Merge intel.com:/data/home/aegl/BK/work/1

into intel.com:/data/home/aegl/BK/linux-ia64-release-2.6.12

8 files changed, 174 insertions, 105 deletions

diff --git a/arch/ia64/kernel/unaligned.c b/arch/ia64/kernel/unaligned.c
index 46dad0d215a3dc..43b45b65ee5a9d 100644
--- a/arch/ia64/kernel/unaligned.c
+++ b/arch/ia64/kernel/unaligned.c
@@ -1380,6 +1380,10 @@ ia64_handle_unaligned (unsigned long ifa, struct pt_regs *regs)
 	 *		- ldX.spill
 	 *		- stX.spill
 	 *	Reason: RNATs are based on addresses
+	 *		- ld16
+	 *		- st16
+	 *	Reason: ld16 and st16 are supposed to occur in a single
+	 *		memory op
 	 *
 	 *	synchronization:
 	 *		- cmpxchg
@@ -1401,6 +1405,10 @@ ia64_handle_unaligned (unsigned long ifa, struct pt_regs *regs)
 	switch (opcode) {
 	      case LDS_OP:
 	      case LDSA_OP:
+		if (u.insn.x)
+			/* oops, really a semaphore op (cmpxchg, etc) */
+			goto failure;
+		/* no break */
 	      case LDS_IMM_OP:
 	      case LDSA_IMM_OP:
 	      case LDFS_OP:
@@ -1425,6 +1433,10 @@ ia64_handle_unaligned (unsigned long ifa, struct pt_regs *regs)
 	      case LDCCLR_OP:
 	      case LDCNC_OP:
 	      case LDCCLRACQ_OP:
+		if (u.insn.x)
+			/* oops, really a semaphore op (cmpxchg, etc) */
+			goto failure;
+		/* no break */
 	      case LD_IMM_OP:
 	      case LDA_IMM_OP:
 	      case LDBIAS_IMM_OP:
@@ -1437,6 +1449,10 @@ ia64_handle_unaligned (unsigned long ifa, struct pt_regs *regs)
 
 	      case ST_OP:
 	      case STREL_OP:
+		if (u.insn.x)
+			/* oops, really a semaphore op (cmpxchg, etc) */
+			goto failure;
+		/* no break */
 	      case ST_IMM_OP:
 	      case STREL_IMM_OP:
 		ret = emulate_store_int(ifa, u.insn, regs);
diff --git a/arch/ia64/pci/pci.c b/arch/ia64/pci/pci.c
index 61d8dd3c9c93a0..88641e5095b5f9 100644
--- a/arch/ia64/pci/pci.c
+++ b/arch/ia64/pci/pci.c
@@ -178,30 +178,6 @@ alloc_pci_controller (int seg)
 	return controller;
 }
 
-static int __devinit
-alloc_resource (char *name, struct resource *root, unsigned long start, unsigned long end,
-		unsigned long flags)
-{
-	struct resource *res;
-
-	res = kmalloc(sizeof(*res), GFP_KERNEL);
-	if (!res)
-		return -ENOMEM;
-
-	memset(res, 0, sizeof(*res));
-	res->name = name;
-	res->start = start;
-	res->end = end;
-	res->flags = flags;
-
-	if (insert_resource(root, res))	{
-		kfree(res);
-		return -EBUSY;
-	}
-
-	return 0;
-}
-
 static u64 __devinit
 add_io_space (struct acpi_resource_address64 *addr)
 {
@@ -254,10 +230,9 @@ struct pci_root_info {
 	char *name;
 };
 
-static acpi_status __devinit
-add_window (struct acpi_resource *res, void *data)
+static __devinit acpi_status add_window(struct acpi_resource *res, void *data)
 {
-	struct pci_root_info *info = (struct pci_root_info *) data;
+	struct pci_root_info *info = data;
 	struct pci_window *window;
 	struct acpi_resource_address64 addr;
 	acpi_status status;
@@ -265,45 +240,71 @@ add_window (struct acpi_resource *res, void *data)
 	struct resource *root;
 
 	status = acpi_resource_to_address64(res, &addr);
-	if (ACPI_SUCCESS(status)) {
-		if (!addr.address_length)
-			return AE_OK;
-
-		if (addr.resource_type == ACPI_MEMORY_RANGE) {
-			flags = IORESOURCE_MEM;
-			root = &iomem_resource;
-			offset = addr.address_translation_offset;
-		} else if (addr.resource_type == ACPI_IO_RANGE) {
-			flags = IORESOURCE_IO;
-			root = &ioport_resource;
-			offset = add_io_space(&addr);
-			if (offset == ~0)
-				return AE_OK;
-		} else
+	if (!ACPI_SUCCESS(status))
+		return AE_OK;
+
+	if (!addr.address_length)
+		return AE_OK;
+
+	if (addr.resource_type == ACPI_MEMORY_RANGE) {
+		flags = IORESOURCE_MEM;
+		root = &iomem_resource;
+		offset = addr.address_translation_offset;
+	} else if (addr.resource_type == ACPI_IO_RANGE) {
+		flags = IORESOURCE_IO;
+		root = &ioport_resource;
+		offset = add_io_space(&addr);
+		if (offset == ~0)
 			return AE_OK;
-
-		window = &info->controller->window[info->controller->windows++];
-		window->resource.flags	= flags;
-		window->resource.start  = addr.min_address_range;
-		window->resource.end    = addr.max_address_range;
-		window->offset		= offset;
-
-		if (alloc_resource(info->name, root, addr.min_address_range + offset,
-			addr.max_address_range + offset, flags))
-			printk(KERN_ERR "alloc 0x%lx-0x%lx from %s for %s failed\n",
-				addr.min_address_range + offset, addr.max_address_range + offset,
-				root->name, info->name);
+	} else
+		return AE_OK;
+
+	window = &info->controller->window[info->controller->windows++];
+	window->resource.name = info->name;
+	window->resource.flags = flags;
+	window->resource.start = addr.min_address_range + offset;
+	window->resource.end = addr.max_address_range + offset;
+	window->resource.child = NULL;
+	window->offset = offset;
+
+	if (insert_resource(root, &window->resource)) {
+		printk(KERN_ERR "alloc 0x%lx-0x%lx from %s for %s failed\n",
+			window->resource.start, window->resource.end,
+			root->name, info->name);
 	}
 
 	return AE_OK;
 }
 
+static void __devinit
+pcibios_setup_root_windows(struct pci_bus *bus, struct pci_controller *ctrl)
+{
+	int i, j;
+
+	j = 0;
+	for (i = 0; i < ctrl->windows; i++) {
+		struct resource *res = &ctrl->window[i].resource;
+		/* HP's firmware has a hack to work around a Windows bug.
+		 * Ignore these tiny memory ranges */
+		if ((res->flags & IORESOURCE_MEM) &&
+		    (res->end - res->start < 16))
+			continue;
+		if (j >= PCI_BUS_NUM_RESOURCES) {
+			printk("Ignoring range [%lx-%lx] (%lx)\n", res->start,
+					res->end, res->flags);
+			continue;
+		}
+		bus->resource[j++] = res;
+	}
+}
+
 struct pci_bus * __devinit
-pci_acpi_scan_root (struct acpi_device *device, int domain, int bus)
+pci_acpi_scan_root(struct acpi_device *device, int domain, int bus)
 {
 	struct pci_root_info info;
 	struct pci_controller *controller;
 	unsigned int windows = 0;
+	struct pci_bus *pbus;
 	char *name;
 
 	controller = alloc_pci_controller(domain);
@@ -312,8 +313,10 @@ pci_acpi_scan_root (struct acpi_device *device, int domain, int bus)
 
 	controller->acpi_handle = device->handle;
 
-	acpi_walk_resources(device->handle, METHOD_NAME__CRS, count_window, &windows);
-	controller->window = kmalloc(sizeof(*controller->window) * windows, GFP_KERNEL);
+	acpi_walk_resources(device->handle, METHOD_NAME__CRS, count_window,
+			&windows);
+	controller->window = kmalloc(sizeof(*controller->window) * windows,
+			GFP_KERNEL);
 	if (!controller->window)
 		goto out2;
 
@@ -324,9 +327,14 @@ pci_acpi_scan_root (struct acpi_device *device, int domain, int bus)
 	sprintf(name, "PCI Bus %04x:%02x", domain, bus);
 	info.controller = controller;
 	info.name = name;
-	acpi_walk_resources(device->handle, METHOD_NAME__CRS, add_window, &info);
+	acpi_walk_resources(device->handle, METHOD_NAME__CRS, add_window,
+			&info);
+
+	pbus = pci_scan_bus(bus, &pci_root_ops, controller);
+	if (pbus)
+		pcibios_setup_root_windows(pbus, controller);
 
-	return pci_scan_bus(bus, &pci_root_ops, controller);
+	return pbus;
 
 out3:
 	kfree(controller->window);
@@ -347,9 +355,9 @@ void pcibios_resource_to_bus(struct pci_dev *dev,
 		struct pci_window *window = &controller->window[i];
 		if (!(window->resource.flags & res->flags))
 			continue;
-		if (window->resource.start > res->start - window->offset)
+		if (window->resource.start > res->start)
 			continue;
-		if (window->resource.end < res->end - window->offset)
+		if (window->resource.end < res->end)
 			continue;
 		offset = window->offset;
 		break;
@@ -371,9 +379,9 @@ void pcibios_bus_to_resource(struct pci_dev *dev,
 		struct pci_window *window = &controller->window[i];
 		if (!(window->resource.flags & res->flags))
 			continue;
-		if (window->resource.start > region->start)
+		if (window->resource.start - window->offset > region->start)
 			continue;
-		if (window->resource.end < region->end)
+		if (window->resource.end - window->offset < region->end)
 			continue;
 		offset = window->offset;
 		break;
diff --git a/arch/ia64/sn/kernel/setup.c b/arch/ia64/sn/kernel/setup.c
index a852077940fe03..f0306b516afba5 100644
--- a/arch/ia64/sn/kernel/setup.c
+++ b/arch/ia64/sn/kernel/setup.c
@@ -67,15 +67,27 @@ extern void snidle(int);
 extern unsigned char acpi_kbd_controller_present;
 
 unsigned long sn_rtc_cycles_per_second;
-
 EXPORT_SYMBOL(sn_rtc_cycles_per_second);
 
+DEFINE_PER_CPU(struct sn_hub_info_s, __sn_hub_info);
+EXPORT_PER_CPU_SYMBOL(__sn_hub_info);
+
 partid_t sn_partid = -1;
 EXPORT_SYMBOL(sn_partid);
 char sn_system_serial_number_string[128];
 EXPORT_SYMBOL(sn_system_serial_number_string);
 u64 sn_partition_serial_number;
 EXPORT_SYMBOL(sn_partition_serial_number);
+u8 sn_partition_id;
+EXPORT_SYMBOL(sn_partition_id);
+u8 sn_system_size;
+EXPORT_SYMBOL(sn_system_size);
+u8 sn_sharing_domain_size;
+EXPORT_SYMBOL(sn_sharing_domain_size);
+u8 sn_coherency_id;
+EXPORT_SYMBOL(sn_coherency_id);
+u8 sn_region_size;
+EXPORT_SYMBOL(sn_region_size);
 
 short physical_node_map[MAX_PHYSNODE_ID];
 
@@ -232,7 +244,7 @@ static void __init sn_check_for_wars(void)
 	} else {
 		for_each_online_node(cnode) {
 			if (is_shub_1_1(cnodeid_to_nasid(cnode)))
-				shub_1_1_found = 1;
+				sn_hub_info->shub_1_1_found = 1;
 		}
 	}
 }
@@ -424,16 +436,14 @@ void __init sn_cpu_init(void)
 	int slice;
 	int cnode;
 	int i;
-	u64 shubtype, nasid_bitmask, nasid_shift;
 	static int wars_have_been_checked;
 
 	memset(pda, 0, sizeof(pda));
-	if (ia64_sn_get_hub_info(0, &shubtype, &nasid_bitmask, &nasid_shift))
+	if (ia64_sn_get_sn_info(0, &sn_hub_info->shub2, &sn_hub_info->nasid_bitmask, &sn_hub_info->nasid_shift,
+				&sn_system_size, &sn_sharing_domain_size, &sn_partition_id,
+				&sn_coherency_id, &sn_region_size))
 		BUG();
-	pda->shub2 = (u8)shubtype;
-	pda->nasid_bitmask = (u16)nasid_bitmask;
-	pda->nasid_shift = (u8)nasid_shift;
-	pda->as_shift = pda->nasid_shift - 2;
+	sn_hub_info->as_shift = sn_hub_info->nasid_shift - 2;
 
 	/*
 	 * The boot cpu makes this call again after platform initialization is
@@ -482,7 +492,7 @@ void __init sn_cpu_init(void)
 		sn_check_for_wars();
 		wars_have_been_checked = 1;
 	}
-	pda->shub_1_1_found = shub_1_1_found;
+	sn_hub_info->shub_1_1_found = shub_1_1_found;
 
 	/*
 	 * Set up addresses of PIO/MEM write status registers.
diff --git a/include/asm-ia64/sn/addrs.h b/include/asm-ia64/sn/addrs.h
index 3ce60976d474b8..c916bd22767a14 100644
--- a/include/asm-ia64/sn/addrs.h
+++ b/include/asm-ia64/sn/addrs.h
@@ -11,6 +11,7 @@
 
 #include <asm/percpu.h>
 #include <asm/sn/types.h>
+#include <asm/sn/arch.h>
 #include <asm/sn/pda.h>
 
 /*
@@ -57,9 +58,9 @@
 /*
  * Define basic shift & mask constants for manipulating NASIDs and AS values.
  */
-#define NASID_BITMASK		(pda->nasid_bitmask)
-#define NASID_SHIFT		(pda->nasid_shift)
-#define AS_SHIFT		(pda->as_shift)
+#define NASID_BITMASK		(sn_hub_info->nasid_bitmask)
+#define NASID_SHIFT		(sn_hub_info->nasid_shift)
+#define AS_SHIFT		(sn_hub_info->as_shift)
 #define AS_BITMASK		0x3UL
 
 #define NASID_MASK              ((u64)NASID_BITMASK << NASID_SHIFT)
diff --git a/include/asm-ia64/sn/arch.h b/include/asm-ia64/sn/arch.h
index bfc922a0ab7179..7c349f07916a1b 100644
--- a/include/asm-ia64/sn/arch.h
+++ b/include/asm-ia64/sn/arch.h
@@ -12,10 +12,34 @@
 #define _ASM_IA64_SN_ARCH_H
 
 #include <asm/types.h>
+#include <asm/percpu.h>
 #include <asm/sn/types.h>
 #include <asm/sn/sn_cpuid.h>
 
 /*
+ * The following defines attributes of the HUB chip. These attributes are
+ * frequently referenced. They are kept in the per-cpu data areas of each cpu.
+ * They are kept together in a struct to minimize cache misses.
+ */
+struct sn_hub_info_s {
+	u8 shub2;
+	u8 nasid_shift;
+	u8 as_shift;
+	u8 shub_1_1_found;
+	u16 nasid_bitmask;
+};
+DECLARE_PER_CPU(struct sn_hub_info_s, __sn_hub_info);
+#define sn_hub_info 	(&__get_cpu_var(__sn_hub_info))
+#define is_shub2()	(sn_hub_info->shub2)
+#define is_shub1()	(sn_hub_info->shub2 == 0)
+
+/*
+ * Use this macro to test if shub 1.1 wars should be enabled
+ */
+#define enable_shub_wars_1_1()	(sn_hub_info->shub_1_1_found)
+
+
+/*
  * This is the maximum number of nodes that can be part of a kernel.
  * Effectively, it's the maximum number of compact node ids (cnodeid_t).
  * This is not necessarily the same as MAX_NASIDS.
diff --git a/include/asm-ia64/sn/pda.h b/include/asm-ia64/sn/pda.h
index 6465e8ab2bccd3..e940d3647c8043 100644
--- a/include/asm-ia64/sn/pda.h
+++ b/include/asm-ia64/sn/pda.h
@@ -37,11 +37,6 @@ typedef struct pda_s {
 	 * Support for SN LEDs
 	 */
 	volatile short	*led_address;
-	u16		nasid_bitmask;
-	u8		shub2;
-	u8		nasid_shift;
-	u8		as_shift;
-	u8		shub_1_1_found;
 	u8		led_state;
 	u8		hb_state;	/* supports blinking heartbeat leds */
 	unsigned int	hb_count;
@@ -53,8 +48,6 @@ typedef struct pda_s {
 	unsigned long pio_write_status_val;
 	volatile unsigned long *pio_shub_war_cam_addr;
 
-	struct bteinfo_s *cpu_bte_if[BTES_PER_NODE];	/* cpu interface order */
-
 	unsigned long	sn_soft_irr[4];
 	unsigned long	sn_in_service_ivecs[4];
 	short		cnodeid_to_nasid_table[MAX_NUMNODES];
@@ -84,12 +77,4 @@ DECLARE_PER_CPU(struct pda_s, pda_percpu);
 
 #define pdacpu(cpu)	(&per_cpu(pda_percpu, cpu))
 
-/*
- * Use this macro to test if shub 1.1 wars should be enabled
- */
-#define enable_shub_wars_1_1()	(pda->shub_1_1_found)
-
-#define is_shub2()	(pda->shub2)
-#define is_shub1()	(pda->shub2 == 0)
-
 #endif /* _ASM_IA64_SN_PDA_H */
diff --git a/include/asm-ia64/sn/sn_cpuid.h b/include/asm-ia64/sn/sn_cpuid.h
index 749f45042cdcc1..685435af170d8d 100644
--- a/include/asm-ia64/sn/sn_cpuid.h
+++ b/include/asm-ia64/sn/sn_cpuid.h
@@ -135,9 +135,10 @@ extern int nasid_slice_to_cpuid(int, int);
 #define nasid_to_cnodeid(nasid)		(physical_node_map[nasid])
 
 /*
- * partition_coherence_id - cget the coherence ID of the current partition
+ * partition_coherence_id - get the coherence ID of the current partition
  */
-#define partition_coherence_id()	(get_nasid() >> 9)
+extern u8 sn_coherency_id;
+#define partition_coherence_id()	(sn_coherency_id)
 
 #endif /* _ASM_IA64_SN_SN_CPUID_H */
 
diff --git a/include/asm-ia64/sn/sn_sal.h b/include/asm-ia64/sn/sn_sal.h
index 4ecc05e8ad5772..88c31b53dc0974 100644
--- a/include/asm-ia64/sn/sn_sal.h
+++ b/include/asm-ia64/sn/sn_sal.h
@@ -18,6 +18,7 @@
 #include <asm/sn/arch.h>
 #include <asm/sn/geo.h>
 #include <asm/sn/nodepda.h>
+#include <asm/sn/shub_mmr.h>
 
 // SGI Specific Calls
 #define  SN_SAL_POD_MODE                           0x02000001
@@ -34,7 +35,7 @@
 #define  SN_SAL_PRINT_ERROR			   0x02000012
 #define  SN_SAL_SET_ERROR_HANDLING_FEATURES	   0x0200001a	// reentrant
 #define  SN_SAL_GET_FIT_COMPT			   0x0200001b	// reentrant
-#define  SN_SAL_GET_HUB_INFO                       0x0200001c
+#define  SN_SAL_GET_SN_INFO                        0x0200001c
 #define  SN_SAL_GET_SAPIC_INFO                     0x0200001d
 #define  SN_SAL_CONSOLE_PUTC                       0x02000021
 #define  SN_SAL_CONSOLE_GETC                       0x02000022
@@ -935,15 +936,24 @@ ia64_sn_get_sapic_info(int sapicid, int *nasid, int *subnode, int *slice)
 /*
  * Returns information about the HUB/SHUB.
  *  In:
- *	arg0 - SN_SAL_GET_HUB_INFO
+ *	arg0 - SN_SAL_GET_SN_INFO
  * 	arg1 - 0 (other values reserved for future use)
  *  Out:
- *	v0 - shub type (0=shub1, 1=shub2)
- *	v1 - masid mask (ex., 0x7ff for 11 bit nasid)
- *	v2 - bit position of low nasid bit
+ *	v0 
+ *		[7:0]   - shub type (0=shub1, 1=shub2)
+ *		[15:8]  - Log2 max number of nodes in entire system (includes
+ *			  C-bricks, I-bricks, etc)
+ *		[23:16] - Log2 of nodes per sharing domain			 
+ * 		[31:24] - partition ID
+ * 		[39:32] - coherency_id
+ * 		[47:40] - regionsize
+ *	v1 
+ *		[15:0]  - nasid mask (ex., 0x7ff for 11 bit nasid)
+ *	 	[23:15] - bit position of low nasid bit
  */
 static inline u64
-ia64_sn_get_hub_info(int fc, u64 *arg1, u64 *arg2, u64 *arg3)
+ia64_sn_get_sn_info(int fc, u8 *shubtype, u16 *nasid_bitmask, u8 *nasid_shift, 
+		u8 *systemsize, u8 *sharing_domain_size, u8 *partid, u8 *coher, u8 *reg)
 {
 	struct ia64_sal_retval ret_stuff;
 
@@ -951,13 +961,22 @@ ia64_sn_get_hub_info(int fc, u64 *arg1, u64 *arg2, u64 *arg3)
 	ret_stuff.v0 = 0;
 	ret_stuff.v1 = 0;
 	ret_stuff.v2 = 0;
-	SAL_CALL_NOLOCK(ret_stuff, SN_SAL_GET_HUB_INFO, fc, 0, 0, 0, 0, 0, 0);
+	SAL_CALL_NOLOCK(ret_stuff, SN_SAL_GET_SN_INFO, fc, 0, 0, 0, 0, 0, 0);
 
 /***** BEGIN HACK - temp til old proms no longer supported ********/
 	if (ret_stuff.status == SALRET_NOT_IMPLEMENTED) {
-		if (arg1) *arg1 = 0;
-		if (arg2) *arg2 = 0x7ff;
-		if (arg3) *arg3 = 38;
+		int nasid = get_sapicid() & 0xfff;;
+#define SH_SHUB_ID_NODES_PER_BIT_MASK 0x001f000000000000UL                                               
+#define SH_SHUB_ID_NODES_PER_BIT_SHFT 48                                                               
+		if (shubtype) *shubtype = 0;
+		if (nasid_bitmask) *nasid_bitmask = 0x7ff;
+		if (nasid_shift) *nasid_shift = 38;
+		if (systemsize) *systemsize = 11;
+		if (sharing_domain_size) *sharing_domain_size = 9;
+		if (partid) *partid = ia64_sn_sysctl_partition_get(nasid);
+		if (coher) *coher = nasid >> 9;
+		if (reg) *reg = (HUB_L((u64 *) LOCAL_MMR_ADDR(SH1_SHUB_ID)) & SH_SHUB_ID_NODES_PER_BIT_MASK) >>
+			SH_SHUB_ID_NODES_PER_BIT_SHFT;
 		return 0;
 	}
 /***** END HACK *******/
@@ -965,9 +984,14 @@ ia64_sn_get_hub_info(int fc, u64 *arg1, u64 *arg2, u64 *arg3)
 	if (ret_stuff.status < 0)
 		return ret_stuff.status;
 
-	if (arg1) *arg1 = ret_stuff.v0;
-	if (arg2) *arg2 = ret_stuff.v1;
-	if (arg3) *arg3 = ret_stuff.v2;
+	if (shubtype) *shubtype = ret_stuff.v0 & 0xff;
+	if (systemsize) *systemsize = (ret_stuff.v0 >> 8) & 0xff;
+	if (sharing_domain_size) *sharing_domain_size = (ret_stuff.v0 >> 16) & 0xff;
+	if (partid) *partid = (ret_stuff.v0 >> 24) & 0xff;
+	if (coher) *coher = (ret_stuff.v0 >> 32) & 0xff;
+	if (reg) *reg = (ret_stuff.v0 >> 40) & 0xff;
+	if (nasid_bitmask) *nasid_bitmask = (ret_stuff.v1 & 0xffff);
+	if (nasid_shift) *nasid_shift = (ret_stuff.v1 >> 16) & 0xff;
 	return 0;
 }