ppc64 iommu rewrite part 1/5

PATCH New Benjamin Herrenschmidt 1 participant 1 message Feb 27, 2004

lore

Share & tools

lore

Benjamin Herrenschmidt

author

2004-02-27 12:07 UTC #0 lore raw

PATCH 1/1 ppc64 iommu rewrite part 1/5

This patch fixes the accounting of physical RAM.

On machines with an IO hole (like Apple G5 with more than 2Gb of RAM, but also some pSeries) we failed to properly account for the real amount of physical RAM and inform the zone allocator of our hole size.

During the process, I included Anton slaughtering of the guard page we had in the first 256Mb kernel segment, thus allowing this segment to be mapped with large pages as it should be.

# This is a BitKeeper generated diff -Nru style patch. #
# ChangeSet

#   2004/02/27 21:55:55+11:00 [email protected] 
#   ppc64: Fix physical memory accounting on machines with IO hole and
#   remove guard page, kernel mapping is now 100% large pages
# 
# include/asm-ppc64/page.h
#   2004/02/27 21:43:23+11:00 [email protected] +3 -0
#   ppc64: Fix physical memory accounting on machines with IO hole and
#   remove guard page, kernel mapping is now 100% large pages
# 
# arch/ppc64/mm/numa.c
#   2004/02/27 21:43:23+11:00 [email protected] +22 -7
#   ppc64: Fix physical memory accounting on machines with IO hole and
#   remove guard page, kernel mapping is now 100% large pages
# 
# arch/ppc64/mm/init.c
#   2004/02/27 21:43:23+11:00 [email protected] +76 -18
#   ppc64: Fix physical memory accounting on machines with IO hole and
#   remove guard page, kernel mapping is now 100% large pages
# 
# arch/ppc64/mm/hash_utils.c
#   2004/02/27 21:43:23+11:00 [email protected] +16 -12
#   ppc64: Fix physical memory accounting on machines with IO hole and
#   remove guard page, kernel mapping is now 100% large pages
# 
# arch/ppc64/kernel/stab.c
#   2004/02/27 21:43:22+11:00 [email protected] +5 -1
#   ppc64: Fix physical memory accounting on machines with IO hole and
#   remove guard page, kernel mapping is now 100% large pages
# 
# arch/ppc64/kernel/lmb.c
#   2004/02/27 21:43:22+11:00 [email protected] +10 -10
#   ppc64: Fix physical memory accounting on machines with IO hole and
#   remove guard page, kernel mapping is now 100% large pages
# 
diff -Nru a/arch/ppc64/kernel/lmb.c b/arch/ppc64/kernel/lmb.c
--- a/arch/ppc64/kernel/lmb.c	Fri Feb 27 22:42:19 2004
+++ b/arch/ppc64/kernel/lmb.c	Fri Feb 27 22:42:19 2004
@@ -269,11 +269,13 @@
 	return _lmb->memory.size;
 #else
 	struct lmb_region *_mem = &(_lmb->memory);
-	unsigned long idx = _mem->cnt-1;
-	unsigned long lastbase = _mem->region[idx].physbase;
-	unsigned long lastsize = _mem->region[idx].size;
-	
-	return (lastbase + lastsize);
+	unsigned long total = 0;
+	int i;
+
+	/* add all physical memory to the bootmem map */
+	for (i=0; i < _mem->cnt; i++)
+		total += _mem->region[i].size;
+	return total;
 #endif /* CONFIG_MSCHUNKS */
 }

@@ -283,15 +285,13 @@
unsigned long offset = reloc_offset(); struct lmb *_lmb = PTRRELOC(&lmb); struct lmb_region *_mem = &(_lmb->memory);
- unsigned long idx;
+ int idx = _mem->cnt - 1;

- for(idx=_mem->cnt-1; idx >= 0; idx--) {
#ifdef CONFIG_MSCHUNKS
- return (_mem->region[idx].physbase + _mem->region[idx].size);
+ return (_mem->region[idx].physbase + _mem->region[idx].size);
#else
- return (_mem->region[idx].base + _mem->region[idx].size);
+ return (_mem->region[idx].base + _mem->region[idx].size);
#endif /* CONFIG_MSCHUNKS */
- }

return 0;
}

arch/ppc64/kernel/stab.c +5 -1

diff -Nru a/arch/ppc64/kernel/stab.c b/arch/ppc64/kernel/stab.c
@@ -32,10 +32,14 @@
 void stab_initialize(unsigned long stab)
 {
 	unsigned long esid, vsid; 
+	int seg0_largepages = 0;
 
 	esid = GET_ESID(KERNELBASE);
 	vsid = get_kernel_vsid(esid << SID_SHIFT); 
 
+	if (cur_cpu_spec->cpu_features & CPU_FTR_16M_PAGE)
+		seg0_largepages = 1;
+
 	if (cur_cpu_spec->cpu_features & CPU_FTR_SLB) {
 		/* Invalidate the entire SLB & all the ERATS */
 #ifdef CONFIG_PPC_ISERIES

@@ -44,7 +48,7 @@
 		asm volatile("isync":::"memory");
 		asm volatile("slbmte  %0,%0"::"r" (0) : "memory");
 		asm volatile("isync; slbia; isync":::"memory");
-		make_slbe(esid, vsid, 0, 1);
+		make_slbe(esid, vsid, seg0_largepages, 1);
 		asm volatile("isync":::"memory");
 #endif
 	} else {

arch/ppc64/mm/hash_utils.c +16 ~~-12~~

diff -Nru a/arch/ppc64/mm/hash_utils.c b/arch/ppc64/mm/hash_utils.c
@@ -123,6 +123,7 @@
 	unsigned long table, htab_size_bytes;
 	unsigned long pteg_count;
 	unsigned long mode_rw;
+	int i, use_largepages = 0;
 
 	/*
 	 * Calculate the required size of the htab.  We want the number of

@@ -165,18 +166,21 @@
 
 	mode_rw = _PAGE_ACCESSED | _PAGE_COHERENT | PP_RWXX;
 
-	/* XXX we currently map kernel text rw, should fix this */
-	if ((cur_cpu_spec->cpu_features & CPU_FTR_16M_PAGE)
-	    && systemcfg->physicalMemorySize > 256*MB) {
-		create_pte_mapping((unsigned long)KERNELBASE, 
-				   KERNELBASE + 256*MB, mode_rw, 0);
-		create_pte_mapping((unsigned long)KERNELBASE + 256*MB, 
-				   KERNELBASE + (systemcfg->physicalMemorySize), 
-				   mode_rw, 1);
-	} else {
-		create_pte_mapping((unsigned long)KERNELBASE, 
-				   KERNELBASE+(systemcfg->physicalMemorySize), 
-				   mode_rw, 0);
+	/* On U3 based machines, we need to reserve the DART area and
+	 * _NOT_ map it to avoid cache paradoxes as it's remapped non
+	 * cacheable later on
+	 */
+	if (cur_cpu_spec->cpu_features & CPU_FTR_16M_PAGE)
+		use_largepages = 1;
+
+	/* add all physical memory to the bootmem map */
+	for (i=0; i < lmb.memory.cnt; i++) {
+		unsigned long base, size;
+
+		base = lmb.memory.region[i].physbase + KERNELBASE;
+		size = lmb.memory.region[i].size;
+
+		create_pte_mapping(base, base + size, mode_rw, use_largepages);
 	}
 }
 #undef KB

arch/ppc64/mm/init.c +76 ~~-18~~

134 lines not on this page · raw message

arch/ppc64/mm/numa.c +22 -7

diff -Nru a/arch/ppc64/mm/numa.c b/arch/ppc64/mm/numa.c
@@ -30,6 +30,7 @@
 
 struct pglist_data node_data[MAX_NUMNODES];
 bootmem_data_t plat_node_bdata[MAX_NUMNODES];
+static unsigned long node0_io_hole_size;
 
 EXPORT_SYMBOL(node_data);
 EXPORT_SYMBOL(numa_memory_lookup_table);

@@ -205,8 +206,15 @@
 
 static void __init setup_nonnuma(void)
 {
+	unsigned long top_of_ram = lmb_end_of_DRAM();
+	unsigned long total_ram = lmb_phys_mem_size();
 	unsigned long i;
 
+	printk(KERN_INFO "Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
+	       top_of_ram, total_ram);
+	printk(KERN_INFO "Memory hole size: %ldMB\n",
+	       (top_of_ram - total_ram) >> 20);
+
 	for (i = 0; i < NR_CPUS; i++)
 		map_cpu_to_node(i, 0);

@@ -215,8 +223,10 @@
 	node_data[0].node_start_pfn = 0;
 	node_data[0].node_spanned_pages = lmb_end_of_DRAM() / PAGE_SIZE;
 
-	for (i = 0 ; i < lmb_end_of_DRAM(); i += MEMORY_INCREMENT)
+	for (i = 0 ; i < top_of_ram; i += MEMORY_INCREMENT)
 		numa_memory_lookup_table[i >> MEMORY_INCREMENT_SHIFT] = 0;
+
+	node0_io_hole_size = top_of_ram - total_ram;
 }
 
 void __init do_init_bootmem(void)

@@ -309,11 +319,12 @@
 void __init paging_init(void)
 {
 	unsigned long zones_size[MAX_NR_ZONES];
-	int i, nid;
+	unsigned long zholes_size[MAX_NR_ZONES];
 	struct page *node_mem_map; 
+	int nid;
 
-	for (i = 1; i < MAX_NR_ZONES; i++)
-		zones_size[i] = 0;
+	memset(zones_size, 0, sizeof(zones_size));
+	memset(zholes_size, 0, sizeof(zholes_size));
 
 	for (nid = 0; nid < numnodes; nid++) {
 		unsigned long start_pfn;

@@ -323,8 +334,12 @@
 		end_pfn = plat_node_bdata[nid].node_low_pfn;
 
 		zones_size[ZONE_DMA] = end_pfn - start_pfn;
-		dbg("free_area_init node %d %lx %lx\n", nid,
-				zones_size[ZONE_DMA], start_pfn);
+		zholes_size[ZONE_DMA] = 0;
+		if (nid == 0)
+			zholes_size[ZONE_DMA] = node0_io_hole_size;
+
+		dbg("free_area_init node %d %lx %lx (hole: %lx)\n", nid,
+		    zones_size[ZONE_DMA], start_pfn, zholes_size[ZONE_DMA]);
 
 		/* 
 		 * Give this empty node a dummy struct page to avoid

@@ -337,6 +352,6 @@
 			node_mem_map = NULL;
 
 		free_area_init_node(nid, NODE_DATA(nid), node_mem_map,
-				    zones_size, start_pfn, NULL);
+				    zones_size, start_pfn, zholes_size);
 	}
 }

include/asm-ppc64/page.h +3 -0

diff -Nru a/include/asm-ppc64/page.h b/include/asm-ppc64/page.h
@@ -163,6 +163,9 @@
 
 #define __pa(x) ((unsigned long)(x)-PAGE_OFFSET)
 
+/* Not 100% correct, for use by /dev/mem only */
+extern int page_is_ram(unsigned long physaddr);
+
 #endif /* __ASSEMBLY__ */
 
 #ifdef MODULE