[U-Boot] common/memsize.c: Increase save array for supporting memory size > 4GB

Message ID 1528875163-22096-1-git-send-email-tien.fong.chee@intel.com
State Superseded
Delegated to: Tom Rini
Headers show
Series
  • [U-Boot] common/memsize.c: Increase save array for supporting memory size > 4GB
Related show

Commit Message

Chee, Tien Fong June 13, 2018, 7:32 a.m.
From: Tien Fong Chee <tien.fong.chee@intel.com>

In ARM 64-bits, memory size can be supported is more than 4GB,
hence increasing save array is needed to cope with testing larger memory.

Signed-off-by: Tien Fong Chee <tien.fong.chee@intel.com>
---
 common/memsize.c |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

Comments

Tom Rini June 18, 2018, 4:59 p.m. | #1
On Wed, Jun 13, 2018 at 03:32:43PM +0800, tien.fong.chee@intel.com wrote:

> From: Tien Fong Chee <tien.fong.chee@intel.com>
> 
> In ARM 64-bits, memory size can be supported is more than 4GB,
> hence increasing save array is needed to cope with testing larger memory.
> 
> Signed-off-by: Tien Fong Chee <tien.fong.chee@intel.com>
> ---
>  common/memsize.c |    2 +-
>  1 files changed, 1 insertions(+), 1 deletions(-)
> 
> diff --git a/common/memsize.c b/common/memsize.c
> index 5670e95..b091203 100644
> --- a/common/memsize.c
> +++ b/common/memsize.c
> @@ -26,7 +26,7 @@ DECLARE_GLOBAL_DATA_PTR;
>  long get_ram_size(long *base, long maxsize)
>  {
>  	volatile long *addr;
> -	long           save[31];
> +	long           save[BITS_PER_LONG];
>  	long           save_base;
>  	long           cnt;
>  	long           val;

Since BITS_PER_LONG is 32 or 64, shouldn't we use B_P_L - 1 here?  Or
are you saying there's also a case where this is wrong on 32bit today?
Thanks!
Chee, Tien Fong June 19, 2018, 5:52 a.m. | #2
On Mon, 2018-06-18 at 12:59 -0400, Tom Rini wrote:
> On Wed, Jun 13, 2018 at 03:32:43PM +0800, tien.fong.chee@intel.com
> wrote:
> 
> > 
> > From: Tien Fong Chee <tien.fong.chee@intel.com>
> > 
> > In ARM 64-bits, memory size can be supported is more than 4GB,
> > hence increasing save array is needed to cope with testing larger
> > memory.
> > 
> > Signed-off-by: Tien Fong Chee <tien.fong.chee@intel.com>
> > ---
> >  common/memsize.c |    2 +-
> >  1 files changed, 1 insertions(+), 1 deletions(-)
> > 
> > diff --git a/common/memsize.c b/common/memsize.c
> > index 5670e95..b091203 100644
> > --- a/common/memsize.c
> > +++ b/common/memsize.c
> > @@ -26,7 +26,7 @@ DECLARE_GLOBAL_DATA_PTR;
> >  long get_ram_size(long *base, long maxsize)
> >  {
> >  	volatile long *addr;
> > -	long           save[31];
> > +	long           save[BITS_PER_LONG];
> >  	long           save_base;
> >  	long           cnt;
> >  	long           val;
> Since BITS_PER_LONG is 32 or 64, shouldn't we use B_P_L - 1 here?  Or
> are you saying there's also a case where this is wrong on 32bit
> today?
As long as the array is large enough to cope with shifting
implementation, then it should be fine. For 32-bit, only 31 units in
array required for storing 31 shifting values, and this apply for 64-
bit also as long as the implementation of first shifting value is not
change(cnt = (maxsize / sizeof(long)) >> 1).
IMO, for simplifying and safety purpose(may be one day implementation
change to "cnt = (maxsize / sizeof(long))", above B_P_L is still
workable.
> Thanks!
>
Marek Vasut June 19, 2018, 6:01 a.m. | #3
On 06/19/2018 07:52 AM, Chee, Tien Fong wrote:
> On Mon, 2018-06-18 at 12:59 -0400, Tom Rini wrote:
>> On Wed, Jun 13, 2018 at 03:32:43PM +0800, tien.fong.chee@intel.com
>> wrote:
>>
>>>
>>> From: Tien Fong Chee <tien.fong.chee@intel.com>
>>>
>>> In ARM 64-bits, memory size can be supported is more than 4GB,
>>> hence increasing save array is needed to cope with testing larger
>>> memory.
>>>
>>> Signed-off-by: Tien Fong Chee <tien.fong.chee@intel.com>
>>> ---
>>>  common/memsize.c |    2 +-
>>>  1 files changed, 1 insertions(+), 1 deletions(-)
>>>
>>> diff --git a/common/memsize.c b/common/memsize.c
>>> index 5670e95..b091203 100644
>>> --- a/common/memsize.c
>>> +++ b/common/memsize.c
>>> @@ -26,7 +26,7 @@ DECLARE_GLOBAL_DATA_PTR;
>>>  long get_ram_size(long *base, long maxsize)
>>>  {
>>>  	volatile long *addr;
>>> -	long           save[31];
>>> +	long           save[BITS_PER_LONG];
>>>  	long           save_base;
>>>  	long           cnt;
>>>  	long           val;
>> Since BITS_PER_LONG is 32 or 64, shouldn't we use B_P_L - 1 here?  Or
>> are you saying there's also a case where this is wrong on 32bit
>> today?
> As long as the array is large enough to cope with shifting
> implementation, then it should be fine. For 32-bit, only 31 units in
> array required for storing 31 shifting values, and this apply for 64-
> bit also as long as the implementation of first shifting value is not
> change(cnt = (maxsize / sizeof(long)) >> 1).
> IMO, for simplifying and safety purpose(may be one day implementation
> change to "cnt = (maxsize / sizeof(long))", above B_P_L is still
> workable.

That's BS reasoning and just sloppy programming. I agree with Tom.

Patch

diff --git a/common/memsize.c b/common/memsize.c
index 5670e95..b091203 100644
--- a/common/memsize.c
+++ b/common/memsize.c
@@ -26,7 +26,7 @@  DECLARE_GLOBAL_DATA_PTR;
 long get_ram_size(long *base, long maxsize)
 {
 	volatile long *addr;
-	long           save[31];
+	long           save[BITS_PER_LONG];
 	long           save_base;
 	long           cnt;
 	long           val;