[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Xen-devel] high memory manipulation patch



This patch adds support for high memory to /proc/xen/memory_target,
and it restores the mem= linux boot parameter.  This is for 2.4.26 only.
I haven't looked at 2.6 yet.  (remmeber that CONFIG_HIGHMEM
is not on by default in the xen0 and xenU configs)
I have racks of 2g and 4g machines, hence my interest.

The 'mem=' param tells linux the maximum amount of memory it can use.
The feature originated for hosts where the normal mem detection mechanisms
failed.   These days its main use in stock linux is to artificially
constrain how much memory linux uses for testing low memory systems.

With this patch, mem= now means the max memory linux could ever use.
When a domain is created with less than that the mem= value, linux will
behave as though that 'missing' memory is 'allocated'.   To give the
domain more memory, first run setdomainmaxmem in Dom-0 to raise Xen's
limit, then write the new total to /proc/xen/memory_target in the
domain.   When mem= is not explicitly set, it defaults to the 
boottime size of the domain.

This is useful in order to create domains at times when the memory resources
the new domain could use in the future are currently in use by other domains.

Hopefully an xm god can replace the setdomainmaxmem script with a new
maxmem argument to xm create.

This bit illustrates booting dom-1 with 100m and growing it to 2g:

     dom-0# xm create name=dom-1 memory=100 extra='-b mem=2g'
     dom-0# setdomainmaxmem 1 2g

then in domain 1,
     dom-1# mount /proc
     dom-1# grep Mem[TF] /proc/meminfo
     MemTotal:      2097152 kB
     MemFree:         57352 kB
     dom-1# echo 2g > /proc/xen/memory_target 
     Granted 1948MB new mem by xen. Domain now has 2048MB
     dom-1# grep Mem[TF] /proc/meminfo
     MemTotal:      2097152 kB
     MemFree:       2052104 kB

David



===== linux-2.4.26-xen-sparse/arch/xen/drivers/balloon/balloon.c 1.18 vs edited 
=====
*** /tmp/balloon.c-1.18-11013   Mon Aug  2 10:57:36 2004
--- edited/linux-2.4.26-xen-sparse/arch/xen/drivers/balloon/balloon.c   Fri Aug 
 6 11:13:24 2004
***************
*** 17,22 ****
--- 17,24 ----
  #include <linux/mman.h>
  #include <linux/smp_lock.h>
  #include <linux/pagemap.h>
+ #include <linux/bootmem.h>
+ #include <linux/highmem.h>
  #include <linux/vmalloc.h>
  
  #include <asm/hypervisor.h>
***************
*** 39,45 ****
  
  static struct proc_dir_entry *balloon_pde;
  unsigned long credit;
! static unsigned long current_pages, max_pages;
  
  static inline pte_t *get_ptep(unsigned long addr)
  {
--- 41,47 ----
  
  static struct proc_dir_entry *balloon_pde;
  unsigned long credit;
! static unsigned long current_pages, most_seen_pages;
  
  static inline pte_t *get_ptep(unsigned long addr)
  {
***************
*** 69,109 ****
      parray = (unsigned long *)vmalloc(num_pages * sizeof(unsigned long));
      if ( parray == NULL )
      {
!         printk("inflate_balloon: Unable to vmalloc parray\n");
!         return 0;
      }
  
      currp = parray;
  
!     for ( i = 0; i < num_pages; i++ )
      {
!         /* NB. Should be GFP_ATOMIC for a less aggressive inflation. */
!         vaddr = __get_free_page(GFP_KERNEL);
  
          /* If allocation fails then free all reserved pages. */
!         if ( vaddr == 0 )
          {
!             printk("Unable to inflate balloon by %ld, only %ld pages free.",
                     num_pages, i);
              currp = parray;
!             for(j = 0; j < i; j++){
!                 free_page(*currp++);
              }
              goto cleanup;
          }
  
!         *currp++ = vaddr;
      }
  
  
!     currp = parray;
!     for ( i = 0; i < num_pages; i++ )
      {
!         curraddr = *currp;
!         *currp = virt_to_machine(*currp) >> PAGE_SHIFT;
!         queue_l1_entry_update(get_ptep(curraddr), 0);
!         phys_to_machine_mapping[__pa(curraddr) >> PAGE_SHIFT] = DEAD;
!         currp++;
      }
  
      XEN_flush_page_update_queue();
--- 71,113 ----
      parray = (unsigned long *)vmalloc(num_pages * sizeof(unsigned long));
      if ( parray == NULL )
      {
!         printk(KERN_ERR "inflate_balloon: Unable to vmalloc parray\n");
!         return -EFAULT;
      }
  
      currp = parray;
  
!     for ( i = 0; i < num_pages; i++, currp++ )
      {
!       struct page *page = alloc_page(GFP_HIGHUSER);
!       unsigned long pfn =  page - mem_map;
  
          /* If allocation fails then free all reserved pages. */
!         if ( page == 0 )
          {
!             printk(KERN_ERR "Unable to inflate balloon by %ld, only %ld pages 
free.",
                     num_pages, i);
              currp = parray;
!             for(j = 0; j < i; j++, ++currp){
!                 __free_page((struct page *) (mem_map + *currp));
              }
+           ret = -EFAULT;
              goto cleanup;
          }
  
!         *currp = pfn;
      }
  
  
!     for ( i = 0, currp = parray; i < num_pages; i++, currp++ )
      {
!       unsigned long mfn = phys_to_machine_mapping[*currp];
!         curraddr = page_address(mem_map + *currp);
!       if (curraddr)
!             queue_l1_entry_update(get_ptep(curraddr), 0);
! 
!         phys_to_machine_mapping[*currp] = DEAD;
!         *currp = mfn;
      }
  
      XEN_flush_page_update_queue();
***************
*** 112,118 ****
                                  parray, num_pages, 0);
      if ( unlikely(ret != num_pages) )
      {
!         printk("Unable to inflate balloon, error %lx\n", ret);
          goto cleanup;
      }
  
--- 116,122 ----
                                  parray, num_pages, 0);
      if ( unlikely(ret != num_pages) )
      {
!         printk(KERN_ERR "Unable to inflate balloon, error %lx\n", ret);
          goto cleanup;
      }
  
***************
*** 130,136 ****
   * phys->machine mapping table looking for DEAD entries and populates
   * them.
   */
! static unsigned long process_new_pages(unsigned long * parray, 
                                         unsigned long num)
  {
      /* currently, this function is rather simplistic as 
--- 134,140 ----
   * phys->machine mapping table looking for DEAD entries and populates
   * them.
   */
! static unsigned long process_returned_pages(unsigned long * parray, 
                                         unsigned long num)
  {
      /* currently, this function is rather simplistic as 
***************
*** 140,146 ****
       * incorporated here.
       */
       
!     unsigned long tot_pages = start_info.nr_pages;   
      unsigned long * curr = parray;
      unsigned long num_installed;
      unsigned long i;
--- 144,150 ----
       * incorporated here.
       */
       
!     unsigned long tot_pages = most_seen_pages;   
      unsigned long * curr = parray;
      unsigned long num_installed;
      unsigned long i;
***************
*** 152,180 ****
          {
              phys_to_machine_mapping[i] = *curr;
              queue_machphys_update(*curr, i);
!             queue_l1_entry_update(
                  get_ptep((unsigned long)__va(i << PAGE_SHIFT)),
                  ((*curr) << PAGE_SHIFT) | pgprot_val(PAGE_KERNEL));
  
!             *curr = (unsigned long)__va(i << PAGE_SHIFT);
              curr++;
              num_installed++;
          }
      }
  
-     /*
-      * This is tricky (and will also change for machine addrs that 
-      * are mapped to not previously released addresses). We free pages
-      * that were allocated by get_free_page (the mappings are different 
-      * now, of course).
-      */
-     curr = parray;
-     for ( i = 0; i < num_installed; i++ )
-     {
-         free_page(*curr);
-         curr++;
-     }
- 
      return num_installed;
  }
  
--- 156,173 ----
          {
              phys_to_machine_mapping[i] = *curr;
              queue_machphys_update(*curr, i);
!           if (i<max_low_pfn)
!               queue_l1_entry_update(
                  get_ptep((unsigned long)__va(i << PAGE_SHIFT)),
                  ((*curr) << PAGE_SHIFT) | pgprot_val(PAGE_KERNEL));
  
!             __free_page(mem_map + i);
! 
              curr++;
              num_installed++;
          }
      }
  
      return num_installed;
  }
  
***************
*** 185,198 ****
  
      if ( num_pages > credit )
      {
!         printk("Can not allocate more pages than previously released.\n");
          return -EAGAIN;
      }
  
      parray = (unsigned long *)vmalloc(num_pages * sizeof(unsigned long));
      if ( parray == NULL )
      {
!         printk("inflate_balloon: Unable to vmalloc parray\n");
          return 0;
      }
  
--- 178,192 ----
  
      if ( num_pages > credit )
      {
!         printk(KERN_ERR "deflate_balloon: %d pages > %d credit.\n",
!                       num_pages, credit);
          return -EAGAIN;
      }
  
      parray = (unsigned long *)vmalloc(num_pages * sizeof(unsigned long));
      if ( parray == NULL )
      {
!         printk(KERN_ERR "deflate_balloon: Unable to vmalloc parray\n");
          return 0;
      }
  
***************
*** 202,215 ****
                                  parray, num_pages, 0);
      if ( unlikely(ret != num_pages) )
      {
!         printk("Unable to deflate balloon, error %lx\n", ret);
          goto cleanup;
      }
  
!     if ( (ret = process_new_pages(parray, num_pages)) < num_pages )
      {
!         printk("Unable to deflate balloon by specified %lx pages, only 
%lx.\n",
!                num_pages, ret);
          goto cleanup;
      }
  
--- 196,211 ----
                                  parray, num_pages, 0);
      if ( unlikely(ret != num_pages) )
      {
!         printk(KERN_ERR "deflate_balloon: xen increase_reservation err %lx\n",
!                       ret);
          goto cleanup;
      }
  
!     if ( (ret = process_returned_pages(parray, num_pages)) < num_pages )
      {
!         printk(KERN_WARNING
!          "deflate_balloon: restored only %lx of %lx pages.\n",
!            ret, num_pages);
          goto cleanup;
      }
  
***************
*** 224,243 ****
  
  #define PAGE_TO_MB_SHIFT 8
  
  static int balloon_write(struct file *file, const char *buffer,
                           u_long count, void *data)
  {
      char memstring[64], *endchar;
      int len, i;
!     unsigned long pages;
!     unsigned long long target;
  
      /* Only admin can play with the balloon :) */
      if ( !capable(CAP_SYS_ADMIN) )
          return -EPERM;
  
      if (count>sizeof memstring) {
!           return -EFBIG;
      }
  
      len = strnlen_user(buffer, count);
--- 220,389 ----
  
  #define PAGE_TO_MB_SHIFT 8
  
+ /*
+  * pagetable_extend() mimics pagetable_init() from arch/xen/mm/init.c 
+  * The loops do go through all of low memory (ZONE_NORMAL).  The
+  * old pages have _PAGE_PRESENT set and so get skipped.
+  * If low memory is not full, the new pages are used to fill it, going
+  * from cur_low_pfn to low_pfn.   high memory is not direct mapped so
+  * no extension is needed for new high memory.
+  */
+ 
+ static void pagetable_extend (int cur_low_pfn, int newpages)
+ {
+     unsigned long vaddr, end;
+     pgd_t *kpgd, *pgd, *pgd_base;
+     int i, j, k;
+     pmd_t *kpmd, *pmd;
+     pte_t *kpte, *pte, *pte_base;
+     int low_pfn = min(cur_low_pfn+newpages,(int)max_low_pfn);
+ 
+     /*
+      * This can be zero as well - no problem, in that case we exit
+      * the loops anyway due to the PTRS_PER_* conditions.
+      */
+     end = (unsigned long)__va(low_pfn*PAGE_SIZE);
+ 
+     pgd_base = init_mm.pgd;
+     i = __pgd_offset(PAGE_OFFSET);
+     pgd = pgd_base + i;
+ 
+     for (; i < PTRS_PER_PGD; pgd++, i++) {
+         vaddr = i*PGDIR_SIZE;
+         if (end && (vaddr >= end))
+             break;
+         pmd = (pmd_t *)pgd;
+         for (j = 0; j < PTRS_PER_PMD; pmd++, j++) {
+             vaddr = i*PGDIR_SIZE + j*PMD_SIZE;
+             if (end && (vaddr >= end))
+                 break;
+ 
+             /* Filled in for us already? */
+             if ( pmd_val(*pmd) & _PAGE_PRESENT )
+                 continue;
+ 
+             pte_base = pte = (pte_t *) __get_free_page(GFP_KERNEL);
+ 
+             for (k = 0; k < PTRS_PER_PTE; pte++, k++) {
+                 vaddr = i*PGDIR_SIZE + j*PMD_SIZE + k*PAGE_SIZE;
+                 if (end && (vaddr >= end))
+                     break;
+                 *pte = mk_pte_phys(__pa(vaddr), PAGE_KERNEL);
+             }
+             kpgd = pgd_offset_k((unsigned long)pte_base);
+             kpmd = pmd_offset(kpgd, (unsigned long)pte_base);
+             kpte = pte_offset(kpmd, (unsigned long)pte_base);
+             queue_l1_entry_update(kpte,
+                                   (*(unsigned long *)kpte)&~_PAGE_RW);
+             set_pmd(pmd, __pmd(_KERNPG_TABLE + __pa(pte_base)));
+             XEN_flush_page_update_queue();
+         }
+     }
+ }
+ 
+ /*
+  * claim_new_pages() asks xen to increase this domain's memory  reservation
+  * and return a list of the new pages of memory.  This new pages are
+  * added to the free list of the memory manager.
+  *
+  * Available RAM does not normally change while Linux runs.  To make this 
work,
+  * the linux mem= boottime command line param must say how big memory could
+  * possibly grow.  Then setup_arch() in arch/xen/kernel/setup.c
+  * sets max_pfn, max_low_pfn and the zones according to
+  * this max memory size.   The page tables themselves can only be
+  * extended after xen has assigned new pages to this domain.
+  */
+ 
+ static unsigned long
+ claim_new_pages(unsigned long num_pages)
+ {
+     unsigned long new_page_cnt, pfn;
+     unsigned long * parray, *curr;
+ 
+     if (most_seen_pages+num_pages> max_pfn)
+         num_pages = max_pfn-most_seen_pages;
+     if (num_pages==0) return 0;
+ 
+     parray = (unsigned long *)vmalloc(num_pages * sizeof(unsigned long));
+     if ( parray == NULL )
+     {
+         printk(KERN_ERR "claim_new_pages: Unable to vmalloc parray\n");
+         return 0;
+     }
+ 
+     XEN_flush_page_update_queue();
+     new_page_cnt = HYPERVISOR_dom_mem_op(MEMOP_increase_reservation, 
+                                 parray, num_pages, 0);
+     if (new_page_cnt != num_pages)
+     {
+         printk(KERN_WARNING
+             "claim_new_pages: xen granted only %lu of %lu requested pages\n",
+             new_page_cnt, num_pages);
+ 
+       /* XXX
+        * avoid xen lockup when user forgot to setdomainmaxmem.  xen
+        * usually can dribble out a few pages and then hangs
+        */
+       if (new_page_cnt < 1000) {
+             printk(KERN_WARNING "Remember to use setdomainmaxmem\n");
+           HYPERVISOR_dom_mem_op(MEMOP_decrease_reservation, 
+                                 parray, new_page_cnt, 0);
+             return -EFAULT;
+       }
+     }
+     memcpy(phys_to_machine_mapping+most_seen_pages, parray,
+             new_page_cnt * sizeof(unsigned long));
+ 
+     pagetable_extend(most_seen_pages,new_page_cnt);
+ 
+     for (pfn = most_seen_pages, curr = parray;
+           pfn < most_seen_pages+new_page_cnt;
+             pfn++, curr++ )
+     {
+         struct page *page = mem_map + pfn;
+ 
+ #ifndef CONFIG_HIGHMEM
+       if (pfn>=max_low_pfn) {
+             printk(KERN_WARNING "Warning only %ldMB will be used.\n",
+                pfn>>PAGE_TO_MB_SHIFT);
+             printk(KERN_WARNING "Use a HIGHMEM enabled kernel.\n");
+           break;
+       }
+ #endif
+       queue_machphys_update(*curr, pfn);
+       XEN_flush_page_update_queue();
+       if (pfn<max_low_pfn)  {
+               queue_l1_entry_update(get_ptep((unsigned long)__va(pfn << 
PAGE_SHIFT)),
+                       ((*curr) << PAGE_SHIFT) | pgprot_val(PAGE_KERNEL));
+               XEN_flush_page_update_queue();
+               }
+ 
+         /* this next bit mimics arch/xen/mm/init.c:one_highpage_init() */
+         ClearPageReserved(page);
+         if (pfn>=max_low_pfn) set_bit(PG_highmem, &page->flags);
+         set_page_count(page, 1);
+         __free_page(page);
+     }
+ 
+     vfree(parray);
+ 
+     return new_page_cnt;
+ }
+ 
  static int balloon_write(struct file *file, const char *buffer,
                           u_long count, void *data)
  {
      char memstring[64], *endchar;
      int len, i;
!     unsigned long target;
!     unsigned long long targetbytes;
  
      /* Only admin can play with the balloon :) */
      if ( !capable(CAP_SYS_ADMIN) )
          return -EPERM;
  
      if (count>sizeof memstring) {
!         return -EFBIG;
      }
  
      len = strnlen_user(buffer, count);
***************
*** 248,300 ****
  
      endchar = memstring;
      for(i=0; i<len; ++i,++endchar) {
!           if ('0'>memstring[i] || memstring[i]>'9') break;
      }
      if (i==0) return -EBADMSG;
  
!     target = memparse(memstring,&endchar);
!     pages = target >> PAGE_SHIFT;
  
!     if (pages < current_pages) {
!           int change = inflate_balloon(current_pages-pages);
!           if (change<0) return change;
! 
!           current_pages -= change;
!           printk("Relinquish %dMB to xen. Domain now has %ldMB\n",
!                   change>>PAGE_TO_MB_SHIFT, current_pages>>PAGE_TO_MB_SHIFT);
!     }
!     else if (pages > current_pages) {
!           int change = deflate_balloon(min(pages,max_pages) - current_pages);
!           if (change<0) return change;
! 
!           current_pages += change;
!           printk("Reclaim %dMB from xen. Domain now has %ldMB\n",
!                   change>>PAGE_TO_MB_SHIFT, current_pages>>PAGE_TO_MB_SHIFT);
      }
  
      return len;
  }
  
  
  static int balloon_read(char *page, char **start, off_t off,
!         int count, int *eof, void *data)
  {
!       int len;
!       len = sprintf(page,"%lu\n",current_pages<<PAGE_SHIFT);
  
!       if (len <= off+count) *eof = 1;
!       *start = page + off;
!       len -= off;
!       if (len>count) len = count;
!       if (len<0) len = 0;
!       return len;
  }
  
  static int __init init_module(void)
  {
      printk(KERN_ALERT "Starting Xen Balloon driver\n");
  
!     max_pages = current_pages = start_info.nr_pages;
      if ( (balloon_pde = create_xen_proc_entry("memory_target", 0644)) == NULL 
)
      {
          printk(KERN_ALERT "Unable to create balloon driver proc entry!");
--- 394,459 ----
  
      endchar = memstring;
      for(i=0; i<len; ++i,++endchar) {
!         if ('0'>memstring[i] || memstring[i]>'9') break;
      }
      if (i==0) return -EBADMSG;
  
!     targetbytes = memparse(memstring,&endchar);
!     target = targetbytes >> PAGE_SHIFT;
! 
!     if (target < current_pages) {
!         int change = inflate_balloon(current_pages-target);
!         if (change<=0) return change;
! 
!         current_pages -= change;
!         printk(KERN_INFO "Relinquish %dMB to xen. Domain now has %luMB\n",
!             change>>PAGE_TO_MB_SHIFT, current_pages>>PAGE_TO_MB_SHIFT);
!     }
!     else if (target > current_pages) {
!         int change, reclaim = min(target,most_seen_pages) - current_pages;
! 
!         if (reclaim) {
!             change = deflate_balloon( reclaim);
!             if (change<=0) return change;
!             current_pages += change;
!             printk(KERN_INFO "Reclaim %dMB from xen. Domain now has %luMB\n",
!                 change>>PAGE_TO_MB_SHIFT, current_pages>>PAGE_TO_MB_SHIFT);
!         }
  
!         if (most_seen_pages<target) {
!             int growth = claim_new_pages(target-most_seen_pages);
!           if (growth<=0) return growth;
!             most_seen_pages += growth;
!             current_pages += growth;
!             printk(KERN_INFO "Granted %dMB new mem by xen. Domain now has 
%luMB\n",
!                 growth>>PAGE_TO_MB_SHIFT, current_pages>>PAGE_TO_MB_SHIFT);
!         }
      }
  
+ 
      return len;
  }
  
  
  static int balloon_read(char *page, char **start, off_t off,
!       int count, int *eof, void *data)
  {
!     int len;
!     len = sprintf(page,"%lu\n",current_pages<<PAGE_SHIFT);
  
!     if (len <= off+count) *eof = 1;
!     *start = page + off;
!     len -= off;
!     if (len>count) len = count;
!     if (len<0) len = 0;
!     return len;
  }
  
  static int __init init_module(void)
  {
      printk(KERN_ALERT "Starting Xen Balloon driver\n");
  
!     most_seen_pages = current_pages = min(start_info.nr_pages,max_pfn);
      if ( (balloon_pde = create_xen_proc_entry("memory_target", 0644)) == NULL 
)
      {
          printk(KERN_ALERT "Unable to create balloon driver proc entry!");
***************
*** 303,308 ****
--- 462,478 ----
  
      balloon_pde->write_proc = balloon_write;
      balloon_pde->read_proc = balloon_read;
+ 
+     /* 
+      * make a new phys map if mem= says xen can give us memory  to grow
+      */
+     if (max_pfn > start_info.nr_pages) {
+         extern unsigned long *phys_to_machine_mapping;
+         unsigned long *newmap;
+         newmap = (unsigned long *)vmalloc(max_pfn * sizeof(unsigned long));
+         phys_to_machine_mapping = memcpy(newmap, phys_to_machine_mapping,
+             start_info.nr_pages * sizeof(unsigned long));
+     }
  
      return 0;
  }
===== linux-2.4.26-xen-sparse/arch/xen/kernel/setup.c 1.51 vs edited =====
*** /tmp/setup.c-1.51-11013     Wed Jul 14 10:26:02 2004
--- edited/linux-2.4.26-xen-sparse/arch/xen/kernel/setup.c      Fri Aug  6 
10:55:06 2004
***************
*** 120,129 ****
  static char command_line[COMMAND_LINE_SIZE];
  char saved_command_line[COMMAND_LINE_SIZE];
  
! static void __init parse_mem_cmdline (char ** cmdline_p)
  {
      char c = ' ', *to = command_line, *from = saved_command_line;
      int len = 0;
  
      /* Save unparsed command line copy for /proc/cmdline */
      memcpy(saved_command_line, start_info.cmd_line, COMMAND_LINE_SIZE);
--- 120,134 ----
  static char command_line[COMMAND_LINE_SIZE];
  char saved_command_line[COMMAND_LINE_SIZE];
  
! /* parse_mem_cmdline()
!  * returns the value of the mem= boot param converted to pages or 0
!  */ 
! static int __init parse_mem_cmdline (char ** cmdline_p)
  {
      char c = ' ', *to = command_line, *from = saved_command_line;
      int len = 0;
+     unsigned long long bytes;
+     int mem_param = 0;
  
      /* Save unparsed command line copy for /proc/cmdline */
      memcpy(saved_command_line, start_info.cmd_line, COMMAND_LINE_SIZE);
***************
*** 145,152 ****
              } else if (!memcmp(from+4, "exactmap", 8)) {
                  from += 8+4;
              } else {
!                 (void)memparse(from+4, &from);
!                 if (*from == '@')
                      (void)memparse(from+1, &from);
              }
          }
--- 150,158 ----
              } else if (!memcmp(from+4, "exactmap", 8)) {
                  from += 8+4;
              } else {
!                 bytes = memparse(from+4, &from);
!                 mem_param = bytes>>PAGE_SHIFT;
!               if (*from == '@')
                      (void)memparse(from+1, &from);
              }
          }
***************
*** 160,165 ****
--- 166,173 ----
      }
      *to = '\0';
      *cmdline_p = command_line;
+ 
+     return mem_param;
  }
  
  /*
***************
*** 194,200 ****
  
  void __init setup_arch(char **cmdline_p)
  {
!     unsigned long bootmap_size, start_pfn, max_low_pfn;
  
      extern void hypervisor_callback(void);
      extern void failsafe_callback(void);
--- 202,210 ----
  
  void __init setup_arch(char **cmdline_p)
  {
!     unsigned long bootmap_size, start_pfn, lmax_low_pfn;
!     int mem_param;  /* user specified memory size in pages */
!     int boot_pfn;   /* low pages available for bootmem */
  
      extern void hypervisor_callback(void);
      extern void failsafe_callback(void);
***************
*** 252,258 ****
      init_mm.end_data = (unsigned long) &_edata;
      init_mm.brk = (unsigned long) &_end;
  
!     parse_mem_cmdline(cmdline_p);
  
  #define PFN_UP(x)     (((x) + PAGE_SIZE-1) >> PAGE_SHIFT)
  #define PFN_DOWN(x)   ((x) >> PAGE_SHIFT)
--- 262,277 ----
      init_mm.end_data = (unsigned long) &_edata;
      init_mm.brk = (unsigned long) &_end;
  
!     /* The mem= kernel command line param overrides the detected amount
!      * of memory.   For xenolinux, if this override is larger than detected
!      * memory, then boot using only detected memory and make provisions to
!      * use all of the override value.   The hypervisor can give this
!      * domain more memory later on and it will be added to the free
!      * lists at that time.   See claim_new_pages() in
!      * arch/xen/drivers/balloon/balloon.c
!      */
!     mem_param = parse_mem_cmdline(cmdline_p);
!     if (!mem_param) mem_param = start_info.nr_pages;
  
  #define PFN_UP(x)     (((x) + PAGE_SIZE-1) >> PAGE_SHIFT)
  #define PFN_DOWN(x)   ((x) >> PAGE_SHIFT)
***************
*** 269,277 ****
      /*
       * Determine low and high memory ranges:
       */
!     max_low_pfn = max_pfn = start_info.nr_pages;
!     if (max_low_pfn > MAXMEM_PFN) {
!         max_low_pfn = MAXMEM_PFN;
  #ifndef CONFIG_HIGHMEM
          /* Maximum memory usable is what is directly addressable */
          printk(KERN_WARNING "Warning only %ldMB will be used.\n",
--- 288,296 ----
      /*
       * Determine low and high memory ranges:
       */
!     lmax_low_pfn = max_pfn = mem_param;
!     if (lmax_low_pfn > MAXMEM_PFN) {
!         lmax_low_pfn = MAXMEM_PFN;
  #ifndef CONFIG_HIGHMEM
          /* Maximum memory usable is what is directly addressable */
          printk(KERN_WARNING "Warning only %ldMB will be used.\n",
***************
*** 314,324 ****
       * bootstrap page table. We are guaranteed to get >=512kB unused 'padding'
       * for our own use after all bootstrap elements (see hypervisor-if.h).
       */
!     bootmap_size = init_bootmem(start_pfn, max_low_pfn);
!     free_bootmem(0, PFN_PHYS(max_low_pfn));
      reserve_bootmem(__pa(&_stext), 
                      PFN_PHYS(start_pfn) + bootmap_size + PAGE_SIZE-1 - 
                      __pa(&_stext));
  
  #ifdef CONFIG_BLK_DEV_INITRD
      if ( start_info.mod_start != 0 )
--- 333,351 ----
       * bootstrap page table. We are guaranteed to get >=512kB unused 'padding'
       * for our own use after all bootstrap elements (see hypervisor-if.h).
       */
!     boot_pfn = min((int)start_info.nr_pages,lmax_low_pfn);
!     bootmap_size = init_bootmem(start_pfn,boot_pfn);
!     free_bootmem(0, PFN_PHYS(boot_pfn));
      reserve_bootmem(__pa(&_stext), 
                      PFN_PHYS(start_pfn) + bootmap_size + PAGE_SIZE-1 - 
                      __pa(&_stext));
+ 
+     /* init_bootmem() set the global max_low_pfn to boot_pfn.  Now 
max_low_pfn 
+      * can be set to the override value.
+      */
+     max_low_pfn = lmax_low_pfn;
+ 
+ 
  
  #ifdef CONFIG_BLK_DEV_INITRD
      if ( start_info.mod_start != 0 )
===== linux-2.4.26-xen-sparse/arch/xen/mm/init.c 1.20 vs edited =====
*** /tmp/init.c-1.20-11013      Wed Jun  9 11:46:56 2004
--- edited/linux-2.4.26-xen-sparse/arch/xen/mm/init.c   Thu Aug  5 17:00:28 2004
***************
*** 219,229 ****
      pmd_t *kpmd, *pmd;
      pte_t *kpte, *pte, *pte_base;
  
      /*
       * This can be zero as well - no problem, in that case we exit
       * the loops anyway due to the PTRS_PER_* conditions.
       */
!     end = (unsigned long)__va(max_low_pfn*PAGE_SIZE);
  
      pgd_base = init_mm.pgd;
      i = __pgd_offset(PAGE_OFFSET);
--- 219,235 ----
      pmd_t *kpmd, *pmd;
      pte_t *kpte, *pte, *pte_base;
  
+     /* create tables only for boot_pfn frames.  max_low_pfn may be sized for
+      * pages yet to be allocated from the hypervisor, or it may be set
+      * to override the start_info amount of memory
+      */
+     int boot_pfn = min(start_info.nr_pages,max_low_pfn);
+ 
      /*
       * This can be zero as well - no problem, in that case we exit
       * the loops anyway due to the PTRS_PER_* conditions.
       */
!     end = (unsigned long)__va(boot_pfn *PAGE_SIZE);
  
      pgd_base = init_mm.pgd;
      i = __pgd_offset(PAGE_OFFSET);
***************
*** 308,314 ****
      pagetable_init();
  
      zone_sizes_init();
- 
      /* Switch to the real shared_info page, and clear the dummy page. */
      set_fixmap(FIX_SHARED_INFO, start_info.shared_info);
      HYPERVISOR_shared_info = (shared_info_t *)fix_to_virt(FIX_SHARED_INFO);
--- 314,319 ----
***************
*** 368,378 ****
  #endif
      int reservedpages, pfn;
  
      /* this will put all low memory onto the freelists */
      totalram_pages += free_all_bootmem();
  
      reservedpages = 0;
!     for (pfn = 0; pfn < max_low_pfn; pfn++) {
          /*
           * Only count reserved RAM pages
           */
--- 373,390 ----
  #endif
      int reservedpages, pfn;
  
+     /* add only boot_pfn pages of low memory to free list.
+      * max_low_pfn may be sized for
+      * pages yet to be allocated from the hypervisor, or it may be set
+      * to override the start_info amount of memory
+      */
+     int boot_pfn = min(start_info.nr_pages,max_low_pfn);
+ 
      /* this will put all low memory onto the freelists */
      totalram_pages += free_all_bootmem();
  
      reservedpages = 0;
!     for (pfn = 0; pfn < boot_pfn ; pfn++) {
          /*
           * Only count reserved RAM pages
           */
***************
*** 380,386 ****
              reservedpages++;
      }
  #ifdef CONFIG_HIGHMEM
!     for (pfn = highend_pfn-1; pfn >= highstart_pfn; pfn--)
          one_highpage_init((struct page *) (mem_map + pfn), pfn, bad_ppro);
      totalram_pages += totalhigh_pages;
  #endif
--- 392,398 ----
              reservedpages++;
      }
  #ifdef CONFIG_HIGHMEM
!     for (pfn = start_info.nr_pages-1; pfn >= highstart_pfn; pfn--)
          one_highpage_init((struct page *) (mem_map + pfn), pfn, bad_ppro);
      totalram_pages += totalhigh_pages;
  #endif
***************
*** 460,470 ****
  
  void si_meminfo(struct sysinfo *val)
  {
!     val->totalram = totalram_pages;
      val->sharedram = 0;
      val->freeram = nr_free_pages();
      val->bufferram = atomic_read(&buffermem_pages);
!     val->totalhigh = totalhigh_pages;
      val->freehigh = nr_free_highpages();
      val->mem_unit = PAGE_SIZE;
      return;
--- 472,482 ----
  
  void si_meminfo(struct sysinfo *val)
  {
!     val->totalram = max_pfn;
      val->sharedram = 0;
      val->freeram = nr_free_pages();
      val->bufferram = atomic_read(&buffermem_pages);
!     val->totalhigh = max_pfn-max_low_pfn;
      val->freehigh = nr_free_highpages();
      val->mem_unit = PAGE_SIZE;
      return;
*** /dev/null   Tue Jul 13 09:55:35 2004
--- setdomainmaxmem     Fri Aug  6 11:40:20 2004
***************
*** 0 ****
--- 1,34 ----
+ #!/usr/bin/env perl
+ 
+ use strict;
+ require "sys/ioctl.ph";
+ 
+ sub SIZEOF_HYPERCALL () { 24; }
+ sub STRUCT_PRIVCMD_HYPERCALL () {"L P";}
+ sub IOCTL_PRIVCMD_HYPERCALL ()
+         { &_IOC( &_IOC_NONE, ord('P'), 0, SIZEOF_HYPERCALL );}
+ sub __HYPERVISOR_dom0_op () {7;}
+ sub DOM0_INTERFACE_VERSION () {0xaaaa0010;}
+ sub DOM0_SETDOMAINMAXMEM () {28;}
+ sub STRUCT_DOM0_OP_PREFIX () {"L L";}
+ sub STRUCT_SETDOMAINMAXMEM () {STRUCT_DOM0_OP_PREFIX."L x4 L";}
+ sub XEN_PRIVCMD () {"/proc/xen/privcmd";}
+ 
+ sub setdomainmaxmem($$) {
+     my ($domain,$bytes) = @_;
+     my $msg = pack(STRUCT_SETDOMAINMAXMEM,DOM0_SETDOMAINMAXMEM,
+         DOM0_INTERFACE_VERSION,  $domain, $bytes);
+     my $cmd = pack(STRUCT_PRIVCMD_HYPERCALL,__HYPERVISOR_dom0_op,$msg);
+     open(XEN,XEN_PRIVCMD) or die "$!\n";
+     ioctl(XEN, IOCTL_PRIVCMD_HYPERCALL, $cmd) or die "ioctl: $!";
+     close XEN;
+ }
+ 
+ my ($bytes,$suffix) = $ARGV[1] =~ m/(^\d+)([mMkKgG])/;
+ $bytes<<=10 if $suffix =~ m/[kK]/;
+ $bytes<<=20 if $suffix =~ m/[mM]/;
+ $bytes<<=30 if $suffix =~ m/[gG]/;
+ 
+ printf "set domain $ARGV[0] to $bytes\n";
+ setdomainmaxmem($ARGV[0],$bytes);
+ 


-------------------------------------------------------
This SF.Net email is sponsored by OSTG. Have you noticed the changes on
Linux.com, ITManagersJournal and NewsForge in the past few weeks? Now,
one more big change to announce. We are now OSTG- Open Source Technology
Group. Come see the changes on the new OSTG site. www.ostg.com
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxxxx
https://lists.sourceforge.net/lists/listinfo/xen-devel


 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.