2018-03-16 02:59:41

by Jia He

Subject: [PATCH] Revert "mm/memblock.c: hardcode the end_pfn being -1"

This reverts commit 379b03b7fa05f7db521b7732a52692448a3c34fe.

Commit 864b75f9d6b0 ("mm/page_alloc: fix memmap_init_zone pageblock
alignment") introduced boot hang issues in arm/arm64 machines, so
Ard Biesheuvel reverted in commit 3e04040df6d4. But there is a
preparation patch for commit 864b75f9d6b0. So just revert it for
the sake of caution.

Signed-off-by: Jia He <[email protected]>
---
mm/memblock.c | 10 +++++-----
1 file changed, 5 insertions(+), 5 deletions(-)

diff --git a/mm/memblock.c b/mm/memblock.c
index b6ba6b7..5a9ca2a 100644
--- a/mm/memblock.c
+++ b/mm/memblock.c
@@ -1107,7 +1107,7 @@ unsigned long __init_memblock memblock_next_valid_pfn(unsigned long pfn,
struct memblock_type *type = &memblock.memory;
unsigned int right = type->cnt;
unsigned int mid, left = 0;
- phys_addr_t addr = PFN_PHYS(++pfn);
+ phys_addr_t addr = PFN_PHYS(pfn + 1);

do {
mid = (right + left) / 2;
@@ -1118,15 +1118,15 @@ unsigned long __init_memblock memblock_next_valid_pfn(unsigned long pfn,
type->regions[mid].size))
left = mid + 1;
else {
- /* addr is within the region, so pfn is valid */
- return pfn;
+ /* addr is within the region, so pfn + 1 is valid */
+ return min(pfn + 1, max_pfn);
}
} while (left < right);

if (right == type->cnt)
- return -1UL;
+ return max_pfn;
else
- return PHYS_PFN(type->regions[right].base);
+ return min(PHYS_PFN(type->regions[right].base), max_pfn);
}

/**
--
2.7.4

2018-03-16 09:08:27

by Michal Hocko

Subject: Re: [PATCH] Revert "mm/memblock.c: hardcode the end_pfn being -1"

On Thu 15-03-18 19:56:06, Jia He wrote:
> This reverts commit 379b03b7fa05f7db521b7732a52692448a3c34fe.
>
> Commit 864b75f9d6b0 ("mm/page_alloc: fix memmap_init_zone pageblock
> alignment") introduced boot hang issues in arm/arm64 machines, so
> Ard Biesheuvel reverted in commit 3e04040df6d4. But there is a
> preparation patch for commit 864b75f9d6b0. So just revert it for
> the sake of caution.

Why? Is there anything wrong with this one?

>
> Signed-off-by: Jia He <[email protected]>
> ---
> mm/memblock.c | 10 +++++-----
> 1 file changed, 5 insertions(+), 5 deletions(-)
>
> diff --git a/mm/memblock.c b/mm/memblock.c
> index b6ba6b7..5a9ca2a 100644
> --- a/mm/memblock.c
> +++ b/mm/memblock.c
> @@ -1107,7 +1107,7 @@ unsigned long __init_memblock memblock_next_valid_pfn(unsigned long pfn,
> struct memblock_type *type = &memblock.memory;
> unsigned int right = type->cnt;
> unsigned int mid, left = 0;
> - phys_addr_t addr = PFN_PHYS(++pfn);
> + phys_addr_t addr = PFN_PHYS(pfn + 1);
>
> do {
> mid = (right + left) / 2;
> @@ -1118,15 +1118,15 @@ unsigned long __init_memblock memblock_next_valid_pfn(unsigned long pfn,
> type->regions[mid].size))
> left = mid + 1;
> else {
> - /* addr is within the region, so pfn is valid */
> - return pfn;
> + /* addr is within the region, so pfn + 1 is valid */
> + return min(pfn + 1, max_pfn);
> }
> } while (left < right);
>
> if (right == type->cnt)
> - return -1UL;
> + return max_pfn;
> else
> - return PHYS_PFN(type->regions[right].base);
> + return min(PHYS_PFN(type->regions[right].base), max_pfn);
> }
>
> /**
> --
> 2.7.4
>

--
Michal Hocko
SUSE Labs

2018-03-16 09:29:15

by Jia He

Subject: Re: [PATCH] Revert "mm/memblock.c: hardcode the end_pfn being -1"

On 3/16/2018 5:06 PM, Michal Hocko Wrote:
> On Thu 15-03-18 19:56:06, Jia He wrote:
>> This reverts commit 379b03b7fa05f7db521b7732a52692448a3c34fe.
>>
>> Commit 864b75f9d6b0 ("mm/page_alloc: fix memmap_init_zone pageblock
>> alignment") introduced boot hang issues in arm/arm64 machines, so
>> Ard Biesheuvel reverted in commit 3e04040df6d4. But there is a
>> preparation patch for commit 864b75f9d6b0. So just revert it for
>> the sake of caution.
> Why? Is there anything wrong with this one?
I don't think there might be anything wrong. Justin for the sake of caution.
Please ignore this patch if you prefer to keep 379b03b7fa.
But seems parameter *max_pfn* is useless and can be removed in this case?

Cheers,
Jia
>> Signed-off-by: Jia He <[email protected]>
>> ---
>> mm/memblock.c | 10 +++++-----
>> 1 file changed, 5 insertions(+), 5 deletions(-)
>>
>> diff --git a/mm/memblock.c b/mm/memblock.c
>> index b6ba6b7..5a9ca2a 100644
>> --- a/mm/memblock.c
>> +++ b/mm/memblock.c
>> @@ -1107,7 +1107,7 @@ unsigned long __init_memblock memblock_next_valid_pfn(unsigned long pfn,
>> struct memblock_type *type = &memblock.memory;
>> unsigned int right = type->cnt;
>> unsigned int mid, left = 0;
>> - phys_addr_t addr = PFN_PHYS(++pfn);
>> + phys_addr_t addr = PFN_PHYS(pfn + 1);
>>
>> do {
>> mid = (right + left) / 2;
>> @@ -1118,15 +1118,15 @@ unsigned long __init_memblock memblock_next_valid_pfn(unsigned long pfn,
>> type->regions[mid].size))
>> left = mid + 1;
>> else {
>> - /* addr is within the region, so pfn is valid */
>> - return pfn;
>> + /* addr is within the region, so pfn + 1 is valid */
>> + return min(pfn + 1, max_pfn);
>> }
>> } while (left < right);
>>
>> if (right == type->cnt)
>> - return -1UL;
>> + return max_pfn;
>> else
>> - return PHYS_PFN(type->regions[right].base);
>> + return min(PHYS_PFN(type->regions[right].base), max_pfn);
>> }
>>
>> /**
>> --
>> 2.7.4
>>

2018-03-16 10:28:53

by Michal Hocko

Subject: Re: [PATCH] Revert "mm/memblock.c: hardcode the end_pfn being -1"

On Fri 16-03-18 17:26:57, Jia He wrote:
>
>
> On 3/16/2018 5:06 PM, Michal Hocko Wrote:
> > On Thu 15-03-18 19:56:06, Jia He wrote:
> > > This reverts commit 379b03b7fa05f7db521b7732a52692448a3c34fe.
> > >
> > > Commit 864b75f9d6b0 ("mm/page_alloc: fix memmap_init_zone pageblock
> > > alignment") introduced boot hang issues in arm/arm64 machines, so
> > > Ard Biesheuvel reverted in commit 3e04040df6d4. But there is a
> > > preparation patch for commit 864b75f9d6b0. So just revert it for
> > > the sake of caution.
> > Why? Is there anything wrong with this one?
> I don't think there might be anything wrong. Justin for the sake of caution.
> Please ignore this patch if you prefer to keep 379b03b7fa.

We do not revert just if the patch is correct. I do not have strong
preference for the patch but I also do not see anything wrong with it.

> But seems parameter *max_pfn* is useless and can be removed in this case?

A patch for that is already sitting in mmotm tree

> Cheers,
> Jia
> > > Signed-off-by: Jia He <[email protected]>
> > > ---
> > > mm/memblock.c | 10 +++++-----
> > > 1 file changed, 5 insertions(+), 5 deletions(-)
> > >
> > > diff --git a/mm/memblock.c b/mm/memblock.c
> > > index b6ba6b7..5a9ca2a 100644
> > > --- a/mm/memblock.c
> > > +++ b/mm/memblock.c
> > > @@ -1107,7 +1107,7 @@ unsigned long __init_memblock memblock_next_valid_pfn(unsigned long pfn,
> > > struct memblock_type *type = &memblock.memory;
> > > unsigned int right = type->cnt;
> > > unsigned int mid, left = 0;
> > > - phys_addr_t addr = PFN_PHYS(++pfn);
> > > + phys_addr_t addr = PFN_PHYS(pfn + 1);
> > > do {
> > > mid = (right + left) / 2;
> > > @@ -1118,15 +1118,15 @@ unsigned long __init_memblock memblock_next_valid_pfn(unsigned long pfn,
> > > type->regions[mid].size))
> > > left = mid + 1;
> > > else {
> > > - /* addr is within the region, so pfn is valid */
> > > - return pfn;
> > > + /* addr is within the region, so pfn + 1 is valid */
> > > + return min(pfn + 1, max_pfn);
> > > }
> > > } while (left < right);
> > > if (right == type->cnt)
> > > - return -1UL;
> > > + return max_pfn;
> > > else
> > > - return PHYS_PFN(type->regions[right].base);
> > > + return min(PHYS_PFN(type->regions[right].base), max_pfn);
> > > }
> > > /**
> > > --
> > > 2.7.4
> > >

--
Michal Hocko
SUSE Labs