RE: [PATCH] mm: cma: split cma-reserved in dmesg log

[PINTU KUMAR]

Hi,

----- Original Message -----
> From: Gioh Kim <gioh.kim@xxxxxxx>
> To: Pintu Kumar <pintu.k@xxxxxxxxxxx>; akpm@xxxxxxxxxxxxxxxxxxxx;
hannes@xxxxxxxxxxx; riel@xxxxxxxxxx; mgorman@xxxxxxx;
vdavydov@xxxxxxxxxxxxx; nasa4836@xxxxxxxxx; ddstreet@xxxxxxxx;
m.szyprowski@xxxxxxxxxxx; mina86@xxxxxxxxxx; iamjoonsoo.kim@xxxxxxx; aneesh.
kumar@xxxxxxxxxxxxxxxxxx; lauraa@xxxxxxxxxxxxxx; rientjes@xxxxxxxxxx;
vbabka@xxxxxxx; sasha.levin@xxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; linux-
mm@xxxxxxxxx
> Cc: cpgs@xxxxxxxxxxx; pintu_agarwal@xxxxxxxxx; vishnu.ps@xxxxxxxxxxx;
rohit.kr@xxxxxxxxxxx; ed.savinay@xxxxxxxxxxx; 이건호 <gunho.lee@xxxxxxx>
> Sent: Tuesday, 21 October 2014 6:17 AM
> Subject: Re: [PATCH] mm: cma: split cma-reserved in dmesg log
> 
> 
> 
> 2014-10-20 오후 4:33, Pintu Kumar 쓴 글:
>> When the system boots up, in the dmesg logs we can see
>> the memory statistics along with total reserved as below.
>> Memory: 458840k/458840k available, 65448k reserved, 0K highmem
>> 
>> When CMA is enabled, still the total reserved memory remains the same.
>> However, the CMA memory is not considered as reserved.
>> But, when we see /proc/meminfo, the CMA memory is part of free memory.
>> This creates confusion.
>> This patch corrects the problem by properly substracting the CMA reserved
>> memory from the total reserved memory in dmesg logs.
>> 
>> Below is the dmesg snaphot from an arm based device with 512MB RAM and
>> 12MB single CMA region.
>> 
>> Before this change:
>> Memory: 458840k/458840k available, 65448k reserved, 0K highmem
>> 
>> After this change:
>> Memory: 458840k/458840k available, 53160k reserved, 12288k cma-reserved,
0K 
> highmem
>> 
>> Signed-off-by: Pintu Kumar <pintu.k@xxxxxxxxxxx>
>> Signed-off-by: Vishnu Pratap Singh <vishnu.ps@xxxxxxxxxxx>
>> ---
>>   include/linux/swap.h | 3 +++
>>   mm/cma.c            | 2 ++
>>   mm/page_alloc.c      | 8 ++++++++
>>   3 files changed, 13 insertions(+)
>> 
>> diff --git a/include/linux/swap.h b/include/linux/swap.h
>> index 37a585b..beb84be 100644
>> --- a/include/linux/swap.h
>> +++ b/include/linux/swap.h
>> @@ -295,6 +295,9 @@ static inline void
workingset_node_shadows_dec(struct 
> radix_tree_node *node)
>>   /* linux/mm/page_alloc.c */
>>   extern unsigned long totalram_pages;
>>   extern unsigned long totalreserve_pages;
>> +#ifdef CONFIG_CMA
>> +extern unsigned long totalcma_pages;
>> +#endif

Ok, as per Andrew Morton comment, will remove CONFIG_CMA, 
But then I need to put it under: include/linux/mm.h
In that case, it will solve the problem for CMA and non-CMA case.
Because, mm.h is already included in cma.c

>>   extern unsigned long dirty_balance_reserve;
>>   extern unsigned long nr_free_buffer_pages(void);
>>   extern unsigned long nr_free_pagecache_pages(void);
>> diff --git a/mm/cma.c b/mm/cma.c
>> index 963bc4a..73fe7be 100644
>> --- a/mm/cma.c
>> +++ b/mm/cma.c
>> @@ -45,6 +45,7 @@ struct cma {
>>   static struct cma cma_areas[MAX_CMA_AREAS];
>>   static unsigned cma_area_count;
>>   static DEFINE_MUTEX(cma_mutex);
>> +unsigned long totalcma_pages __read_mostly;
> 
> I think __read_mostly is not good here.
> Cma areas often are rare
> but we cannot expect how many cma areas exists.
> 

Firstly, I want to move this to mm/page_alloc.c, so that it can be visible
for non-CMA cases.
Next, the purpose this variable is not only during init time.
Just like totalram_pages, I wanted to retain this variable to use it to
populate the CMA info, during /proc/meminfo.
Like:
CMATotal: (using totalcma_pages)
CMAFree:  (using NR_FREE_CMA_PAGES)
I will post these changes in the next patch series.
Please let me know your comments.

>>   
>>   phys_addr_t cma_get_base(struct cma *cma)
>>   {
>> @@ -288,6 +289,7 @@ int __init cma_declare_contiguous(phys_addr_t base,
>>       if (ret)
>>           goto err;
>>   
>> +    totalcma_pages += (size / PAGE_SIZE);
>>       pr_info("Reserved %ld MiB at %08lx\n", (unsigned 
> long)size / SZ_1M,
>>           (unsigned long)base);
>>       return 0;
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index dd73f9a..c6165ac 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -5521,6 +5521,9 @@ void __init mem_init_print_info(const char *str)
>>       pr_info("Memory: %luK/%luK available "
>>             "(%luK kernel code, %luK rwdata, %luK rodata, "
>>             "%luK init, %luK bss, %luK reserved"
>> +#ifdef CONFIG_CMA
>> +        ", %luK cma-reserved"
>> +#endif
>>   #ifdef    CONFIG_HIGHMEM
>>             ", %luK highmem"
>>   #endif
>> @@ -5528,7 +5531,12 @@ void __init mem_init_print_info(const char *str)
>>             nr_free_pages() << (PAGE_SHIFT-10), physpages << 
> (PAGE_SHIFT-10),
>>             codesize >> 10, datasize >> 10, rosize >> 
> 10,
>>             (init_data_size + init_code_size) >> 10, bss_size 
>>> 10,
>> +#ifdef CONFIG_CMA
>> +          (physpages - totalram_pages - totalcma_pages) << 
> (PAGE_SHIFT-10),
>> +          totalcma_pages << (PAGE_SHIFT-10),
>> +#else
>>             (physpages - totalram_pages) << (PAGE_SHIFT-10),
>> +#endif
>>   #ifdef    CONFIG_HIGHMEM
>>             totalhigh_pages << (PAGE_SHIFT-10),
>>   #endif
>> 
> 
> I basically agree with your point.
> But CMA feature is not popular yet, so memory develoers probably doesn't 
> like this.
> 

Ok agree. If we move totalcma_pages declaration to page_alloc.c and mm.h,
then we can get rid of CONFIG_CMA, to make it neat.


> I'm not sure but I think there is a debugfs file for cma.
> Can you use it?
> 

As of now, I think there is no debugfs for cma.
However, we can make one if required.

> Or what do you think about making another proc file to show cma area size
and 
> address?
> For instance,
> 
> # cat /proc/cmainfo
> CMATotal:    400kB
> 0x10000000  300kB
> 0x20000000  100kB
> 

I think this is not required.
For multiple CMA regions, this can be found under:
/sys/kernel/debug/memblock/reserved
However, as I said, we can populate this information under: /proc/meminfo
I think capturing it at one place will be better. For non-CMA cases, it
will be hidden.
Thus summary of CMA info can be seen in meminfo, and detailed information
can be seen in memblock/reserved.
Do, let me know if you have any other idea?

> 
> --
> To unsubscribe, send a message with 'unsubscribe linux-mm' in
> the body to majordomo@xxxxxxxxxx  For more info on Linux MM,
> see: http://www.linux-mm.org/ .
> Don't email: <a href=mailto:"dont@xxxxxxxxx";> 
> email@xxxxxxxxx </a>
>

--
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/