Re: [PATCH] mm: append __GFP_COMP flag for trace_malloc

[Vlastimil Babka <vbabka@suse.cz>]

On 4/28/21 5:05 AM, Xiongwei Song wrote:
> On Tue, Apr 27, 2021 at 7:26 PM Matthew Wilcox <willy@infradead.org> wrote:
>>
>> On Tue, Apr 27, 2021 at 01:30:48PM +0800, Xiongwei Song wrote:
>> > Hi Mattew,
>> >
>> > One more thing I should explain, the kmalloc_order() appends the
>> > __GFP_COMP flags,
>> > not by the caller.
>> >
>> > void *kmalloc_order(size_t size, gfp_t flags, unsigned int order)
>> > {
>> > ...........................................................
>> >
>> > flags |= __GFP_COMP;
>> > page = alloc_pages(flags, order);
>> > ...........................................................
>> > return ret;
>> > }
>> > EXPORT_SYMBOL(kmalloc_order);
>> >
>> > #ifdef CONFIG_TRACING
>> > void *kmalloc_order_trace(size_t size, gfp_t flags, unsigned int order)
>> > {
>> > void *ret = kmalloc_order(size, flags, order);
>> > trace_kmalloc(_RET_IP_, ret, size, PAGE_SIZE << order, flags);
>> > return ret;
>> > }
>> > EXPORT_SYMBOL(kmalloc_order_trace);
>> > #endif
>>
>> Yes, I understood that.  What I don't understand is why appending the
>> __GFP_COMP to the trace would have been less confusing for you.
>>
>> Suppose I have some code which calls:
>>
>>         kmalloc(10 * 1024, GFP_ATOMIC|__GFP_NOWARN|__GFP_NOMEMALLOC);
>>
>> and I see in my logs
>>
>>      0.08%  call_site=ffffffff851d0cb0 ptr=0xffff8c04a4ca0000 bytes_req=10176 bytes_alloc=16384 gfp_flags=GFP_ATOMIC|__GFP_NOWARN|__GFP_NOMEMALLOC|__GFP_COMP
>>
>> That seems to me _more_ confusing because I would wonder "Where did that
>> __GFP_COMP come from?"
> 
> Thank you for the comments. But I disagree.

FTR, I agree with Matthew. This is a tracepoint for kmalloc() so I would expect
to see what flags were passed to kmalloc().
If I wanted to see how the flags translated to page allocator's flags, I would
have used a page allocator's tracepoint which would show me that.

> When I use trace, I hope I can get the precise data rather than something
> changed that I don't know , then I can get the correct conclusion or
> direction on my issue.

It's precise from the point of the caller.

> Here my question is what the trace events are for if they don't provide the
> real situation? I think that's not graceful and friendly.
> 
> From my perspective, it'd be better to know my flags changed before checking
> code lines one by one. In other words, I need a warning to reminder me on this,
> then I can know quickly my process might do some incorrect things.

Your process should not care about __GFP_COMP if you use properly
kmalloc()+kfree(). Once you start caring about __GFP_COMP, you should be using
page allocator's API, not kmalloc().

> Regards,
> Xiongwei
>