Xen devel - Aug 2005 - [PATCH] [x86_64] Fixing PGT_va_mutable and PGT_va_unknown

This patch ensures that type_info is always greater than PGT_va_unknown
if PGT_va_mutable is set.

Today, some (rare) cases (#PF in the kernel address space, such vmalloc
handling) are unnecessarily sent to the emulation code.

int ptwr_do_page_fault(struct domain *d, unsigned long addr, 
                       struct cpu_user_regs *regs)
{
    unsigned long    pfn;
...
    if ( unlikely(l2_idx >= PGT_va_unknown) )
        goto emulate; /* Urk! This L1 is mapped in multiple L2 slots! */

Signed-off-by: Jun Nakajima <jun.nakajima@intel.com>

diff -r f2e241ecf1c3 -r f89a34a65e7d xen/include/asm-x86/mm.h
--- a/xen/include/asm-x86/mm.h  Mon Aug 15 01:10:47 2005
+++ b/xen/include/asm-x86/mm.h  Mon Aug 15 03:31:27 2005
@@ -88,14 +88,11 @@
 #elif defined(__x86_64__)
  /* The 27 most significant bits of virt address if this is a page
table. */
 #define PGT_va_shift        32
-#define PGT_va_mask         ((unsigned
long)((1U<<28)-1)<<PGT_va_shift)
+#define PGT_va_mask         ((unsigned
long)((1U<<29)-1)<<PGT_va_shift)
  /* Is the back pointer still mutable (i.e. not fixed yet)? */
- /* Use PML4 slot for HYPERVISOR_VIRT_START.  
-    18 = L4_PAGETABLE_SHIFT - L2_PAGETABLE_SHIFT */
-#define PGT_va_mutable      ((unsigned
long)(256U<<18)<<PGT_va_shift)
+#define PGT_va_mutable      ((unsigned long)(1U<<28)<<PGT_va_shift)
  /* Is the back pointer unknown (e.g., p.t. is mapped at multiple VAs)?
*/
- /* Use PML4 slot for HYPERVISOR_VIRT_START + 1 */
-#define PGT_va_unknown      ((unsigned
long)(257U<<18)<<PGT_va_shift)
+#define PGT_va_unknown      ((unsigned
long)((1U<<28)-1)<<PGT_va_shift)
 #endif
 
  /* 16-bit count of uses of this frame as its current type. */

Jun
---
Intel Open Source Technology Center 


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 15 Aug 2005, at 08:11, Nakajima, Jun wrote:
> This patch ensures that type_info is always greater than PGT_va_unknown
> if PGT_va_mutable is set.
Okay, but I see no reason to change PGT_va_mask. You already include 28 
bits where you actually need only 27 bits of backptr.

  -- Keir


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

On 15 Aug 2005, at 08:50, Nakajima, Jun wrote:
>> Okay, but I see no reason to change PGT_va_mask. You already include
>> 28 bits where you actually need only 27 bits of backptr.
>
> The current one
> #define PGT_va_mask         ((unsigned
long)((1U<<28)-1)<<PGT_va_shift)
> includes only 27 bits (see -1).
(1<<n)-1 sets the n least significant bits, not n-1.

  -- Keir


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Keir Fraser wrote:
> On 15 Aug 2005, at 08:11, Nakajima, Jun wrote:
> 
>> This patch ensures that type_info is always greater than
>> PGT_va_unknown if PGT_va_mutable is set.
> 
> Okay, but I see no reason to change PGT_va_mask. You already include
> 28 bits where you actually need only 27 bits of backptr.
The current one
#define PGT_va_mask         ((unsigned
long)((1U<<28)-1)<<PGT_va_shift)
includes only 27 bits (see -1). 
> 
>   -- Keir

Jun
---
Intel Open Source Technology Center

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Keir Fraser wrote:
> On 15 Aug 2005, at 08:50, Nakajima, Jun wrote:
> 
>>> Okay, but I see no reason to change PGT_va_mask. You already
include
>>> 28 bits where you actually need only 27 bits of backptr.
>> 
>> The current one
>> #define PGT_va_mask         ((unsigned
>> long)((1U<<28)-1)<<PGT_va_shift) includes only 27 bits (see
-1).
> 
> (1<<n)-1 sets the n least significant bits, not n-1.
You are right. I need to go to bed now. 
> 
>   -- Keir
Jun
---
Intel Open Source Technology Center

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel