Re: [PATCH v7u1 26/31] x86: Don't enable swiotlb if there is not enough ram for it
From: Eric W. Biederman
Date: Mon Jan 07 2013 - 22:13:45 EST
Yinghai Lu <yinghai@xxxxxxxxxx> writes:
> On Mon, Jan 7, 2013 at 6:22 PM, Eric W. Biederman <ebiederm@xxxxxxxxxxxx> wrote:
>> Yinghai I sat down and read your patch and the approach you are taking
>> is totally wrong.
> Thanks for check the patch, did you check v3?
I looked at the version of the patch you had as an attachment. I don't
know the version number but it was the latest version of the patch I saw
in this thread.
After looking at things having a function enoung_mem_for_swiotlb()
in pci_swiotlb_detect_override() and pic_swiotlb_detect_4gb is brittle
>> The problem is that swiotlb_init() in lib/swiotlb.c does not know how to
>> fail without panic'ing the system.
> I did not put panic in swiotlb, now I put panic in amd_iommu ops init
> when it need extra
> swiotlb for unhandled devices by AMD IOMMU.
But the only reason you need to touch this code at all is that
swiotlb_init() calls panic() if you don't have 64MB of memory below 4G.
>> Which leaves two valid approaches.
>> - Create a variant of swiotlb_init that can fail for use on x86 and
>> handle the failure.
>> - Delay initializing the swiotlb until someone actually needs a mapping
>> from it.
>> Delaying the initialization of the swiotlb is out because the code
>> needs an early memory allocation to get a large chunk of contiguous
>> memory for the bounce buffers.
>> Which means the panics that occurr in swiotlb_init() need to be delayed
>> until someone something actually needs bounce buffers from the swiotlb.
>> Although arguably what should actually happen instead of panic() is that
>> swiotlb_map_single should simply fail early when it was not possible to
>> preallocate bounce buffers.
> do you mean: actually needed dma buffer is much less than swiotlb
> buffer aka 64M?
I meant we should detect failure to allocate bounce buffers in in
swiotlb_init() instead of panicing.
I meant swiotlb_map_single() should either panic or simply fail.
If I have read lib/swiotlb.c correctly the only place we allocate a
bounce buffer is in swiotlb_map_single. If there are more places we can
allocate bounce buffers those need to be handled as well.
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/