question about page_to_phys

question about page_to_phys

[Xiaofeng Ling]

The page_to_phys is defined as
#define page_to_phys(page)   (phys_to_machine(page_to_pseudophys(page)))
so it return machine addresss
while virt_to_phys return psedophys. include/asm-xen/asm-i386/io.h
this is really confusing.
why not define page_to_machine?

Re: question about page_to_phys

[Keir Fraser]

On 1 Sep 2005, at 02:29, Xiaofeng Ling wrote:

> The page_to_phys is defined as
> #define page_to_phys(page)   
> (phys_to_machine(page_to_pseudophys(page)))
> so it return machine addresss
> while virt_to_phys return psedophys. include/asm-xen/asm-i386/io.h
> this is really confusing.
> why not define page_to_machine?

page_to_phys() is used only in places where we want to compare machine 
addresses (e.g., in deciding when to merge I/O requests). It was easier 
to abuse the macro definition than fix every place that it is used.

  -- Keir

Re: question about page_to_phys

[Xiaofeng Ling]

Keir Fraser wrote:
> 
> On 1 Sep 2005, at 10:19, Xiaofeng Ling wrote:
> 
> If a guest is in translate mode it should have no concept of pseudophys
> vs machine. That translation is hidden from it. Hence there should be
> no need to hack the virtual drivers -- page_to_phys() should naturally
> translate to pseudophys if SHADOW_TRANSLATE is configured.
> 
>   -- Keir
> 
so we need such a patch? but how to deal with virt_to_mfn in blkfront.c?

diff -r 9092486cc4b3 linux-2.6-xen-sparse/include/asm-xen/asm-i386/io.h
--- a/linux-2.6-xen-sparse/include/asm-xen/asm-i386/io.h
+++ b/linux-2.6-xen-sparse/include/asm-xen/asm-i386/io.h
@@ -101,7 +101,11 @@
   * Change "struct page" to physical address.
   */
  #define page_to_pseudophys(page) ((dma_addr_t)page_to_pfn(page) << 
PAGE_SHIFT)
+#ifndef CONFIG_SHADOW_TRANSLATE
  #define page_to_phys(page) 
(phys_to_machine(page_to_pseudophys(page)))
+#else
+#define page_to_phys(page)      (page_to_pseudophys(page))
+#endif

  #define bio_to_pseudophys(bio)  (page_to_pseudophys(bio_page((bio))) + \
                                   (unsigned long) bio_offset((bio)))

Re: question about page_to_phys

[Keir Fraser]

On 1 Sep 2005, at 10:56, Xiaofeng Ling wrote:

> so we need such a patch? but how to deal with virt_to_mfn in 
> blkfront.c?

I guess the phys<->mach macros should all be no-ops.

  -- Keir