[SPARC64]: Fix 32-bit truncation which broke sparsemem.
David S. Miller [Wed, 8 Mar 2006 23:57:03 +0000 (15:57 -0800)]
The page->flags manipulations done by the D-cache dirty
state tracking was broken because the constants were not
marked with "UL" to make them 64-bit, which means we were
clobbering the upper 32-bits of page->flags all the time.

This doesn't jive well with sparsemem which stores the
section and indexing information in the top 32-bits of
page->flags.

This is yet another sparc64 bug which has been with us
forever.

While we're here, tidy up some things in bootmem_init()
and paginig_init():

1) Pass min_low_pfn to init_bootmem_node(), it's identical
   to (phys_base >> PAGE_SHIFT) but we should use consistent
   with the variable names we print in CONFIG_BOOTMEM_DEBUG

2) max_mapnr, although no longer used, was being set
   inaccurately, we shouldn't subtract pfn_base any more.

3) All the games with phys_base in the zones_*[] arrays
   we pass to free_area_init_node() are no longer necessary.

Thanks to Josh Grebe and Fabbione for the bug reports
and testing.  Fix also verified locally on an SB2500
which had a memory layout that triggered the same problem.

Signed-off-by: David S. Miller <davem@davemloft.net>

arch/sparc64/mm/init.c

index 5f67b53..b40f647 100644 (file)
@@ -205,8 +205,8 @@ inline void flush_dcache_page_impl(struct page *page)
 }
 
 #define PG_dcache_dirty                PG_arch_1
-#define PG_dcache_cpu_shift    24
-#define PG_dcache_cpu_mask     (256 - 1)
+#define PG_dcache_cpu_shift    24UL
+#define PG_dcache_cpu_mask     (256UL - 1UL)
 
 #if NR_CPUS > 256
 #error D-cache dirty tracking and thread_info->cpu need fixing for > 256 cpus
@@ -901,8 +901,7 @@ static unsigned long __init bootmem_init(unsigned long *pages_avail,
                    min_low_pfn, bootmap_pfn, max_low_pfn);
 #endif
        bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap_pfn,
-                                        (phys_base >> PAGE_SHIFT),
-                                        end_pfn);
+                                        min_low_pfn, end_pfn);
 
        /* Now register the available physical memory with the
         * allocator.
@@ -1311,25 +1310,24 @@ void __init paging_init(void)
        pages_avail = 0;
        last_valid_pfn = end_pfn = bootmem_init(&pages_avail, phys_base);
 
-       max_mapnr = last_valid_pfn - (phys_base >> PAGE_SHIFT);
+       max_mapnr = last_valid_pfn;
 
        kernel_physical_mapping_init();
 
        {
                unsigned long zones_size[MAX_NR_ZONES];
                unsigned long zholes_size[MAX_NR_ZONES];
-               unsigned long npages;
                int znum;
 
                for (znum = 0; znum < MAX_NR_ZONES; znum++)
                        zones_size[znum] = zholes_size[znum] = 0;
 
-               npages = end_pfn - (phys_base >> PAGE_SHIFT);
-               zones_size[ZONE_DMA] = npages;
-               zholes_size[ZONE_DMA] = npages - pages_avail;
+               zones_size[ZONE_DMA] = end_pfn;
+               zholes_size[ZONE_DMA] = end_pfn - pages_avail;
 
                free_area_init_node(0, &contig_page_data, zones_size,
-                                   phys_base >> PAGE_SHIFT, zholes_size);
+                                   __pa(PAGE_OFFSET) >> PAGE_SHIFT,
+                                   zholes_size);
        }
 
        device_scan();