2009-07-14 08:57:36

by Zhao Lei

Subject: [PATCH] Remove initialization of page_table in one_page_table_init()

Only judge if (!page_table) when CONFIG_DEBUG_PAGEALLOC or CONFIG_KMEMCHECK
are defined, then we don't need to set init value for page_table.

Signed-off-by: Zhao Lei <[email protected]>
---
arch/x86/mm/init_32.c | 4 ++--
1 files changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
index 3cd7711..55da09b 100644
--- a/arch/x86/mm/init_32.c
+++ b/arch/x86/mm/init_32.c
@@ -108,13 +108,13 @@ static pmd_t * __init one_md_table_init(pgd_t *pgd)
static pte_t * __init one_page_table_init(pmd_t *pmd)
{
if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
- pte_t *page_table = NULL;
+ pte_t *page_table;

if (after_bootmem) {
#if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
-#endif
if (!page_table)
+#endif
page_table =
(pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
} else
--
1.5.5.3

2009-07-14 10:18:49

by David Rientjes

Subject: Re: [PATCH] Remove initialization of page_table in one_page_table_init()

On Tue, 14 Jul 2009, Zhaolei wrote:

> diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
> index 3cd7711..55da09b 100644
> --- a/arch/x86/mm/init_32.c
> +++ b/arch/x86/mm/init_32.c
> @@ -108,13 +108,13 @@ static pmd_t * __init one_md_table_init(pgd_t *pgd)
> static pte_t * __init one_page_table_init(pmd_t *pmd)
> {
> if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
> - pte_t *page_table = NULL;
> + pte_t *page_table;
>
> if (after_bootmem) {
> #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
> page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
> -#endif
> if (!page_table)
> +#endif
> page_table =
> (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
> } else

Nack, gcc will already optimize the branch out for configs without
CONFIG_DEBUG_PAGEALLOC and CONFIG_KMEMCHECK. The only thing this change
does is make the code less readable.

2009-07-18 14:13:06

by Ingo Molnar

Subject: Re: [PATCH] Remove initialization of page_table in one_page_table_init()

* Zhaolei <[email protected]> wrote:

> Only judge if (!page_table) when CONFIG_DEBUG_PAGEALLOC or CONFIG_KMEMCHECK
> are defined, then we don't need to set init value for page_table.
>
> Signed-off-by: Zhao Lei <[email protected]>
> ---
> arch/x86/mm/init_32.c | 4 ++--
> 1 files changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
> index 3cd7711..55da09b 100644
> --- a/arch/x86/mm/init_32.c
> +++ b/arch/x86/mm/init_32.c
> @@ -108,13 +108,13 @@ static pmd_t * __init one_md_table_init(pgd_t *pgd)
> static pte_t * __init one_page_table_init(pmd_t *pmd)
> {
> if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
> - pte_t *page_table = NULL;
> + pte_t *page_table;
>
> if (after_bootmem) {
> #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
> page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
> -#endif
> if (!page_table)
> +#endif
> page_table =
> (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);

I think the original code was cleaner - putting an #ifdef straight
into the middle of a control block is quite fragile.

Ingo

2009-07-20 03:07:00

by Zhao Lei

Subject: Re: Re: [PATCH] Remove initialization of page_table in one_page_table_init()

Ingo Molnar wrote:
> * Zhaolei <[email protected]> wrote:
>
>> Only judge if (!page_table) when CONFIG_DEBUG_PAGEALLOC or CONFIG_KMEMCHECK
>> are defined, then we don't need to set init value for page_table.
>>
>> Signed-off-by: Zhao Lei <[email protected]>
>> ---
>> arch/x86/mm/init_32.c | 4 ++--
>> 1 files changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
>> index 3cd7711..55da09b 100644
>> --- a/arch/x86/mm/init_32.c
>> +++ b/arch/x86/mm/init_32.c
>> @@ -108,13 +108,13 @@ static pmd_t * __init one_md_table_init(pgd_t *pgd)
>> static pte_t * __init one_page_table_init(pmd_t *pmd)
>> {
>> if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
>> - pte_t *page_table = NULL;
>> + pte_t *page_table;
>>
>> if (after_bootmem) {
>> #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
>> page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
>> -#endif
>> if (!page_table)
>> +#endif
>> page_table =
>> (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
>
> I think the original code was cleaner - putting an #ifdef straight
> into the middle of a control block is quite fragile.
>
> Ingo
>

Hello, Ingo

David Rientjes said that:
Nack, gcc will already optimize the branch out for configs without
CONFIG_DEBUG_PAGEALLOC and CONFIG_KMEMCHECK. The only thing this change
does is make the code less readable.

I tested it with gcc -O2, and proved that David is correct.
Initialization of page_table and additional "if (!page_table)" was removed by optimization.
Please ignore this patch.

Thanks
Zhaolei

2009-07-20 11:51:03

by Ingo Molnar

Subject: Re: Re: [PATCH] Remove initialization of page_table in one_page_table_init()

* Zhaolei <[email protected]> wrote:

> Ingo Molnar wrote:
> > * Zhaolei <[email protected]> wrote:
> >
> >> Only judge if (!page_table) when CONFIG_DEBUG_PAGEALLOC or CONFIG_KMEMCHECK
> >> are defined, then we don't need to set init value for page_table.
> >>
> >> Signed-off-by: Zhao Lei <[email protected]>
> >> ---
> >> arch/x86/mm/init_32.c | 4 ++--
> >> 1 files changed, 2 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/arch/x86/mm/init_32.c b/arch/x86/mm/init_32.c
> >> index 3cd7711..55da09b 100644
> >> --- a/arch/x86/mm/init_32.c
> >> +++ b/arch/x86/mm/init_32.c
> >> @@ -108,13 +108,13 @@ static pmd_t * __init one_md_table_init(pgd_t *pgd)
> >> static pte_t * __init one_page_table_init(pmd_t *pmd)
> >> {
> >> if (!(pmd_val(*pmd) & _PAGE_PRESENT)) {
> >> - pte_t *page_table = NULL;
> >> + pte_t *page_table;
> >>
> >> if (after_bootmem) {
> >> #if defined(CONFIG_DEBUG_PAGEALLOC) || defined(CONFIG_KMEMCHECK)
> >> page_table = (pte_t *) alloc_bootmem_pages(PAGE_SIZE);
> >> -#endif
> >> if (!page_table)
> >> +#endif
> >> page_table =
> >> (pte_t *)alloc_bootmem_low_pages(PAGE_SIZE);
> >
> > I think the original code was cleaner - putting an #ifdef straight
> > into the middle of a control block is quite fragile.
> >
> > Ingo
> >
>
> Hello, Ingo
>
> David Rientjes said that:
> Nack, gcc will already optimize the branch out for configs without
> CONFIG_DEBUG_PAGEALLOC and CONFIG_KMEMCHECK. The only thing this change
> does is make the code less readable.
>
> I tested it with gcc -O2, and proved that David is correct.
> Initialization of page_table and additional "if (!page_table)" was
> removed by optimization. Please ignore this patch.

Yeah - and note that even if GCC _didnt_ optimize it out, the patch
was not acceptable in this form.

Ingo