Re: [PATCH V2 1/1][cleanup] memcg: renaming of mem variable to memcg

[Raghavendra K T <raghukt@linux.vnet.ibm.com>]

On 08/17/2011 06:13 PM, Michal Hocko wrote:
> Sorry for late reply
>
> On Fri 12-08-11 12:36:23, Raghavendra K T wrote:
>>   The memcg code sometimes uses "struct mem_cgroup *mem" and sometimes uses
>>   "struct mem_cgroup *memcg". This patch renames all mem variables to memcg in
>>   source file.
>>
>> Testing : Compile tested with following configurations.
>> 1) make defconfig ARCH=i386 + CONFIG_CGROUP_MEM_RES_CTLR=y
>> CONFIG_CGROUP_MEM_RES_CTLR_SWAP=y CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED=y
>>
>> Binary size Before patch
>> ========================
>>     text	   data	    bss	    dec	    hex	filename
>> 8911169	 520464	1884160	11315793	 acaa51	vmlinux
>>
>> Binary Size After patch
>> =======================
>>     text	   data	    bss	    dec	    hex	filename
>> 8911169	 520464	1884160	11315793	 acaa51	vmlinux
>
> It would be much nicer to see unchanged md5sum. I am not sure how much
> possible is this with current gcc or whether special command line
> parameters have to be used (at least !CONFIG_DEBUG_INFO* is necessary)
> but simple variable rename shouldn't be binary visible.
> I guess that a similar approach was used during 32b and 64b x86
> unification.
>
I agree,  I could get same MD5 sum only in N N N config case (3rd config).
  I am not sure whether we can get same Md5 after lines have been split.
Here is what I tried: static KBUILD_BUILD_TIMESTAMP 
KBUILD_BUILD_VERSION, initramfs source. strip vmlinux. (could not 
disable CONFIG_BUG).
I referred to 32nb 64b unification also, did not get much insight from 
there on same MD5.

>>
>> 2) make defconfig ARCH=i386 + CONFIG_CGROUP_MEM_RES_CTLR=y
>> CONFIG_CGROUP_MEM_RES_CTLR_SWAP=n CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED=n
>
> I would assume the same testing results as above
Yes
8908671	 519808	1884160	11312639	 ac9dff	vmlinux
>
>>
>> 3) make defconfig ARCH=i386  CONFIG_CGROUP_MEM_RES_CTLR=n
>> CONFIG_CGROUP_MEM_RES_CTLR_SWAP=n CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED=n
>
> ditto.

8878794	 517632	1880064	11276490	 ac10ca	vmlinux before and after
>
>>
>> Other sanity check:
>> Bootable configuration on x86 (T60p)  with  CONFIG_CGROUP_MEM_RES_CTLR=y
>> CONFIG_CGROUP_MEM_RES_CTLR_SWAP=y CONFIG_CGROUP_MEM_RES_CTLR_SWAP_ENABLED=y
>> is tesed with basic mounting of memcgroup, creation of child and parallel fault.
>> mkdir -p /cgroup
>> mount -t cgroup none /cgroup -o memory
>> mkdir /cgroup/0
>> echo $$>  /cgroup/0/tasks
>> time ./parallel_fault 2 100000 32
>>
>> real	0m0.025s
>> user	0m0.001s
>> sys	0m0.033s
>
> This looks like a random test. I wouldn't add it to the changelog.
Agree.