fedora core 6 1.2949 + vserver 2.2.0

[linux-2.6.git] / include / asm-x86_64 / tlbflush.h

diff --git a/include/asm-x86_64/tlbflush.h b/include/asm-x86_64/tlbflush.h
index 2e811ac..983bd29 100644 (file)
--- a/include/asm-x86_64/tlbflush.h
+++ b/include/asm-x86_64/tlbflush.h
@@ -1,46 +1,47 @@
 #ifndef _X8664_TLBFLUSH_H
 #define _X8664_TLBFLUSH_H
 
-#include <linux/config.h>
 #include <linux/mm.h>
 #include <asm/processor.h>
 
-#define __flush_tlb()                                                  \
-       do {                                                            \
-               unsigned long tmpreg;                                   \
-                                                                       \
-               __asm__ __volatile__(                                   \
-                       "movq %%cr3, %0;  # flush TLB \n"               \
-                       "movq %0, %%cr3;              \n"               \
-                       : "=r" (tmpreg)                                 \
-                       :: "memory");                                   \
-       } while (0)
+static inline unsigned long get_cr3(void)
+{
+       unsigned long cr3;
+       asm volatile("mov %%cr3,%0" : "=r" (cr3));
+       return cr3;
+}
 
-/*
- * Global pages have to be flushed a bit differently. Not a real
- * performance problem because this does not happen often.
- */
-#define __flush_tlb_global()                                           \
-       do {                                                            \
-               unsigned long tmpreg;                                   \
-                                                                       \
-               __asm__ __volatile__(                                   \
-                       "movq %1, %%cr4;  # turn off PGE     \n"        \
-                       "movq %%cr3, %0;  # flush TLB        \n"        \
-                       "movq %0, %%cr3;                     \n"        \
-                       "movq %2, %%cr4;  # turn PGE back on \n"        \
-                       : "=&r" (tmpreg)                                \
-                       : "r" (mmu_cr4_features & ~X86_CR4_PGE),        \
-                         "r" (mmu_cr4_features)                        \
-                       : "memory");                                    \
-       } while (0)
-
-extern unsigned long pgkern_mask;
-
-#define __flush_tlb_all() __flush_tlb_global()
+static inline void set_cr3(unsigned long cr3)
+{
+       asm volatile("mov %0,%%cr3" :: "r" (cr3) : "memory");
+}
+
+static inline void __flush_tlb(void)
+{
+       set_cr3(get_cr3());
+}
+
+static inline unsigned long get_cr4(void)
+{
+       unsigned long cr4;
+       asm volatile("mov %%cr4,%0" : "=r" (cr4));
+       return cr4;
+}
+
+static inline void set_cr4(unsigned long cr4)
+{
+       asm volatile("mov %0,%%cr4" :: "r" (cr4) : "memory");
+}
+
+static inline void __flush_tlb_all(void)
+{
+       unsigned long cr4 = get_cr4();
+       set_cr4(cr4 & ~X86_CR4_PGE);    /* clear PGE */
+       set_cr4(cr4);                   /* write old PGE again and flush TLBs */
+}
 
 #define __flush_tlb_one(addr) \
-       __asm__ __volatile__("invlpg %0": :"m" (*(char *) addr))
+       __asm__ __volatile__("invlpg (%0)" :: "r" (addr) : "memory")
 
 
 /*
@@ -54,8 +55,9 @@ extern unsigned long pgkern_mask;
  *  - flush_tlb_kernel_range(start, end) flushes a range of kernel pages
  *  - flush_tlb_pgtables(mm, start, end) flushes a range of page tables
  *
- * ..but the x86_64 has somewhat limited tlb flushing capabilities,
- * and page-granular flushes are available only on i486 and up.
+ * x86-64 can only flush individual pages or full VMs. For a range flush
+ * we always do the full VM. Might be worth trying if for a small
+ * range a few INVLPGs in a row are a win.
  */
 
 #ifndef CONFIG_SMP
@@ -106,6 +108,10 @@ static inline void flush_tlb_range(struct vm_area_struct * vma, unsigned long st
 #define TLBSTATE_OK    1
 #define TLBSTATE_LAZY  2
 
+/* Roughly an IPI every 20MB with 4k pages for freeing page table
+   ranges. Cost is about 42k of memory for each CPU. */
+#define ARCH_FREE_PTE_NR 5350  
+
 #endif
 
 #define flush_tlb_kernel_range(start, end) flush_tlb_all()
@@ -113,7 +119,9 @@ static inline void flush_tlb_range(struct vm_area_struct * vma, unsigned long st
 static inline void flush_tlb_pgtables(struct mm_struct *mm,
                                      unsigned long start, unsigned long end)
 {
-       /* x86_64 does not keep any page table caches in TLB */
+       /* x86_64 does not keep any page table caches in a software TLB.
+          The CPUs do in their hardware TLBs, but they are handled
+          by the normal TLB flushing algorithms. */
 }
 
 #endif /* _X8664_TLBFLUSH_H */