On Mon, 18 Jan 2016, Shannon Zhao wrote:
> On 2016/1/16 1:08, Stefano Stabellini wrote:
> > On Fri, 15 Jan 2016, Shannon Zhao wrote:
> >> From: Shannon Zhao <shannon.zhao@xxxxxxxxxx>
> >>
> >> Use page_to_xen_pfn in case of 64KB page.
> >>
> >> Signed-off-by: Shannon Zhao <shannon.zhao@xxxxxxxxxx>
> >> ---
> >> drivers/xen/xlate_mmu.c | 2 +-
> >> 1 file changed, 1 insertion(+), 1 deletion(-)
> >>
> >> diff --git a/drivers/xen/xlate_mmu.c b/drivers/xen/xlate_mmu.c
> >> index 9692656..b9fcc2c 100644
> >> --- a/drivers/xen/xlate_mmu.c
> >> +++ b/drivers/xen/xlate_mmu.c
> >> @@ -227,7 +227,7 @@ int __init xen_xlate_map_ballooned_pages(xen_pfn_t **gfns, void **virt,
> >> return rc;
> >> }
> >> for (i = 0; i < nr_grant_frames; i++)
> >> - pfns[i] = page_to_pfn(pages[i]);
> >> + pfns[i] = page_to_xen_pfn(pages[i]);
> >
> > Shannon, thanks for the patch.
> >
> > Keeping in mind that in the 64K case, kernel pages are 64K but xen pages
> > are still 4K, I think you also need to allocate
> > (nr_grant_frames/XEN_PFN_PER_PAGE) kernel pages (assuming that they are
> > allocated contiguously): nr_grant_frames refers to 4K pages, while
> > xen_xlate_map_ballooned_pages is allocating pages on a 64K granularity
> > (sizeof(pages[0]) == 64K).
> >
> > Be careful that alloc_xenballooned_pages deals with 64K pages (because
> > it deals with kernel pages), while xen_pfn_t is always 4K based (because
> > it deals with Xen pfns).
> >
> > Please test it with and without CONFIG_ARM64_64K_PAGES. Thanks!
> >
> Stefano, thanks for your explanation. How about below patch?
Good work, it looks like you covered all bases, I think it should work,
but I haven't tested it myself. Just one note below.
> diff --git a/drivers/xen/xlate_mmu.c b/drivers/xen/xlate_mmu.c
> index 9692656..e1f7c95 100644
> --- a/drivers/xen/xlate_mmu.c
> +++ b/drivers/xen/xlate_mmu.c
> @@ -207,9 +207,12 @@ int __init xen_xlate_map_ballooned_pages(xen_pfn_t
> **gfns, void **virt,
> void *vaddr;
> int rc;
> unsigned int i;
> + unsigned long nr_pages;
> + xen_pfn_t xen_pfn = 0;
>
> BUG_ON(nr_grant_frames == 0);
> - pages = kcalloc(nr_grant_frames, sizeof(pages[0]), GFP_KERNEL);
> + nr_pages = DIV_ROUND_UP(nr_grant_frames, XEN_PFN_PER_PAGE);
> + pages = kcalloc(nr_pages, sizeof(pages[0]), GFP_KERNEL);
> if (!pages)
> return -ENOMEM;
>
> @@ -218,22 +221,25 @@ int __init xen_xlate_map_ballooned_pages(xen_pfn_t
> **gfns, void **virt,
> kfree(pages);
> return -ENOMEM;
> }
> - rc = alloc_xenballooned_pages(nr_grant_frames, pages);
> + rc = alloc_xenballooned_pages(nr_pages, pages);
> if (rc) {
> - pr_warn("%s Couldn't balloon alloc %ld pfns rc:%d\n",
> __func__,
> - nr_grant_frames, rc);
> + pr_warn("%s Couldn't balloon alloc %ld pages rc:%d\n",
> __func__,
> + nr_pages, rc);
> kfree(pages);
> kfree(pfns);
> return rc;
> }
> - for (i = 0; i < nr_grant_frames; i++)
> - pfns[i] = page_to_pfn(pages[i]);
> + for (i = 0; i < nr_grant_frames; i++) {
> + if ((i % XEN_PFN_PER_PAGE) == 0)
> + xen_pfn = page_to_xen_pfn(pages[i /
> XEN_PFN_PER_PAGE]);
> + pfns[i] = xen_pfn++;
> + }
We might want to:
pfns[i] = pfn_to_gfn(xen_pfn++);
for consistency, even though for autotranslate guests pfn_to_gfn always
returns pfn.
> - vaddr = vmap(pages, nr_grant_frames, 0, PAGE_KERNEL);
> + vaddr = vmap(pages, nr_pages, 0, PAGE_KERNEL);
> if (!vaddr) {
> - pr_warn("%s Couldn't map %ld pfns rc:%d\n", __func__,
> - nr_grant_frames, rc);
> - free_xenballooned_pages(nr_grant_frames, pages);
> + pr_warn("%s Couldn't map %ld pages rc:%d\n", __func__,
> + nr_pages, rc);
> + free_xenballooned_pages(nr_pages, pages);
> kfree(pages);
> kfree(pfns);
> return -ENOMEM;
>
> --
> Shannon
>
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
