Re: [PATCH] Use kzfree in tty buffer management to enforce data sanitization

[Pekka Enberg <penberg@cs.helsinki.fi>]

Hi Linus,

On Sun, 31 May 2009, Alan Cox wrote:
>>>>        memset(buf->data, 0, N_TTY_BUF_SIZE);
>>>>        if (PAGE_SIZE != N_TTY_BUF_SIZE)
>>>>                kfree(...)
>>>>        else
>>>>                free_page(...)
>>>>
>>>>
>>>> but quite frankly, I'm not convinced about these patches at all.
>>> I wonder why the tty code has that N_TTY_BUF_SIZE special casing in
>>> the first place? I think we can probably just get rid of it and thus
>>> we can use kzfree() here if we want to.
>> Some platforms with very large page sizes override the use of page based
>> allocators (eg older ARM would go around allocating 32K). The normal path
>> is 4K or 8K page sized buffers.

Linus Torvalds wrote:
> I think Pekka meant the other way around - why don't we always just use 
> kmalloc(N_TTY_BUF_SIZE)/kfree(), and drop the whole conditional "use page 
> allocator" entirely?
> 
> I suspect the "use page allocator" is historical - ie the tty layer 
> originally always did that, and then when people wanted to suppotr smaller 
> areas than one page, they added the special case. I have this dim memory 
> of the _original_ kmalloc not handling page-sized allocations well (due to 
> embedded size/pointer overheads), but I think all current allocators are 
> perfectly happy to allocate PAGE_SIZE buffers without slop.
> 
> If I'm right, then we could just use kmalloc/kfree unconditionally. Pekka?

Yup, that's what I meant. Even SLAB moves metadata off-slab to make sure 
  we support PAGE_SIZE allocations nicely. SLUB even used to pass 
kmalloc(PAGE_SIZE) directly to the page allocator and will likely do 
that again once Mel Gorman's page allocator optimization patches hit 
mainline.

			Pekka

--
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>