Re: [PATCH v7u1 26/31] x86: Don't enable swiotlb if there is notenough ram for it

[Yinghai Lu]

On Wed, Jan 9, 2013 at 10:01 AM, Shuah Khan <shuahkhan@xxxxxxxxx> wrote:
> After several revisions, I am loosing track. Could you please write a
> change log and explain the change to the existing behavior. If you
> could addresses the following areas, it will be easier figure if we
> are missing something (if any):
>
> 1. What happens when switolb is forced with iommu=soft on a system
> with and without not enough low mem?

panic later when device try to map_single with it.
instead of panic early during swiotlb allocation.

> 2. What happens when swiotlb is not forced, but iommu driver sets
> swiotlb=1 after it gets done with its iommu initialization on a system
> with and without enough low mem.

panic later when device try to map_single with it.
instead of panic early during swiotlb allocation.

> 3. What happens when nopanic is used and when will the system fail?
> Will it fail when driver runs into errors. I am hoping this won't be a
> silent failure. Please see more on this below:

according to eric/konrad, I make all use nopanic path,
and only mips, and x86-xen still panic early when
swiotlb_init_with_tab is called
directly.

hope mips, and x86-xen guys could convert to calling swiotlb_init instead.

>
> I did dma mapping error analysis a few months ago and found several
> drivers that don't check dma mapping errors, don't unmap dma buffers
> etc. Returning mapping error from switolb could cause problems when we
> have drivers that don't check mapping errors. These drivers might not
> fail cleanly either and could cause data corruption.

Now, we are using panic with them, later after those drivers get fixed, we
can change to returning MAP_ERROR instead.

>
> Here is a link to the dma mapping error analysis I did:
>
> http://linuxdriverproject.org/mediawiki/index.php/DMA_Mapping_Error_Analysis

good, looks like need one generic way to fix the problem, like another wrapper?

Thanks

Yinghai
--
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/