[PATCH V1 2/2] xen/grant-table: Use unpopulated contiguous pages instead of real RAM ones

[Oleksandr Tyshchenko <olekstysh@xxxxxxxxx>]

From: Oleksandr Tyshchenko <oleksandr_tyshchenko@xxxxxxxx>

Depends on CONFIG_XEN_UNPOPULATED_ALLOC. If enabled then unpopulated
contiguous pages will be allocated for grant mapping into instead of
ballooning out real RAM pages.

Also fallback to allocate DMAable pages (balloon out real RAM pages)
if we failed to allocate unpopulated contiguous pages. Use recently
introduced is_xen_unpopulated_page() in gnttab_dma_free_pages() to know
what API to use for freeing pages.

Signed-off-by: Oleksandr Tyshchenko <oleksandr_tyshchenko@xxxxxxxx>
---
Please note, I haven't re-checked yet the use-case where the xen-swiotlb
is involved (proposed by Stefano):
https://lore.kernel.org/xen-devel/alpine.DEB.2.22.394.2206031348230.2783803@ubuntu-linux-20-04-desktop/
I will re-check that for next version and add corresponding comment
in the code.

Changes RFC -> V1:
   - update commit subject/description
   - rework to avoid introducing alternative implementation
     of gnttab_dma_alloc(free)_pages(), use IS_ENABLED()
   - implement a fallback to real RAM pages if we failed to allocate
     unpopulated contiguous pages (resolve initial TODO)
   - update according to the API renaming (s/dma/contiguous)
---
 drivers/xen/grant-table.c | 24 ++++++++++++++++++++++++
 1 file changed, 24 insertions(+)

diff --git a/drivers/xen/grant-table.c b/drivers/xen/grant-table.c
index 738029d..15e426b 100644
--- a/drivers/xen/grant-table.c
+++ b/drivers/xen/grant-table.c
@@ -1047,6 +1047,23 @@ int gnttab_dma_alloc_pages(struct gnttab_dma_alloc_args 
*args)
        size_t size;
        int i, ret;
 
+       if (IS_ENABLED(CONFIG_XEN_UNPOPULATED_ALLOC)) {
+               ret = xen_alloc_unpopulated_contiguous_pages(args->dev, 
args->nr_pages,
+                               args->pages);
+               if (ret < 0)
+                       goto fallback;
+
+               ret = gnttab_pages_set_private(args->nr_pages, args->pages);
+               if (ret < 0)
+                       goto fail;
+
+               args->vaddr = page_to_virt(args->pages[0]);
+               args->dev_bus_addr = page_to_phys(args->pages[0]);
+
+               return ret;
+       }
+
+fallback:
        size = args->nr_pages << PAGE_SHIFT;
        if (args->coherent)
                args->vaddr = dma_alloc_coherent(args->dev, size,
@@ -1103,6 +1120,13 @@ int gnttab_dma_free_pages(struct gnttab_dma_alloc_args 
*args)
 
        gnttab_pages_clear_private(args->nr_pages, args->pages);
 
+       if (IS_ENABLED(CONFIG_XEN_UNPOPULATED_ALLOC) &&
+                       is_xen_unpopulated_page(args->pages[0])) {
+               xen_free_unpopulated_contiguous_pages(args->dev, args->nr_pages,
+                               args->pages);
+               return 0;
+       }
+
        for (i = 0; i < args->nr_pages; i++)
                args->frames[i] = page_to_xen_pfn(args->pages[i]);
 
-- 
2.7.4