ARC: support HIGHMEM even without PAE40

Initial HIGHMEM support on ARC was introduced for PAE40 where the low
memory (0x8000_0000 based) and high memory (0x1_0000_0000) were
physically contiguous. So CONFIG_FLATMEM sufficed (despite a peipheral
hole in the middle, which wasted a bit of struct page memory, but things
worked).

However w/o PAE, highmem was not possible and we could only reach
~1.75GB of DDR. Now there is a use case to access ~4GB of DDR w/o PAE40
The idea is to have low memory at canonical 0x8000_0000 and highmem
at 0 so enire 4GB address space is available for physical addressing
This needs additional platform/interconnect mapping to convert
the non contiguous physical addresses into linear bus adresses.

From Linux point of view, non contiguous divide means FLATMEM no
longer works and DISCONTIGMEM is needed to track the pfns in the 2
regions.

This scheme would also work for PAE40, only better in that we don't
waste struct page memory for the peripheral hole.

The DT description will be something like

memory {
...
reg = <0x80000000 0x200000000 /* 512MB: lowmem */
0x00000000 0x10000000>; /* 256MB: highmem */
}

Signed-off-by: Noam Camus <[e-mail]>
Signed-off-by: Vineet Gupta <[e-mail]>

arch/arc/Kconfig

···
 config RWSEM_GENERIC_SPINLOCK
 	def_bool y
 
+config ARCH_DISCONTIGMEM_ENABLE
+	def_bool y
+
 config ARCH_FLATMEM_ENABLE
 	def_bool y
 
···
 
 endchoice
 
+config NODES_SHIFT
+	int "Maximum NUMA Nodes (as a power of 2)"
+	default "1" if !DISCONTIGMEM
+	default "2" if DISCONTIGMEM
+	depends on NEED_MULTIPLE_NODES
+	---help---
+	  Accessing memory beyond 1GB (with or w/o PAE) requires 2 memory
+	  zones.
+
 if ISA_ARCOMPACT
 
 config ARC_COMPACT_IRQ_LEVELS
···
 
 config HIGHMEM
 	bool "High Memory Support"
+	select DISCONTIGMEM
 	help
 	  With ARC 2G:2G address split, only upper 2G is directly addressable by
 	  kernel. Enable this to potentially allow access to rest of 2G and PAE

arch/arc/include/asm/mmzone.h

···
+/*
+ * Copyright (C) 2016 Synopsys, Inc. (www.synopsys.com)
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#ifndef _ASM_ARC_MMZONE_H
+#define _ASM_ARC_MMZONE_H
+
+#ifdef CONFIG_DISCONTIGMEM
+
+extern struct pglist_data node_data[];
+#define NODE_DATA(nid) (&node_data[nid])
+
+static inline int pfn_to_nid(unsigned long pfn)
+{
+	int is_end_low = 1;
+
+	if (IS_ENABLED(CONFIG_ARC_HAS_PAE40))
+		is_end_low = pfn <= virt_to_pfn(0xFFFFFFFFUL);
+
+	/*
+	 * node 0: lowmem:             0x8000_0000   to 0xFFFF_FFFF
+	 * node 1: HIGHMEM w/o  PAE40: 0x0           to 0x7FFF_FFFF
+	 *         HIGHMEM with PAE40: 0x1_0000_0000 to ...
+	 */
+	if (pfn >= ARCH_PFN_OFFSET && is_end_low)
+		return 0;
+
+	return 1;
+}
+
+static inline int pfn_valid(unsigned long pfn)
+{
+	int nid = pfn_to_nid(pfn);
+
+	return (pfn <= node_end_pfn(nid));
+}
+#endif /* CONFIG_DISCONTIGMEM  */
+
+#endif

arch/arc/include/asm/page.h

···
 
 #define ARCH_PFN_OFFSET		virt_to_pfn(CONFIG_LINUX_LINK_BASE)
 
+#ifdef CONFIG_FLATMEM
 #define pfn_valid(pfn)		(((pfn) - ARCH_PFN_OFFSET) < max_mapnr)
+#endif
 
 /*
  * __pa, __va, virt_to_page (ALERT: deprecated, don't use them)
···
 #define __pa(vaddr)  ((unsigned long)(vaddr))
 #define __va(paddr)  ((void *)((unsigned long)(paddr)))
 
-#define virt_to_page(kaddr)	\
-	(mem_map + virt_to_pfn((kaddr) - CONFIG_LINUX_LINK_BASE))
-
+#define virt_to_page(kaddr)	pfn_to_page(virt_to_pfn(kaddr))
 #define virt_addr_valid(kaddr)  pfn_valid(virt_to_pfn(kaddr))
 
 /* Default Permissions for stack/heaps pages (Non Executable) */

arch/arc/mm/init.c

···
 static unsigned long low_mem_sz;
 
 #ifdef CONFIG_HIGHMEM
-static unsigned long min_high_pfn;
+static unsigned long min_high_pfn, max_high_pfn;
 static u64 high_mem_start;
 static u64 high_mem_sz;
+#endif
+
+#ifdef CONFIG_DISCONTIGMEM
+struct pglist_data node_data[MAX_NUMNODES] __read_mostly;
+EXPORT_SYMBOL(node_data);
 #endif
 
 /* User can over-ride above with "mem=nnn[KkMm]" in cmdline */
···
 	/* Last usable page of low mem */
 	max_low_pfn = max_pfn = PFN_DOWN(low_mem_start + low_mem_sz);
 
-#ifdef CONFIG_HIGHMEM
-	min_high_pfn = PFN_DOWN(high_mem_start);
-	max_pfn = PFN_DOWN(high_mem_start + high_mem_sz);
+#ifdef CONFIG_FLATMEM
+	/* pfn_valid() uses this */
+	max_mapnr = max_low_pfn - min_low_pfn;
 #endif
-
-	max_mapnr = max_pfn - min_low_pfn;
 
 	/*------------- bootmem allocator setup -----------------------*/
 
···
 	 * the crash
 	 */
 
-	memblock_add(low_mem_start, low_mem_sz);
+	memblock_add_node(low_mem_start, low_mem_sz, 0);
 	memblock_reserve(low_mem_start, __pa(_end) - low_mem_start);
 
 #ifdef CONFIG_BLK_DEV_INITRD
···
 	zones_size[ZONE_NORMAL] = max_low_pfn - min_low_pfn;
 	zones_holes[ZONE_NORMAL] = 0;
 
-#ifdef CONFIG_HIGHMEM
-	zones_size[ZONE_HIGHMEM] = max_pfn - max_low_pfn;
-
-	/* This handles the peripheral address space hole */
-	zones_holes[ZONE_HIGHMEM] = min_high_pfn - max_low_pfn;
-#endif
-
 	/*
 	 * We can't use the helper free_area_init(zones[]) because it uses
 	 * PAGE_OFFSET to compute the @min_low_pfn which would be wrong
···
 			    zones_holes);	/* holes */
 
 #ifdef CONFIG_HIGHMEM
+	/*
+	 * Populate a new node with highmem
+	 *
+	 * On ARC (w/o PAE) HIGHMEM addresses are actually smaller (0 based)
+	 * than addresses in normal ala low memory (0x8000_0000 based).
+	 * Even with PAE, the huge peripheral space hole would waste a lot of
+	 * mem with single mem_map[]. This warrants a mem_map per region design.
+	 * Thus HIGHMEM on ARC is imlemented with DISCONTIGMEM.
+	 *
+	 * DISCONTIGMEM in turns requires multiple nodes. node 0 above is
+	 * populated with normal memory zone while node 1 only has highmem
+	 */
+	node_set_online(1);
+
+	min_high_pfn = PFN_DOWN(high_mem_start);
+	max_high_pfn = PFN_DOWN(high_mem_start + high_mem_sz);
+
+	zones_size[ZONE_NORMAL] = 0;
+	zones_holes[ZONE_NORMAL] = 0;
+
+	zones_size[ZONE_HIGHMEM] = max_high_pfn - min_high_pfn;
+	zones_holes[ZONE_HIGHMEM] = 0;
+
+	free_area_init_node(1,			/* node-id */
+			    zones_size,		/* num pages per zone */
+			    min_high_pfn,	/* first pfn of node */
+			    zones_holes);	/* holes */
+
 	high_memory = (void *)(min_high_pfn << PAGE_SHIFT);
 	kmap_init();
 #endif
···
 	unsigned long tmp;
 
 	reset_all_zones_managed_pages();
-	for (tmp = min_high_pfn; tmp < max_pfn; tmp++)
+	for (tmp = min_high_pfn; tmp < max_high_pfn; tmp++)
 		free_highmem_page(pfn_to_page(tmp));
 #endif