On 07/08/2015 07:58 PM, David Rientjes wrote:

On Wed, 1 Jul 2015, Sasha Levin wrote:

quoted

quoted

Since we'd BUG at VM_BUG_ON(), this would be something closer to:

	if (unlikely(compound_head(page) != head)) {
		dump_page(page);
		dump_page(head);
		VM_BUG_ON(1);
	}

I was thinking closer to

	if (VM_WARN_ON(compound_head(page) != head)) {
		...
		BUG();
	}

so we prefix all output with the typical warning diagnostics, emit 
whatever page, vma, etc output we want, and then finally die.  The final 
BUG() here would have to be replaced by something that suppresses the 
repeated output.

If it's really just a warning, then no BUG() needed.

How is that simpler than getting it all under VM_BUG()? Just like the regular
WARN() does.

quoted

quoted

But my point here was that while one *could* do it that way, no one does because
it's not intuitive. We both agree that in the example above it would be useful to
see both 'page' and 'head', and yet the code that was written didn't dump any of
them. Why? No one wants to write debug code unless it's easy and short.

pr_alert("%pZp %pZv", page, vma) isn't shorter than dump_page(page); 
dump_vma(vma), but it would be a line shorter.  I'm not sure that the 
former is easier, though, and it prevents us from ever expanding dump_*() 
functions for conditional output.

I'm not objecting to leaving dump_*() for these trivial cases.


Thanks,
Sasha

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