Re: [RFC PATCH] memory driver: make phys_index/end_phys_index reflect the start/end section number

[Li Zhong]

On Thu, 2014-04-03 at 11:06 +0800, Zhang Yanfei wrote:
> On 04/03/2014 10:37 AM, Li Zhong wrote:
> > On Thu, 2014-04-03 at 09:37 +0800, Zhang Yanfei wrote:
> >> Add ccing
> >>
> >> On 04/02/2014 04:56 PM, Li Zhong wrote:
> >>> I noticed the phys_index and end_phys_index under 
> >>> /sys/devices/system/memory/memoryXXX/ have the same value, e.g.
> >>> (for the test machine, one memory block has 8 sections, that is 
> >>>  sections_per_block equals 8)
> >>>
> >>> # cd /sys/devices/system/memory/memory100/
> >>> # cat phys_index end_phys_index 
> >>> 00000064
> >>> 00000064
> >>>
> >>> Seems they should reflect the start/end section number respectively, which 
> >>> also matches what is said in Documentation/memory-hotplug.txt
> >>
> > Hi Yanfei, 
> > 
> > Thanks for the review. 
> > 
> >> Indeed. I've noticed this before. The value in 'end_phys_index' doesn't
> >> match what it really means. But, the name itself is vague, it looks like
> >> it is the index of some page frame. (we keep this name for compatibility?)
> > 
> > I guess so, Dave just reminded me that the RFC would also break
> > userspace..
> > 
> > And now my plan is: 
> >  leave the code unchanged
> >  update the document, state the end_phys_index/phys_index are the same,
> > and means the memory block index
> 
> Ah. I doubt whether there is userspace tool which is using the two sysfiles?
> for example, the memory100 directory itself can tell us which block it is.
> So why there is the two files under it give the same meaning.
> 
> If there is userspace tool using the two files, does it use 'end_phys_index'
> in the correct way? That said, if a userspace tool knows what the 'end_phys_index'
> really mean, does it still need it since we have 'phys_index' with the same value?

For the end_phys_index, I totally agree with you. If somebody tries to
use it, say as the end section number, he should finally be able to find
that the value is not what he expects. But who knows ...

For phys_index, I guess it is also reasonable for some userspace tool to
just loop for each memoryXXX directory under /sys/devices/system/memory,
and use the phys_index as the memory block index for the directory,
instead of extracting the XXX from the directory name memoryXXX

Don't know whether there are some generic rules for handling such kind
of compatibility issues...


> >  [optional] create two new files start_sec_nr, end_sec_nr if needed
> 
> These two are the really meaningful sysfiles for userspace, IMO.

OK, I see. 

> 
> > 
> > Do you have any other suggestions? 
> 
> No. I think we should wait for other guys to comment more.

OK, let's wait.

Thanks, Zhong

> 
> Thanks.
> 
> > 
> > Thanks, Zhong
> > 
> >>
> >> The corresponding member in struct memory_block is:
> >>
> >> struct memory_block {
> >>         unsigned long start_section_nr;
> >>         unsigned long end_section_nr;
> >> ...
> >>
> >> The two members seem to have the right name, and have the right value in kernel.
> >>
> >>
> >>>
> >>> This patch tries to modify that so the two files could show the start/end
> >>> section number of the memory block.
> >>>
> >>> After this change, output of the above example looks like:
> >>>
> >>> # cat phys_index end_phys_index 
> >>> 00000320
> >>> 00000327
> >>>
> >>> Signed-off-by: Li Zhong <zhong@xxxxxxxxxxxxxxxxxx>
> >>> ---
> >>>  drivers/base/memory.c | 4 ++--
> >>>  1 file changed, 2 insertions(+), 2 deletions(-)
> >>>
> >>> diff --git a/drivers/base/memory.c b/drivers/base/memory.c
> >>> index bece691..b10f2fa 100644
> >>> --- a/drivers/base/memory.c
> >>> +++ b/drivers/base/memory.c
> >>> @@ -114,7 +114,7 @@ static ssize_t show_mem_start_phys_index(struct device *dev,
> >>>  	struct memory_block *mem = to_memory_block(dev);
> >>>  	unsigned long phys_index;
> >>>  
> >>> -	phys_index = mem->start_section_nr / sections_per_block;
> >>> +	phys_index = mem->start_section_nr;
> >>>  	return sprintf(buf, "%08lx\n", phys_index);
> >>>  }
> >>>  
> >>> @@ -124,7 +124,7 @@ static ssize_t show_mem_end_phys_index(struct device *dev,
> >>>  	struct memory_block *mem = to_memory_block(dev);
> >>>  	unsigned long phys_index;
> >>>  
> >>> -	phys_index = mem->end_section_nr / sections_per_block;
> >>> +	phys_index = mem->end_section_nr;
> >>>  	return sprintf(buf, "%08lx\n", phys_index);
> >>>  }
> >>>  
> >>>
> >>>
> >>> --
> >>> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> >>> the body of a message to majordomo@xxxxxxxxxxxxxxx
> >>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> >>> Please read the FAQ at  http://www.tux.org/lkml/
> >>>
> >>
> >>
> > 
> > 
> > .
> > 
> 
> 


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/