sets max_pfn_mapped. So
>>> the assignment to 0 is not necessary, remove it.
>>
>> NAK.
>>
>> for 32bit or Xen, max_pfn_mapped is set way before in head_32.S and
>> xen-enlighen.
>
> Hi Yinghai
>
> I might be wrong, but just from the code in init_mem_mapping only:
>
>  410         /* the ISA range is always mapped regardless of memory holes */
>  411         init_memory_mapping(0, ISA_END_ADDRESS);
>  412
>  413         /* xen has big range in reserved near end of ram, skip it at first.*/
>  414         addr = memblock_find_in_range(ISA_END_ADDRESS, end, PMD_SIZE, PMD_SIZE);
>  415         real_end = addr + PMD_SIZE;
>  416
>  417         /* step_size need to be small so pgt_buf from BRK could cover it */
>  418         step_size = PMD_SIZE;
>  419         max_pfn_mapped = 0; /* will get exact value next */
>
> Line 411 set max_pfn_mapped, and then line 419 set it to zero again, so
> why keep the later assignment?

the comment says: /* will get exact value next */

For x86 32bit path, and xen set bigger max_pfn_mapped before.

Yinghai

Re: [PATCH] x86: mm: Remove unnecessary assignment for max_pfn_mapped:q

[Ingo Molnar]

* Yinghai Lu <yinghai@kernel.org> wrote:

> On Fri, May 10, 2013 at 3:28 AM, Zhang Yanfei
> <zhangyanfei@cn.fujitsu.com> wrote:
> > ?? 2013??05??10?? 17:27, Yinghai Lu ????:
> >> On Fri, May 10, 2013 at 2:01 AM, Zhang Yanfei
> >> <zhangyanfei@cn.fujitsu.com> wrote:
> >>> init_memory_mapping will set max_pfn_mapped:
> >>> int_memory_mapping
> >>>   --> add_pfn_range_mapped
> >>>     --> max_pfn_mapped = max(max_pfn_mapped, end_pfn)
> >>>
> >>> In init_mem_mapping, before we set max_pfn_mapped to 0, we
> >>> have already called init_memory_mapping to setup pagetable
> >>> for [0, ISA_END_ADDRESS], and that sets max_pfn_mapped. So
> >>> the assignment to 0 is not necessary, remove it.
> >>
> >> NAK.
> >>
> >> for 32bit or Xen, max_pfn_mapped is set way before in head_32.S and
> >> xen-enlighen.
> >
> > Hi Yinghai
> >
> > I might be wrong, but just from the code in init_mem_mapping only:
> >
> >  410         /* the ISA range is always mapped regardless of memory holes */
> >  411         init_memory_mapping(0, ISA_END_ADDRESS);
> >  412
> >  413         /* xen has big range in reserved near end of ram, skip it at first.*/
> >  414         addr = memblock_find_in_range(ISA_END_ADDRESS, end, PMD_SIZE, PMD_SIZE);
> >  415         real_end = addr + PMD_SIZE;
> >  416
> >  417         /* step_size need to be small so pgt_buf from BRK could cover it */
> >  418         step_size = PMD_SIZE;
> >  419         max_pfn_mapped = 0; /* will get exact value next */
> >
> > Line 411 set max_pfn_mapped, and then line 419 set it to zero again, so
> > why keep the later assignment?
> 
> the comment says: /* will get exact value next */

And what does that comment mean??

next where? Which call? How? What's the logic? Where is it described 
accurately so that people who haven't seen the code (for a while) can 
understand it?

> For x86 32bit path, and xen set bigger max_pfn_mapped before.

and what does this mean? Perhaps:

 "32-bit x86 and Xen set a bigger value for max_pfn_mapped before this 
  call."

?

And where and why does it matter that we set a bigger max_pfn_mapped 
before?

The dependencies in this code are incestous.

Thanks,

	Ingo

Re: [PATCH] x86: mm: Remove unnecessary assignment for max_pfn_mapped

[Zhang Yanfei]

于 2013年05月10日 22:57, Yinghai Lu 写道:
> On Fri, May 10, 2013 at 3:28 AM, Zhang Yanfei
> <zhangyanfei@cn.fujitsu.com> wrote:
>> 于 2013年05月10日 17:27, Yinghai Lu 写道:
>>> On Fri, May 10, 2013 at 2:01 AM, Zhang Yanfei
>>> <zhangyanfei@cn.fujitsu.com> wrote:
>>>> init_memory_mapping will set max_pfn_mapped:
>>>> int_memory_mapping
>>>>   --> add_pfn_range_mapped
>>>>     --> max_pfn_mapped = max(max_pfn_mapped, end_pfn)
>>>>
>>>> In init_mem_mapping, before we set max_pfn_mapped to 0, we
>>>> have already called init_memory_mapping to setup pagetable
>>>> for [0, ISA_END_ADDRESS], and that sets max_pfn_mapped. So
>>>> the assignment to 0 is not necessary, remove it.
>>>
>>> NAK.
>>>
>>> for 32bit or Xen, max_pfn_mapped is set way before in head_32.S and
>>> xen-enlighen.
>>
>> Hi Yinghai
>>
>> I might be wrong, but just from the code in init_mem_mapping only:
>>
>>  410         /* the ISA range is always mapped regardless of memory holes */
>>  411         init_memory_mapping(0, ISA_END_ADDRESS);
>>  412
>>  413         /* xen has big range in reserved near end of ram, skip it at first.*/
>>  414         addr = memblock_find_in_range(ISA_END_ADDRESS, end, PMD_SIZE, PMD_SIZE);
>>  415         real_end = addr + PMD_SIZE;
>>  416
>>  417         /* step_size need to be small so pgt_buf from BRK could cover it */
>>  418         step_size = PMD_SIZE;
>>  419         max_pfn_mapped = 0; /* will get exact value next */
>>
>> Line 411 set max_pfn_mapped, and then line 419 set it to zero again, so
>> why keep the later assignment?
> 
> the comment says: /* will get exact value next */
> 
> For x86 32bit path, and xen set bigger max_pfn_mapped before.
> 

Hello Yinghai,

I kindly understand what you mean now. But I think we can put the reset of
max_pfn_mapped at the beginning of init_mem_mapping. That looks more logical.

See patch below, please.

Thanks
Zhang

------
>From dbd213301d800299d256c5da65a1d598902ed826 Mon Sep 17 00:00:00 2001
From: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
Date: Wed, 15 May 2013 11:48:19 +0800
Subject: [PATCH] x86, mm: Reset max_pfn_mapped before we create direct mappings

We use init_mem_mapping to create direct mappings from scratch. But
in 32bit or xen, we may have already set max_pfn_mapped before in
head_32.S and or xen-enlighten. So put the reset at beginning not
in the middle of init_mem_mapping seems more logical.

Also add comments to explain the reset.

Signed-off-by: Zhang Yanfei <zhangyanfei@cn.fujitsu.com>
---
 arch/x86/mm/init.c |    9 ++++++++-
 1 files changed, 8 insertions(+), 1 deletions(-)

diff --git a/arch/x86/mm/init.c b/arch/x86/mm/init.c
index fdc5dca..f2e5d28 100644
--- a/arch/x86/mm/init.c
+++ b/arch/x86/mm/init.c
@@ -407,6 +407,14 @@ void __init init_mem_mapping(void)
 	end = max_low_pfn << PAGE_SHIFT;
 #endif
 
+	/*
+	 * In 32bit or xen, max_pfn_mapped has been set way before in
+	 * head_32.S or xen-enlighten. So here we reset it since we will
+	 * create direct mappings from scratch. And it will get its finally
+	 * exact value after we finish the direct mapping in this function.
+	 */
+	max_pfn_mapped = 0;
+
 	/* the ISA range is always mapped regardless of memory holes */
 	init_memory_mapping(0, ISA_END_ADDRESS);
 
@@ -416,7 +424,6 @@ void __init init_mem_mapping(void)
 
 	/* step_size need to be small so pgt_buf from BRK could cover it */
 	step_size = PMD_SIZE;
-	max_pfn_mapped = 0; /* will get exact value next */
 	min_pfn_mapped = real_end >> PAGE_SHIFT;
 	last_start = start = real_end;
 	while (last_start > ISA_END_ADDRESS) {
-- 
1.7.1