Messages in this thread | | | Date | Thu, 11 Sep 2008 00:09:43 +0900 | Subject | Re: [PATCH] x86: avoid unnecessary low zone allocation in AMD IOMMU's alloc_coherent | From | FUJITA Tomonori <> |
| |
On Wed, 10 Sep 2008 16:52:49 +0200 Joerg Roedel <joerg.roedel@amd.com> wrote:
> On Wed, Sep 10, 2008 at 11:39:00PM +0900, FUJITA Tomonori wrote: > > > btw, in tip/x86/iommu, GART's alloc_coherent always does virtual > > mappings to allocate a size-aligned memory (as DMA-mapping.txt > > defines). > > > > Because someone strongly insisted, I modified GART's alloc_coherent to > > do so but as I said again and again, it's completely meaningless (only > > POWER IOMMU does it and drivers don't depend on such requirement). > > > > I guess that it would be better to do virtual mappings only when > > necessary as the current mainline does since GART I/O space is > > precious in some systems. But I don't care much. What's your opinion > > (as a AMD developer)? > > Very true. My original rewrite did the mapping only when necessary too. > What were the reasons to do the mapping always?
As I said above, it's for allocating a size-aligned memory. Look at the description of pci_alloc_consistent in DMA-mapping.txt:
The cpu return address and the DMA bus master address are both guaranteed to be aligned to the smallest PAGE_SIZE order which is greater than or equal to the requested size. This invariant exists (for example) to guarantee that if you allocate a chunk which is smaller than or equal to 64 kilobytes, the extent of the buffer you receive will not cross a 64K boundary.
You can't do this with __get_free_pages easily (you need some hacks to do this). You can do this via iommu_area_alloc() for free.
Well, actually you agreed with adding such requirement (though I said again and again that it's totally meaningless...):
http://lkml.org/lkml/2008/7/24/162
| |