1.1 --- a/xen/arch/i386/traps.c	Tue Apr 15 23:19:11 2003 +0000
     1.2 +++ b/xen/arch/i386/traps.c	Wed Apr 16 09:49:33 2003 +0000
     1.3 @@ -259,14 +259,14 @@ asmlinkage void do_page_fault(struct pt_
     1.4  
     1.5      __asm__ __volatile__ ("movl %%cr2,%0" : "=r" (addr) : );
     1.6  
     1.7 -    if ( unlikely(!(regs->xcs & 3)) )
     1.8 -        goto fault_in_hypervisor;
     1.9 -
    1.10      if ( unlikely(addr > PAGE_OFFSET) )
    1.11          goto fault_in_xen_space;
    1.12  
    1.13   bounce_fault:
    1.14  
    1.15 +    if ( unlikely(!(regs->xcs & 3)) )
    1.16 +        goto fault_in_hypervisor;
    1.17 +
    1.18      ti = p->thread.traps + 14;
    1.19      gtb->flags = GTBF_TRAP_CR2; /* page fault pushes %cr2 */
    1.20      gtb->cr2        = addr;
    1.21 @@ -275,7 +275,12 @@ asmlinkage void do_page_fault(struct pt_
    1.22      gtb->eip        = ti->address;
    1.23      return; 
    1.24  
    1.25 -
    1.26 +    /*
    1.27 +     * FAULT IN XEN ADDRESS SPACE:
    1.28 +     *  We only deal with one kind -- a fault in the shadow LDT mapping.
    1.29 +     *  If this occurs we pull a mapping from the guest's LDT, if it is
    1.30 +     *  valid. Otherwise we send the fault up to the guest OS to be handled.
    1.31 +     */
    1.32   fault_in_xen_space:
    1.33  
    1.34      if ( (addr < LDT_VIRT_START) || 
    1.35 @@ -316,9 +321,11 @@ asmlinkage void do_page_fault(struct pt_
    1.36          page->flags |= PGT_ldt_page;
    1.37      }
    1.38  
    1.39 +    /* Success! */
    1.40      get_page_type(page);
    1.41      get_page_tot(page);
    1.42      p->mm.perdomain_pt[l1_table_offset(off)+16] = mk_l1_pgentry(l1e|_PAGE_RW);
    1.43 +    p->mm.shadow_ldt_mapcnt++;
    1.44  
    1.45      spin_unlock(&p->page_lock);
    1.46      return;

     2.1 --- a/xen/common/memory.c	Tue Apr 15 23:19:11 2003 +0000
     2.2 +++ b/xen/common/memory.c	Wed Apr 16 09:49:33 2003 +0000
     2.3 @@ -209,7 +209,9 @@ struct list_head free_list;
     2.4  spinlock_t free_list_lock = SPIN_LOCK_UNLOCKED;
     2.5  unsigned int free_pfns;
     2.6  
     2.7 -static int tlb_flush[NR_CPUS];
     2.8 +/* Used to defer flushing of memory structures. */
     2.9 +static int flush_tlb[NR_CPUS] __cacheline_aligned;
    2.10 +
    2.11  
    2.12  /*
    2.13   * init_frametable:
    2.14 @@ -222,7 +224,7 @@ void __init init_frametable(unsigned lon
    2.15      unsigned long page_index;
    2.16      unsigned long flags;
    2.17  
    2.18 -    memset(tlb_flush, 0, sizeof(tlb_flush));
    2.19 +    memset(flush_tlb, 0, sizeof(flush_tlb));
    2.20  
    2.21      max_page = nr_pages;
    2.22      frame_table_size = nr_pages * sizeof(struct pfn_info);
    2.23 @@ -247,6 +249,34 @@ void __init init_frametable(unsigned lon
    2.24  }
    2.25  
    2.26  
    2.27 +static void __invalidate_shadow_ldt(void)
    2.28 +{
    2.29 +    int i;
    2.30 +    unsigned long pfn;
    2.31 +    struct pfn_info *page;
    2.32 +    
    2.33 +    current->mm.shadow_ldt_mapcnt = 0;
    2.34 +
    2.35 +    for ( i = 16; i < 32; i++ )
    2.36 +    {
    2.37 +        pfn = l1_pgentry_to_pagenr(current->mm.perdomain_pt[i]);
    2.38 +        if ( pfn == 0 ) continue;
    2.39 +        current->mm.perdomain_pt[i] = mk_l1_pgentry(0);
    2.40 +        page = frame_table + pfn;
    2.41 +        ASSERT((page->flags & PG_type_mask) == PGT_ldt_page);
    2.42 +        ASSERT((page->flags & PG_domain_mask) == current->domain);
    2.43 +        ASSERT((page->type_count != 0) && (page->tot_count != 0));
    2.44 +        put_page_type(page);
    2.45 +        put_page_tot(page);                
    2.46 +    }
    2.47 +}
    2.48 +static inline void invalidate_shadow_ldt(void)
    2.49 +{
    2.50 +    if ( current->mm.shadow_ldt_mapcnt != 0 )
    2.51 +        __invalidate_shadow_ldt();
    2.52 +}
    2.53 +
    2.54 +
    2.55  /* Return original refcnt, or -1 on error. */
    2.56  static int inc_page_refcnt(unsigned long page_nr, unsigned int type)
    2.57  {
    2.58 @@ -283,6 +313,7 @@ static int inc_page_refcnt(unsigned long
    2.59      return get_page_type(page);
    2.60  }
    2.61  
    2.62 +
    2.63  /* Return new refcnt, or -1 on error. */
    2.64  static int dec_page_refcnt(unsigned long page_nr, unsigned int type)
    2.65  {
    2.66 @@ -373,6 +404,7 @@ static int get_l2_table(unsigned long pa
    2.67      return ret;
    2.68  }
    2.69  
    2.70 +
    2.71  static int get_l1_table(unsigned long page_nr)
    2.72  {
    2.73      l1_pgentry_t *p_l1_entry, l1_entry;
    2.74 @@ -408,6 +440,7 @@ static int get_l1_table(unsigned long pa
    2.75      return ret;
    2.76  }
    2.77  
    2.78 +
    2.79  static int get_page(unsigned long page_nr, int writeable)
    2.80  {
    2.81      struct pfn_info *page;
    2.82 @@ -450,6 +483,7 @@ static int get_page(unsigned long page_n
    2.83      return(0);
    2.84  }
    2.85  
    2.86 +
    2.87  static void put_l2_table(unsigned long page_nr)
    2.88  {
    2.89      l2_pgentry_t *p_l2_entry, l2_entry;
    2.90 @@ -469,6 +503,7 @@ static void put_l2_table(unsigned long p
    2.91      unmap_domain_mem(p_l2_entry);
    2.92  }
    2.93  
    2.94 +
    2.95  static void put_l1_table(unsigned long page_nr)
    2.96  {
    2.97      l1_pgentry_t *p_l1_entry, l1_entry;
    2.98 @@ -492,6 +527,7 @@ static void put_l1_table(unsigned long p
    2.99      unmap_domain_mem(p_l1_entry-1);
   2.100  }
   2.101  
   2.102 +
   2.103  static void put_page(unsigned long page_nr, int writeable)
   2.104  {
   2.105      struct pfn_info *page;
   2.106 @@ -502,10 +538,19 @@ static void put_page(unsigned long page_
   2.107             ((page_type_count(page) != 0) && 
   2.108              ((page->flags & PG_type_mask) == PGT_writeable_page) &&
   2.109              ((page->flags & PG_need_flush) == PG_need_flush)));
   2.110 -    if ( writeable && (put_page_type(page) == 0) )
   2.111 +    if ( writeable )
   2.112      {
   2.113 -        tlb_flush[smp_processor_id()] = 1;
   2.114 -        page->flags &= ~PG_need_flush;
   2.115 +        if ( put_page_type(page) == 0 )
   2.116 +        {
   2.117 +            flush_tlb[smp_processor_id()] = 1;
   2.118 +            page->flags &= ~PG_need_flush;
   2.119 +        }
   2.120 +    }
   2.121 +    else if ( unlikely(((page->flags & PG_type_mask) == PGT_ldt_page) &&
   2.122 +                       (page_type_count(page) != 0)) )
   2.123 +    {
   2.124 +        /* We expect this is rare so we just blow the entire shadow LDT. */
   2.125 +        invalidate_shadow_ldt();
   2.126      }
   2.127      put_page_tot(page);
   2.128  }
   2.129 @@ -685,15 +730,17 @@ static int do_extended_command(unsigned 
   2.130          {
   2.131              put_l2_table(pagetable_val(current->mm.pagetable) >> PAGE_SHIFT);
   2.132              current->mm.pagetable = mk_pagetable(pfn << PAGE_SHIFT);
   2.133 +            invalidate_shadow_ldt();
   2.134 +            flush_tlb[smp_processor_id()] = 1;
   2.135          }
   2.136          else
   2.137          {
   2.138              MEM_LOG("Error while installing new baseptr %08lx %d", ptr, err);
   2.139          }
   2.140 -        /* fall through */
   2.141 +        break;
   2.142          
   2.143      case PGEXT_TLB_FLUSH:
   2.144 -        tlb_flush[smp_processor_id()] = 1;
   2.145 +        flush_tlb[smp_processor_id()] = 1;
   2.146          break;
   2.147      
   2.148      case PGEXT_INVLPG:
   2.149 @@ -702,7 +749,6 @@ static int do_extended_command(unsigned 
   2.150  
   2.151      case PGEXT_SET_LDT:
   2.152      {
   2.153 -        int i;
   2.154          unsigned long ents = val >> PGEXT_CMD_SHIFT;
   2.155          if ( ((ptr & (PAGE_SIZE-1)) != 0) || 
   2.156               (ents > 8192) ||
   2.157 @@ -717,20 +763,8 @@ static int do_extended_command(unsigned 
   2.158          {
   2.159              if ( current->mm.ldt_ents != 0 )
   2.160              {
   2.161 -                /* Tear down the old LDT. */
   2.162 -                for ( i = 16; i < 32; i++ )
   2.163 -                {
   2.164 -                    pfn = l1_pgentry_to_pagenr(current->mm.perdomain_pt[i]);
   2.165 -                    if ( pfn == 0 ) continue;
   2.166 -                    current->mm.perdomain_pt[i] = mk_l1_pgentry(0);
   2.167 -                    page = frame_table + pfn;
   2.168 -                    ASSERT((page->flags & PG_type_mask) == PGT_ldt_page);
   2.169 -                    ASSERT((page->flags & PG_domain_mask) == current->domain);
   2.170 -                    ASSERT((page->type_count != 0) && (page->tot_count != 0));
   2.171 -                    put_page_type(page);
   2.172 -                    put_page_tot(page);                
   2.173 -                }
   2.174 -                tlb_flush[smp_processor_id()] = 1;
   2.175 +                invalidate_shadow_ldt();
   2.176 +                flush_tlb[smp_processor_id()] = 1;
   2.177              }
   2.178              current->mm.ldt_base = ptr;
   2.179              current->mm.ldt_ents = ents;
   2.180 @@ -748,6 +782,7 @@ static int do_extended_command(unsigned 
   2.181      return err;
   2.182  }
   2.183  
   2.184 +
   2.185  int do_process_page_updates(page_update_request_t *ureqs, int count)
   2.186  {
   2.187      page_update_request_t req;
   2.188 @@ -860,9 +895,9 @@ int do_process_page_updates(page_update_
   2.189          ureqs++;
   2.190      }
   2.191  
   2.192 -    if ( tlb_flush[smp_processor_id()] )
   2.193 +    if ( flush_tlb[smp_processor_id()] )
   2.194      {
   2.195 -        tlb_flush[smp_processor_id()] = 0;
   2.196 +        flush_tlb[smp_processor_id()] = 0;
   2.197          __write_cr3_counted(pagetable_val(current->mm.pagetable));
   2.198  
   2.199      }

     3.1 --- a/xen/include/xeno/sched.h	Tue Apr 15 23:19:11 2003 +0000
     3.2 +++ b/xen/include/xeno/sched.h	Wed Apr 16 09:49:33 2003 +0000
     3.3 @@ -30,7 +30,7 @@ struct mm_struct {
     3.4      l1_pgentry_t *perdomain_pt;
     3.5      pagetable_t  pagetable;
     3.6      /* Current LDT details. */
     3.7 -    unsigned long ldt_base, ldt_ents;
     3.8 +    unsigned long ldt_base, ldt_ents, shadow_ldt_mapcnt;
     3.9      /* Next entry is passed to LGDT on domain switch. */
    3.10      char gdt[6];
    3.11  };