[MIPS] 64-bit Sibyte kernels need DMA32.

Sibyte SOCs only have 32-bit PCI.  Due to the sparse use of the address
space only the first 1GB of memory is mapped at physical addresses
below 1GB.  If a system has more than 1GB of memory 32-bit DMA will
not be able to reach all of it.

For now this patch is good enough to keep Sibyte users happy but it seems
eventually something like swiotlb will be needed for Sibyte.

Signed-off-by: Ralf Baechle <ralf@linux-mips.org>

4 files changed, 82 insertions, 37 deletions

diff --git a/arch/mips/Kconfig b/arch/mips/Kconfig
index 7750829b416a..4c6ba7b30a68 100644
--- a/arch/mips/Kconfig
+++ b/arch/mips/Kconfig
@@ -515,6 +515,7 @@ config SIBYTE_SWARM
         select SYS_SUPPORTS_HIGHMEM
         select SYS_SUPPORTS_KGDB
         select SYS_SUPPORTS_LITTLE_ENDIAN
+        select ZONE_DMA32 if 64BIT
 
 config SIBYTE_LITTLESUR
         bool "Sibyte BCM91250C2-LittleSur"
@@ -565,6 +566,7 @@ config SIBYTE_BIGSUR
         select SYS_SUPPORTS_BIG_ENDIAN
         select SYS_SUPPORTS_HIGHMEM
         select SYS_SUPPORTS_LITTLE_ENDIAN
+        select ZONE_DMA32 if 64BIT
 
 config SNI_RM
         bool "SNI RM200/300/400"
@@ -1664,6 +1666,9 @@ config ARCH_DISCONTIGMEM_ENABLE
           or have huge holes in the physical address space for other reasons.
           See <file:Documentation/vm/numa> for more.
 
+config ARCH_POPULATES_NODE_MAP
+        def_bool y
+
 config ARCH_SPARSEMEM_ENABLE
         bool
         select SPARSEMEM_STATIC
@@ -1969,6 +1974,9 @@ config I8253
 config PCSPEAKER
         bool
 
+config ZONE_DMA32
+        bool
+
 source "drivers/pcmcia/Kconfig"
 
 source "drivers/pci/hotplug/Kconfig"
diff --git a/arch/mips/kernel/setup.c b/arch/mips/kernel/setup.c
index a06a27d6cfcd..7f6ddcb5d485 100644
--- a/arch/mips/kernel/setup.c
+++ b/arch/mips/kernel/setup.c
@@ -269,7 +269,7 @@ static void __init bootmem_init(void)
 
 static void __init bootmem_init(void)
 {
-        unsigned long reserved_end;
+        unsigned long init_begin, reserved_end;
         unsigned long mapstart = ~0UL;
         unsigned long bootmap_size;
         int i;
@@ -342,6 +342,35 @@ static void __init bootmem_init(void)
          */
         bootmap_size = init_bootmem_node(NODE_DATA(0), mapstart,
                                          min_low_pfn, max_low_pfn);
+
+
+        init_begin = PFN_UP(__pa_symbol(&__init_begin));
+        for (i = 0; i < boot_mem_map.nr_map; i++) {
+                unsigned long start, end;
+
+                start = PFN_UP(boot_mem_map.map[i].addr);
+                end = PFN_DOWN(boot_mem_map.map[i].addr
+                                + boot_mem_map.map[i].size);
+
+                if (start <= init_begin)
+                        start = init_begin;
+                if (start >= end)
+                        continue;
+
+#ifndef CONFIG_HIGHMEM
+                if (end > max_low_pfn)
+                        end = max_low_pfn;
+
+                /*
+                 * ... finally, is the area going away?
+                 */
+                if (end <= start)
+                        continue;
+#endif
+
+                add_active_range(0, start, end);
+        }
+
         /*
          * Register fully available low RAM pages with the bootmem allocator.
          */
diff --git a/arch/mips/mm/dma-default.c b/arch/mips/mm/dma-default.c
index 33519ce49540..ae76795685cc 100644
--- a/arch/mips/mm/dma-default.c
+++ b/arch/mips/mm/dma-default.c
@@ -40,16 +40,38 @@ static inline int cpu_is_noncoherent_r10000(struct device *dev)
                current_cpu_type() == CPU_R12000);
 }
 
+static gfp_t massage_gfp_flags(const struct device *dev, gfp_t gfp)
+{
+        /* ignore region specifiers */
+        gfp &= ~(__GFP_DMA | __GFP_DMA32 | __GFP_HIGHMEM);
+
+#ifdef CONFIG_ZONE_DMA32
+        if (dev == NULL)
+                gfp |= __GFP_DMA;
+        else if (dev->coherent_dma_mask < DMA_BIT_MASK(24))
+                gfp |= __GFP_DMA;
+        else
+#endif
+#ifdef CONFIG_ZONE_DMA32
+             if (dev->coherent_dma_mask < DMA_BIT_MASK(32))
+                gfp |= __GFP_DMA32;
+        else
+#endif
+                ;
+
+        /* Don't invoke OOM killer */
+        gfp |= __GFP_NORETRY;
+
+        return gfp;
+}
+
 void *dma_alloc_noncoherent(struct device *dev, size_t size,
         dma_addr_t * dma_handle, gfp_t gfp)
 {
         void *ret;
 
-        /* ignore region specifiers */
-        gfp &= ~(__GFP_DMA | __GFP_HIGHMEM);
+        gfp = massage_gfp_flags(dev, gfp);
 
-        if (dev == NULL || (dev->coherent_dma_mask < 0xffffffff))
-                gfp |= GFP_DMA;
         ret = (void *) __get_free_pages(gfp, get_order(size));
 
         if (ret != NULL) {
@@ -67,11 +89,8 @@ void *dma_alloc_coherent(struct device *dev, size_t size,
 {
         void *ret;
 
-        /* ignore region specifiers */
-        gfp &= ~(__GFP_DMA | __GFP_HIGHMEM);
+        gfp = massage_gfp_flags(dev, gfp);
 
-        if (dev == NULL || (dev->coherent_dma_mask < 0xffffffff))
-                gfp |= GFP_DMA;
         ret = (void *) __get_free_pages(gfp, get_order(size));
 
         if (ret) {
@@ -343,7 +362,7 @@ int dma_supported(struct device *dev, u64 mask)
          * so we can't guarantee allocations that must be
          * within a tighter range than GFP_DMA..
          */
-        if (mask < 0x00ffffff)
+        if (mask < DMA_BIT_MASK(24))
                 return 0;
 
         return 1;
diff --git a/arch/mips/mm/init.c b/arch/mips/mm/init.c
index ec3b9e9f30f4..480dec04f552 100644
--- a/arch/mips/mm/init.c
+++ b/arch/mips/mm/init.c
@@ -347,11 +347,8 @@ static int __init page_is_ram(unsigned long pagenr)
 
 void __init paging_init(void)
 {
-        unsigned long zones_size[MAX_NR_ZONES] = { 0, };
-#ifndef CONFIG_FLATMEM
-        unsigned long zholes_size[MAX_NR_ZONES] = { 0, };
-        unsigned long i, j, pfn;
-#endif
+        unsigned long max_zone_pfns[MAX_NR_ZONES];
+        unsigned long lastpfn;
 
         pagetable_init();
 
@@ -361,35 +358,27 @@ void __init paging_init(void)
         kmap_coherent_init();
 
 #ifdef CONFIG_ZONE_DMA
-        if (min_low_pfn < MAX_DMA_PFN && MAX_DMA_PFN <= max_low_pfn) {
-                zones_size[ZONE_DMA] = MAX_DMA_PFN - min_low_pfn;
-                zones_size[ZONE_NORMAL] = max_low_pfn - MAX_DMA_PFN;
-        } else if (max_low_pfn < MAX_DMA_PFN)
-                zones_size[ZONE_DMA] = max_low_pfn - min_low_pfn;
-        else
+        max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
 #endif
-        zones_size[ZONE_NORMAL] = max_low_pfn - min_low_pfn;
-
+#ifdef CONFIG_ZONE_DMA32
+        max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
+#endif
+        max_zone_pfns[ZONE_NORMAL] = max_low_pfn;
+        lastpfn = max_low_pfn;
 #ifdef CONFIG_HIGHMEM
-        zones_size[ZONE_HIGHMEM] = highend_pfn - highstart_pfn;
+        max_zone_pfns[ZONE_HIGHMEM] = highend_pfn;
+        lastpfn = highend_pfn;
 
-        if (cpu_has_dc_aliases && zones_size[ZONE_HIGHMEM]) {
+        if (cpu_has_dc_aliases && max_low_pfn != highend_pfn) {
                 printk(KERN_WARNING "This processor doesn't support highmem."
-                       " %ldk highmem ignored\n", zones_size[ZONE_HIGHMEM]);
-                zones_size[ZONE_HIGHMEM] = 0;
+                       " %ldk highmem ignored\n",
+                       (highend_pfn - max_low_pfn) << (PAGE_SHIFT - 10));
+                max_zone_pfns[ZONE_HIGHMEM] = max_low_pfn;
+                lastpfn = max_low_pfn;
         }
 #endif
 
-#ifdef CONFIG_FLATMEM
-        free_area_init(zones_size);
-#else
-        pfn = min_low_pfn;
-        for (i = 0; i < MAX_NR_ZONES; i++)
-                for (j = 0; j < zones_size[i]; j++, pfn++)
-                        if (!page_is_ram(pfn))
-                                zholes_size[i]++;
-        free_area_init_node(0, NODE_DATA(0), zones_size, 0, zholes_size);
-#endif
+        free_area_init_nodes(max_zone_pfns);
 }
 
 static struct kcore_list kcore_mem, kcore_vmalloc;