Re: [PATCH v3 02/10] swiotlb: Factor out slot allocation and free

[Lu Baolu <baolu.lu@linux.intel.com>]

Hi,

On 4/24/19 10:45 PM, Christoph Hellwig wrote:
> On Tue, Apr 23, 2019 at 03:32:16PM +0800, Lu Baolu wrote:
>> When we add the bounce buffer between IOVA and physical buffer, the
>> bounced buffer must starts from the same offset in a page, otherwise,
>> IOMMU can't work here.
> 
> Why?  Even with the odd hardware descriptors typical in Intel land that
> only allow offsets in the first page, not the following ones, having
> a zero offset where we previously had one should be fine.
> 

This is not VT-d specific. It's just how generic IOMMU works.

Normally, IOMMU works in paging mode. So if a driver issues DMA with
IOVA  0xAAAA0123, IOMMU can remap it with a physical address 0xBBBB0123.
But we should never expect IOMMU to remap 0xAAAA0123 with physical
address of 0xBBBB0000. That's the reason why I said that IOMMU will not
work there.

                                 swiotlb                  System
                 IOVA          bounce page                Memory
              .---------.      .---------.             .---------.
              |         |      |         |             |         |
              |         |      |         |             |         |
buffer_start .---------.      .---------.buffer_start .---------.
              |         |----->|         |             |         |
              |         |      |         |************>|         |
              |         |      |         |   swiotlb   |         |
  IOMMU Page  '---------'      '---------'   mapping   '---------'
   Boundary   |         |                              |         |
              |         |                              |         |
              |         |                              |         |
              |         |----------------------------->|         |
              |         |        IOMMU mapping         |         |
              |         |                              |         |
  IOMMU Page  .---------.                              .---------.
   Boundary   |         |                              |         |
              |         |                              |         |
              |         |----------------------------->|         |
              |         |         IOMMU mapping        |         |
              |         |                              |         |
              |         |                              |         |
  IOMMU Page  .---------.      .---------.             .---------.
   Boundary   |         |      |         |             |         |
              |         |      |         |************>|         |
              |         |----->|         |   swiotlb   |         |
   buffer_end '---------'      '---------'   mapping   '---------'
              |         |      |         |             |         |
              |         |      |         |             |         |
              '---------'      '---------'             '---------'


This is the whole view of iommu bounce page. I expect the buffer_start
returned by swiotlb_tbl_map_single() starts from the same page_offset as
the buffer_start in IOVA.

Best regards,
Lu Baolu