Merge branch 'for-2.6.23' of master.kernel.org:/pub/scm/linux/kernel/git/arnd/cell-2.6

* 'for-2.6.23' of master.kernel.org:/pub/scm/linux/kernel/git/arnd/cell-2.6: (37 commits)
  [CELL] spufs: rework list management and associated locking
  [CELL] oprofile: add support to OProfile for profiling CELL BE SPUs
  [CELL] oprofile: enable SPU switch notification to detect currently active SPU tasks
  [CELL] spu_base: locking cleanup
  [CELL] cell: indexing of SPUs based on firmware vicinity properties
  [CELL] spufs: integration of SPE affinity with the scheduller
  [CELL] cell: add placement computation for scheduling of affinity contexts
  [CELL] spufs: extension of spu_create to support affinity definition
  [CELL] cell: add hardcoded spu vicinity information for QS20
  [CELL] cell: add vicinity information on spus
  [CELL] cell: add per BE structure with info about its SPUs
  [CELL] spufs: use find_first_bit() instead of sched_find_first_bit()
  [CELL] spufs: remove unused file argument from spufs_run_spu()
  [CELL] spufs: change decrementer restore timing
  [CELL] spufs: dont halt decrementer at restore step 47
  [CELL] spufs: limit saving MFC_CNTL bits
  [CELL] spufs: fix read and write for decr_status file
  [CELL] spufs: fix decr_status meanings
  [CELL] spufs: remove needless context save/restore code
  [CELL] spufs: fix array size of channel index
  ...

arch/powerpc/configs/cell_defconfig

@@ -1455,7 +1455,8 @@ CONFIG_HAS_DMA=y
 # Instrumentation Support
 #
 CONFIG_PROFILING=y
-CONFIG_OPROFILE=y
+CONFIG_OPROFILE=m
+CONFIG_OPROFILE_CELL=y
 # CONFIG_KPROBES is not set
 
 #

arch/powerpc/kernel/crash.c

@@ -219,6 +219,72 @@ void crash_kexec_secondary(struct pt_regs *regs)
 	cpus_in_sr = CPU_MASK_NONE;
 }
 #endif
+#ifdef CONFIG_SPU_BASE
+
+#include <asm/spu.h>
+#include <asm/spu_priv1.h>
+
+struct crash_spu_info {
+	struct spu *spu;
+	u32 saved_spu_runcntl_RW;
+	u32 saved_spu_status_R;
+	u32 saved_spu_npc_RW;
+	u64 saved_mfc_sr1_RW;
+	u64 saved_mfc_dar;
+	u64 saved_mfc_dsisr;
+};
+
+#define CRASH_NUM_SPUS	16	/* Enough for current hardware */
+static struct crash_spu_info crash_spu_info[CRASH_NUM_SPUS];
+
+static void crash_kexec_stop_spus(void)
+{
+	struct spu *spu;
+	int i;
+	u64 tmp;
+
+	for (i = 0; i < CRASH_NUM_SPUS; i++) {
+		if (!crash_spu_info[i].spu)
+			continue;
+
+		spu = crash_spu_info[i].spu;
+
+		crash_spu_info[i].saved_spu_runcntl_RW =
+			in_be32(&spu->problem->spu_runcntl_RW);
+		crash_spu_info[i].saved_spu_status_R =
+			in_be32(&spu->problem->spu_status_R);
+		crash_spu_info[i].saved_spu_npc_RW =
+			in_be32(&spu->problem->spu_npc_RW);
+
+		crash_spu_info[i].saved_mfc_dar    = spu_mfc_dar_get(spu);
+		crash_spu_info[i].saved_mfc_dsisr  = spu_mfc_dsisr_get(spu);
+		tmp = spu_mfc_sr1_get(spu);
+		crash_spu_info[i].saved_mfc_sr1_RW = tmp;
+
+		tmp &= ~MFC_STATE1_MASTER_RUN_CONTROL_MASK;
+		spu_mfc_sr1_set(spu, tmp);
+
+		__delay(200);
+	}
+}
+
+void crash_register_spus(struct list_head *list)
+{
+	struct spu *spu;
+
+	list_for_each_entry(spu, list, full_list) {
+		if (WARN_ON(spu->number >= CRASH_NUM_SPUS))
+			continue;
+
+		crash_spu_info[spu->number].spu = spu;
+	}
+}
+
+#else
+static inline void crash_kexec_stop_spus(void)
+{
+}
+#endif /* CONFIG_SPU_BASE */
 
 void default_machine_crash_shutdown(struct pt_regs *regs)
 {
@@ -254,6 +320,7 @@ void default_machine_crash_shutdown(struct pt_regs *regs)
 	crash_save_cpu(regs, crashing_cpu);
 	crash_kexec_prepare_cpus(crashing_cpu);
 	cpu_set(crashing_cpu, cpus_in_crash);
+	crash_kexec_stop_spus();
 	if (ppc_md.kexec_cpu_down)
 		ppc_md.kexec_cpu_down(1, 0);
 }

arch/powerpc/kernel/time.c

@@ -122,6 +122,7 @@ extern struct timezone sys_tz;
 static long timezone_offset;
 
 unsigned long ppc_proc_freq;
+EXPORT_SYMBOL(ppc_proc_freq);
 unsigned long ppc_tb_freq;
 
 static u64 tb_last_jiffy __cacheline_aligned_in_smp;

arch/powerpc/oprofile/Kconfig

@@ -15,3 +15,10 @@ config OPROFILE
 
 	  If unsure, say N.
 
+config OPROFILE_CELL
+	bool "OProfile for Cell Broadband Engine"
+	depends on (SPU_FS = y && OPROFILE = m) || (SPU_FS = y && OPROFILE = y) || (SPU_FS = m && OPROFILE = m)
+	default y
+	help
+	  Profiling of Cell BE SPUs requires special support enabled
+	  by this option.

arch/powerpc/oprofile/Makefile

@@ -11,7 +11,9 @@ DRIVER_OBJS := $(addprefix ../../../drivers/oprofile/, \
 		timer_int.o )
 
 oprofile-y := $(DRIVER_OBJS) common.o backtrace.o
-oprofile-$(CONFIG_PPC_CELL_NATIVE) += op_model_cell.o
+oprofile-$(CONFIG_OPROFILE_CELL) += op_model_cell.o \
+		cell/spu_profiler.o cell/vma_map.o \
+		cell/spu_task_sync.o
 oprofile-$(CONFIG_PPC64) += op_model_rs64.o op_model_power4.o op_model_pa6t.o
 oprofile-$(CONFIG_FSL_BOOKE) += op_model_fsl_booke.o
 oprofile-$(CONFIG_6xx) += op_model_7450.o

arch/powerpc/oprofile/cell/pr_util.h

+ /*
+ * Cell Broadband Engine OProfile Support
+ *
+ * (C) Copyright IBM Corporation 2006
+ *
+ * Author: Maynard Johnson <maynardj@us.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#ifndef PR_UTIL_H
+#define PR_UTIL_H
+
+#include <linux/cpumask.h>
+#include <linux/oprofile.h>
+#include <asm/cell-pmu.h>
+#include <asm/spu.h>
+
+#include "../../platforms/cell/cbe_regs.h"
+
+/* Defines used for sync_start */
+#define SKIP_GENERIC_SYNC 0
+#define SYNC_START_ERROR -1
+#define DO_GENERIC_SYNC 1
+
+struct spu_overlay_info {	/* map of sections within an SPU overlay */
+	unsigned int vma;	/* SPU virtual memory address from elf */
+	unsigned int size;	/* size of section from elf */
+	unsigned int offset;	/* offset of section into elf file */
+	unsigned int buf;
+};
+
+struct vma_to_fileoffset_map {	/* map of sections within an SPU program */
+	struct vma_to_fileoffset_map *next;	/* list pointer */
+	unsigned int vma;	/* SPU virtual memory address from elf */
+	unsigned int size;	/* size of section from elf */
+	unsigned int offset;	/* offset of section into elf file */
+	unsigned int guard_ptr;
+	unsigned int guard_val;
+        /*
+	 * The guard pointer is an entry in the _ovly_buf_table,
+	 * computed using ovly.buf as the index into the table.  Since
+	 * ovly.buf values begin at '1' to reference the first (or 0th)
+	 * entry in the _ovly_buf_table, the computation subtracts 1
+	 * from ovly.buf.
+	 * The guard value is stored in the _ovly_buf_table entry and
+	 * is an index (starting at 1) back to the _ovly_table entry
+	 * that is pointing at this _ovly_buf_table entry.  So, for
+	 * example, for an overlay scenario with one overlay segment
+	 * and two overlay sections:
+	 *      - Section 1 points to the first entry of the
+	 *        _ovly_buf_table, which contains a guard value
+	 *        of '1', referencing the first (index=0) entry of
+	 *        _ovly_table.
+	 *      - Section 2 points to the second entry of the
+	 *        _ovly_buf_table, which contains a guard value
+	 *        of '2', referencing the second (index=1) entry of
+	 *        _ovly_table.
+	 */
+
+};
+
+/* The three functions below are for maintaining and accessing
+ * the vma-to-fileoffset map.
+ */
+struct vma_to_fileoffset_map *create_vma_map(const struct spu *spu,
+					     u64 objectid);
+unsigned int vma_map_lookup(struct vma_to_fileoffset_map *map,
+			    unsigned int vma, const struct spu *aSpu,
+			    int *grd_val);
+void vma_map_free(struct vma_to_fileoffset_map *map);
+
+/*
+ * Entry point for SPU profiling.
+ * cycles_reset is the SPU_CYCLES count value specified by the user.
+ */
+int start_spu_profiling(unsigned int cycles_reset);
+
+void stop_spu_profiling(void);
+
+
+/* add the necessary profiling hooks */
+int spu_sync_start(void);
+
+/* remove the hooks */
+int spu_sync_stop(void);
+
+/* Record SPU program counter samples to the oprofile event buffer. */
+void spu_sync_buffer(int spu_num, unsigned int *samples,
+		     int num_samples);
+
+void set_spu_profiling_frequency(unsigned int freq_khz, unsigned int cycles_reset);
+
+#endif	  /* PR_UTIL_H */

arch/powerpc/oprofile/cell/spu_profiler.c

+/*
+ * Cell Broadband Engine OProfile Support
+ *
+ * (C) Copyright IBM Corporation 2006
+ *
+ * Authors: Maynard Johnson <maynardj@us.ibm.com>
+ *	    Carl Love <carll@us.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+#include <linux/hrtimer.h>
+#include <linux/smp.h>
+#include <linux/slab.h>
+#include <asm/cell-pmu.h>
+#include "pr_util.h"
+
+#define TRACE_ARRAY_SIZE 1024
+#define SCALE_SHIFT 14
+
+static u32 *samples;
+
+static int spu_prof_running;
+static unsigned int profiling_interval;
+
+#define NUM_SPU_BITS_TRBUF 16
+#define SPUS_PER_TB_ENTRY   4
+#define SPUS_PER_NODE	     8
+
+#define SPU_PC_MASK	     0xFFFF
+
+static DEFINE_SPINLOCK(sample_array_lock);
+unsigned long sample_array_lock_flags;
+
+void set_spu_profiling_frequency(unsigned int freq_khz, unsigned int cycles_reset)
+{
+	unsigned long ns_per_cyc;
+
+	if (!freq_khz)
+		freq_khz = ppc_proc_freq/1000;
+
+	/* To calculate a timeout in nanoseconds, the basic
+	 * formula is ns = cycles_reset * (NSEC_PER_SEC / cpu frequency).
+	 * To avoid floating point math, we use the scale math
+	 * technique as described in linux/jiffies.h.  We use
+	 * a scale factor of SCALE_SHIFT, which provides 4 decimal places
+	 * of precision.  This is close enough for the purpose at hand.
+	 *
+	 * The value of the timeout should be small enough that the hw
+	 * trace buffer will not get more then about 1/3 full for the
+	 * maximum user specified (the LFSR value) hw sampling frequency.
+	 * This is to ensure the trace buffer will never fill even if the
+	 * kernel thread scheduling varies under a heavy system load.
+	 */
+
+	ns_per_cyc = (USEC_PER_SEC << SCALE_SHIFT)/freq_khz;
+	profiling_interval = (ns_per_cyc * cycles_reset) >> SCALE_SHIFT;
+
+}
+
+/*
+ * Extract SPU PC from trace buffer entry
+ */
+static void spu_pc_extract(int cpu, int entry)
+{
+	/* the trace buffer is 128 bits */
+	u64 trace_buffer[2];
+	u64 spu_mask;
+	int spu;
+
+	spu_mask = SPU_PC_MASK;
+
+	/* Each SPU PC is 16 bits; hence, four spus in each of
+	 * the two 64-bit buffer entries that make up the
+	 * 128-bit trace_buffer entry.	Process two 64-bit values
+	 * simultaneously.
+	 * trace[0] SPU PC contents are: 0 1 2 3
+	 * trace[1] SPU PC contents are: 4 5 6 7
+	 */
+
+	cbe_read_trace_buffer(cpu, trace_buffer);
+
+	for (spu = SPUS_PER_TB_ENTRY-1; spu >= 0; spu--) {
+		/* spu PC trace entry is upper 16 bits of the
+		 * 18 bit SPU program counter
+		 */
+		samples[spu * TRACE_ARRAY_SIZE + entry]
+			= (spu_mask & trace_buffer[0]) << 2;
+		samples[(spu + SPUS_PER_TB_ENTRY) * TRACE_ARRAY_SIZE + entry]
+			= (spu_mask & trace_buffer[1]) << 2;
+
+		trace_buffer[0] = trace_buffer[0] >> NUM_SPU_BITS_TRBUF;
+		trace_buffer[1] = trace_buffer[1] >> NUM_SPU_BITS_TRBUF;
+	}
+}
+
+static int cell_spu_pc_collection(int cpu)
+{
+	u32 trace_addr;
+	int entry;
+
+	/* process the collected SPU PC for the node */
+
+	entry = 0;
+
+	trace_addr = cbe_read_pm(cpu, trace_address);
+	while (!(trace_addr & CBE_PM_TRACE_BUF_EMPTY)) {
+		/* there is data in the trace buffer to process */
+		spu_pc_extract(cpu, entry);
+
+		entry++;
+
+		if (entry >= TRACE_ARRAY_SIZE)
+			/* spu_samples is full */
+			break;
+
+		trace_addr = cbe_read_pm(cpu, trace_address);
+	}
+
+	return entry;
+}
+
+
+static enum hrtimer_restart profile_spus(struct hrtimer *timer)
+{
+	ktime_t kt;
+	int cpu, node, k, num_samples, spu_num;
+
+	if (!spu_prof_running)
+		goto stop;
+
+	for_each_online_cpu(cpu) {
+		if (cbe_get_hw_thread_id(cpu))
+			continue;
+
+		node = cbe_cpu_to_node(cpu);
+
+		/* There should only be one kernel thread at a time processing
+		 * the samples.	 In the very unlikely case that the processing
+		 * is taking a very long time and multiple kernel threads are
+		 * started to process the samples.  Make sure only one kernel
+		 * thread is working on the samples array at a time.  The
+		 * sample array must be loaded and then processed for a given
+		 * cpu.	 The sample array is not per cpu.
+		 */
+		spin_lock_irqsave(&sample_array_lock,
+				  sample_array_lock_flags);
+		num_samples = cell_spu_pc_collection(cpu);
+
+		if (num_samples == 0) {
+			spin_unlock_irqrestore(&sample_array_lock,
+					       sample_array_lock_flags);
+			continue;
+		}
+
+		for (k = 0; k < SPUS_PER_NODE; k++) {
+			spu_num = k + (node * SPUS_PER_NODE);
+			spu_sync_buffer(spu_num,
+					samples + (k * TRACE_ARRAY_SIZE),
+					num_samples);
+		}
+
+		spin_unlock_irqrestore(&sample_array_lock,
+				       sample_array_lock_flags);
+
+	}
+	smp_wmb();	/* insure spu event buffer updates are written */
+			/* don't want events intermingled... */
+
+	kt = ktime_set(0, profiling_interval);
+	if (!spu_prof_running)
+		goto stop;
+	hrtimer_forward(timer, timer->base->get_time(), kt);
+	return HRTIMER_RESTART;
+
+ stop:
+	printk(KERN_INFO "SPU_PROF: spu-prof timer ending\n");
+	return HRTIMER_NORESTART;
+}
+
+static struct hrtimer timer;
+/*
+ * Entry point for SPU profiling.
+ * NOTE:  SPU profiling is done system-wide, not per-CPU.
+ *
+ * cycles_reset is the count value specified by the user when
+ * setting up OProfile to count SPU_CYCLES.
+ */
+int start_spu_profiling(unsigned int cycles_reset)
+{
+	ktime_t kt;
+
+	pr_debug("timer resolution: %lu\n", TICK_NSEC);
+	kt = ktime_set(0, profiling_interval);
+	hrtimer_init(&timer, CLOCK_MONOTONIC, HRTIMER_MODE_REL);
+	timer.expires = kt;
+	timer.function = profile_spus;
+
+	/* Allocate arrays for collecting SPU PC samples */
+	samples = kzalloc(SPUS_PER_NODE *
+			  TRACE_ARRAY_SIZE * sizeof(u32), GFP_KERNEL);
+
+	if (!samples)
+		return -ENOMEM;
+
+	spu_prof_running = 1;
+	hrtimer_start(&timer, kt, HRTIMER_MODE_REL);
+
+	return 0;
+}
+
+void stop_spu_profiling(void)
+{
+	spu_prof_running = 0;
+	hrtimer_cancel(&timer);
+	kfree(samples);
+	pr_debug("SPU_PROF: stop_spu_profiling issued\n");
+}

arch/powerpc/oprofile/cell/vma_map.c

+/*
+ * Cell Broadband Engine OProfile Support
+ *
+ * (C) Copyright IBM Corporation 2006
+ *
+ * Author: Maynard Johnson <maynardj@us.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version
+ * 2 of the License, or (at your option) any later version.
+ */
+
+/* The code in this source file is responsible for generating
+ * vma-to-fileOffset maps for both overlay and non-overlay SPU
+ * applications.
+ */
+
+#include <linux/mm.h>
+#include <linux/string.h>
+#include <linux/uaccess.h>
+#include <linux/elf.h>
+#include "pr_util.h"
+
+
+void vma_map_free(struct vma_to_fileoffset_map *map)
+{
+	while (map) {
+		struct vma_to_fileoffset_map *next = map->next;
+		kfree(map);
+		map = next;
+	}
+}
+
+unsigned int
+vma_map_lookup(struct vma_to_fileoffset_map *map, unsigned int vma,
+	       const struct spu *aSpu, int *grd_val)
+{
+	/*
+	 * Default the offset to the physical address + a flag value.
+	 * Addresses of dynamically generated code can't be found in the vma
+	 * map.  For those addresses the flagged value will be sent on to
+	 * the user space tools so they can be reported rather than just
+	 * thrown away.
+	 */
+	u32 offset = 0x10000000 + vma;
+	u32 ovly_grd;
+
+	for (; map; map = map->next) {
+		if (vma < map->vma || vma >= map->vma + map->size)
+			continue;
+
+		if (map->guard_ptr) {
+			ovly_grd = *(u32 *)(aSpu->local_store + map->guard_ptr);
+			if (ovly_grd != map->guard_val)
+				continue;
+			*grd_val = ovly_grd;
+		}
+		offset = vma - map->vma + map->offset;
+		break;
+	}
+
+	return offset;
+}
+
+static struct vma_to_fileoffset_map *
+vma_map_add(struct vma_to_fileoffset_map *map, unsigned int vma,
+	    unsigned int size, unsigned int offset, unsigned int guard_ptr,
+	    unsigned int guard_val)
+{
+	struct vma_to_fileoffset_map *new =
+		kzalloc(sizeof(struct vma_to_fileoffset_map), GFP_KERNEL);
+	if (!new) {
+		printk(KERN_ERR "SPU_PROF: %s, line %d: malloc failed\n",
+		       __FUNCTION__, __LINE__);
+		vma_map_free(map);
+		return NULL;
+	}
+
+	new->next = map;
+	new->vma = vma;
+	new->size = size;
+	new->offset = offset;
+	new->guard_ptr = guard_ptr;
+	new->guard_val = guard_val;
+
+	return new;
+}
+
+
+/* Parse SPE ELF header and generate a list of vma_maps.
+ * A pointer to the first vma_map in the generated list
+ * of vma_maps is returned.  */
+struct vma_to_fileoffset_map *create_vma_map(const struct spu *aSpu,
+					     unsigned long spu_elf_start)
+{
+	static const unsigned char expected[EI_PAD] = {
+		[EI_MAG0] = ELFMAG0,
+		[EI_MAG1] = ELFMAG1,
+		[EI_MAG2] = ELFMAG2,
+		[EI_MAG3] = ELFMAG3,
+		[EI_CLASS] = ELFCLASS32,
+		[EI_DATA] = ELFDATA2MSB,
+		[EI_VERSION] = EV_CURRENT,
+		[EI_OSABI] = ELFOSABI_NONE
+	};
+
+	int grd_val;
+	struct vma_to_fileoffset_map *map = NULL;
+	struct spu_overlay_info ovly;
+	unsigned int overlay_tbl_offset = -1;
+	unsigned long phdr_start, shdr_start;
+	Elf32_Ehdr ehdr;
+	Elf32_Phdr phdr;
+	Elf32_Shdr shdr, shdr_str;
+	Elf32_Sym sym;
+	int i, j;
+	char name[32];
+
+	unsigned int ovly_table_sym = 0;
+	unsigned int ovly_buf_table_sym = 0;
+	unsigned int ovly_table_end_sym = 0;
+	unsigned int ovly_buf_table_end_sym = 0;
+	unsigned long ovly_table;
+	unsigned int n_ovlys;
+
+	/* Get and validate ELF header.	 */
+
+	if (copy_from_user(&ehdr, (void *) spu_elf_start, sizeof (ehdr)))
+		goto fail;
+
+	if (memcmp(ehdr.e_ident, expected, EI_PAD) != 0) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: Unexpected e_ident parsing SPU ELF\n",
+		       __FUNCTION__, __LINE__);
+		goto fail;
+	}
+	if (ehdr.e_machine != EM_SPU) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: Unexpected e_machine parsing SPU ELF\n",
+		       __FUNCTION__,  __LINE__);
+		goto fail;
+	}
+	if (ehdr.e_type != ET_EXEC) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: Unexpected e_type parsing SPU ELF\n",
+		       __FUNCTION__, __LINE__);
+		goto fail;
+	}
+	phdr_start = spu_elf_start + ehdr.e_phoff;
+	shdr_start = spu_elf_start + ehdr.e_shoff;
+
+	/* Traverse program headers.  */
+	for (i = 0; i < ehdr.e_phnum; i++) {
+		if (copy_from_user(&phdr,
+				   (void *) (phdr_start + i * sizeof(phdr)),
+				   sizeof(phdr)))
+			goto fail;
+
+		if (phdr.p_type != PT_LOAD)
+			continue;
+		if (phdr.p_flags & (1 << 27))
+			continue;
+
+		map = vma_map_add(map, phdr.p_vaddr, phdr.p_memsz,
+				  phdr.p_offset, 0, 0);
+		if (!map)
+			goto fail;
+	}
+
+	pr_debug("SPU_PROF: Created non-overlay maps\n");
+	/* Traverse section table and search for overlay-related symbols.  */
+	for (i = 0; i < ehdr.e_shnum; i++) {
+		if (copy_from_user(&shdr,
+				   (void *) (shdr_start + i * sizeof(shdr)),
+				   sizeof(shdr)))
+			goto fail;
+
+		if (shdr.sh_type != SHT_SYMTAB)
+			continue;
+		if (shdr.sh_entsize != sizeof (sym))
+			continue;
+
+		if (copy_from_user(&shdr_str,
+				   (void *) (shdr_start + shdr.sh_link *
+					     sizeof(shdr)),
+				   sizeof(shdr)))
+			goto fail;
+
+		if (shdr_str.sh_type != SHT_STRTAB)
+			goto fail;;
+
+		for (j = 0; j < shdr.sh_size / sizeof (sym); j++) {
+			if (copy_from_user(&sym, (void *) (spu_elf_start +
+						       shdr.sh_offset + j *
+							   sizeof (sym)),
+					   sizeof (sym)))
+				goto fail;
+
+			if (copy_from_user(name, (void *)
+					   (spu_elf_start + shdr_str.sh_offset +
+					    sym.st_name),
+					   20))
+				goto fail;
+
+			if (memcmp(name, "_ovly_table", 12) == 0)
+				ovly_table_sym = sym.st_value;
+			if (memcmp(name, "_ovly_buf_table", 16) == 0)
+				ovly_buf_table_sym = sym.st_value;
+			if (memcmp(name, "_ovly_table_end", 16) == 0)
+				ovly_table_end_sym = sym.st_value;
+			if (memcmp(name, "_ovly_buf_table_end", 20) == 0)
+				ovly_buf_table_end_sym = sym.st_value;
+		}
+	}
+
+	/* If we don't have overlays, we're done.  */
+	if (ovly_table_sym == 0 || ovly_buf_table_sym == 0
+	    || ovly_table_end_sym == 0 || ovly_buf_table_end_sym == 0) {
+		pr_debug("SPU_PROF: No overlay table found\n");
+		goto out;
+	} else {
+		pr_debug("SPU_PROF: Overlay table found\n");
+	}
+
+	/* The _ovly_table symbol represents a table with one entry
+	 * per overlay section.	 The _ovly_buf_table symbol represents
+	 * a table with one entry per overlay region.
+	 * The struct spu_overlay_info gives the structure of the _ovly_table
+	 * entries.  The structure of _ovly_table_buf is simply one
+	 * u32 word per entry.
+	 */
+	overlay_tbl_offset = vma_map_lookup(map, ovly_table_sym,
+					    aSpu, &grd_val);
+	if (overlay_tbl_offset < 0) {
+		printk(KERN_ERR "SPU_PROF: "
+		       "%s, line %d: Error finding SPU overlay table\n",
+		       __FUNCTION__, __LINE__);
+		goto fail;
+	}
+	ovly_table = spu_elf_start + overlay_tbl_offset;
+
+	n_ovlys = (ovly_table_end_sym -
+		   ovly_table_sym) / sizeof (ovly);
+
+	/* Traverse overlay table.  */
+	for (i = 0; i < n_ovlys; i++) {
+		if (copy_from_user(&ovly, (void *)
+				   (ovly_table + i * sizeof (ovly)),
+				   sizeof (ovly)))
+			goto fail;
+
+		/* The ovly.vma/size/offset arguments are analogous to the same
+		 * arguments used above for non-overlay maps.  The final two
+		 * args are referred to as the guard pointer and the guard
+		 * value.
+		 * The guard pointer is an entry in the _ovly_buf_table,
+		 * computed using ovly.buf as the index into the table.	 Since
+		 * ovly.buf values begin at '1' to reference the first (or 0th)
+		 * entry in the _ovly_buf_table, the computation subtracts 1
+		 * from ovly.buf.
+		 * The guard value is stored in the _ovly_buf_table entry and
+		 * is an index (starting at 1) back to the _ovly_table entry
+		 * that is pointing at this _ovly_buf_table entry.  So, for
+		 * example, for an overlay scenario with one overlay segment
+		 * and two overlay sections:
+		 *	- Section 1 points to the first entry of the
+		 *	  _ovly_buf_table, which contains a guard value
+		 *	  of '1', referencing the first (index=0) entry of
+		 *	  _ovly_table.
+		 *	- Section 2 points to the second entry of the
+		 *	  _ovly_buf_table, which contains a guard value
+		 *	  of '2', referencing the second (index=1) entry of
+		 *	  _ovly_table.
+		 */
+		map = vma_map_add(map, ovly.vma, ovly.size, ovly.offset,
+				  ovly_buf_table_sym + (ovly.buf-1) * 4, i+1);
+		if (!map)
+			goto fail;
+	}
+	goto out;
+
+ fail:
+	map = NULL;
+ out:
+	return map;
+}

arch/powerpc/oprofile/common.c

@@ -29,6 +29,8 @@ static struct op_powerpc_model *model;
 static struct op_counter_config ctr[OP_MAX_COUNTER];
 static struct op_system_config sys;
 
+static int op_per_cpu_rc;
+
 static void op_handle_interrupt(struct pt_regs *regs)
 {
 	model->handle_interrupt(regs, ctr);
@@ -36,25 +38,41 @@ static void op_handle_interrupt(struct pt_regs *regs)
 
 static void op_powerpc_cpu_setup(void *dummy)
 {
-	model->cpu_setup(ctr);
+	int ret;
+
+	ret = model->cpu_setup(ctr);
+
+	if (ret != 0)
+		op_per_cpu_rc = ret;
 }
 
 static int op_powerpc_setup(void)
 {
 	int err;
 
+	op_per_cpu_rc = 0;
+
 	/* Grab the hardware */
 	err = reserve_pmc_hardware(op_handle_interrupt);
 	if (err)
 		return err;
 
 	/* Pre-compute the values to stuff in the hardware registers.  */
-	model->reg_setup(ctr, &sys, model->num_counters);
+	op_per_cpu_rc = model->reg_setup(ctr, &sys, model->num_counters);
 
-	/* Configure the registers on all cpus.  */
+	if (op_per_cpu_rc)
+		goto out;
+
+	/* Configure the registers on all cpus.	 If an error occurs on one
+	 * of the cpus, op_per_cpu_rc will be set to the error */
 	on_each_cpu(op_powerpc_cpu_setup, NULL, 0, 1);
 
-	return 0;
+out:	if (op_per_cpu_rc) {
+		/* error on setup release the performance counter hardware */
+		release_pmc_hardware();
+	}
+
+	return op_per_cpu_rc;
 }
 
 static void op_powerpc_shutdown(void)
@@ -64,16 +82,29 @@ static void op_powerpc_shutdown(void)
 
 static void op_powerpc_cpu_start(void *dummy)
 {
-	model->start(ctr);
+	/* If any of the cpus have return an error, set the
+	 * global flag to the error so it can be returned
+	 * to the generic OProfile caller.
+	 */
+	int ret;
+
+	ret = model->start(ctr);
+	if (ret != 0)
+		op_per_cpu_rc = ret;
 }
 
 static int op_powerpc_start(void)
 {
+	op_per_cpu_rc = 0;
+
 	if (model->global_start)
-		model->global_start(ctr);
-	if (model->start)
+		return model->global_start(ctr);
+	if (model->start) {
 		on_each_cpu(op_powerpc_cpu_start, NULL, 0, 1);
-	return 0;
+		return op_per_cpu_rc;
+	}
+	return -EIO; /* No start function is defined for this
+			power architecture */
 }
 
 static inline void op_powerpc_cpu_stop(void *dummy)
@@ -147,11 +178,13 @@ int __init oprofile_arch_init(struct oprofile_operations *ops)
 
 	switch (cur_cpu_spec->oprofile_type) {
 #ifdef CONFIG_PPC64
-#ifdef CONFIG_PPC_CELL_NATIVE
+#ifdef CONFIG_OPROFILE_CELL
 		case PPC_OPROFILE_CELL:
 			if (firmware_has_feature(FW_FEATURE_LPAR))
 				return -ENODEV;
 			model = &op_model_cell;
+			ops->sync_start = model->sync_start;
+			ops->sync_stop = model->sync_stop;
 			break;
 #endif
 		case PPC_OPROFILE_RS64:

arch/powerpc/oprofile/op_model_7450.c

@@ -81,7 +81,7 @@ static void pmc_stop_ctrs(void)
 
 /* Configures the counters on this CPU based on the global
  * settings */
-static void fsl7450_cpu_setup(struct op_counter_config *ctr)
+static int fsl7450_cpu_setup(struct op_counter_config *ctr)
 {
 	/* freeze all counters */
 	pmc_stop_ctrs();
@@ -89,12 +89,14 @@ static void fsl7450_cpu_setup(struct op_counter_config *ctr)
 	mtspr(SPRN_MMCR0, mmcr0_val);
 	mtspr(SPRN_MMCR1, mmcr1_val);
 	mtspr(SPRN_MMCR2, mmcr2_val);
+
+	return 0;
 }
 
 #define NUM_CTRS 6
 
 /* Configures the global settings for the countes on all CPUs. */
-static void fsl7450_reg_setup(struct op_counter_config *ctr,
+static int fsl7450_reg_setup(struct op_counter_config *ctr,
 			     struct op_system_config *sys,
 			     int num_ctrs)
 {
@@ -126,10 +128,12 @@ static void fsl7450_reg_setup(struct op_counter_config *ctr,
 		| mmcr1_event6(ctr[5].event);
 
 	mmcr2_val = 0;
+
+	return 0;
 }
 
 /* Sets the counters on this CPU to the chosen values, and starts them */
-static void fsl7450_start(struct op_counter_config *ctr)
+static int fsl7450_start(struct op_counter_config *ctr)
 {
 	int i;
 
@@ -148,6 +152,8 @@ static void fsl7450_start(struct op_counter_config *ctr)
 	pmc_start_ctrs();
 
 	oprofile_running = 1;
+
+	return 0;
 }
 
 /* Stop the counters on this CPU */
@@ -193,7 +199,7 @@ static void fsl7450_handle_interrupt(struct pt_regs *regs,
 	/* The freeze bit was set by the interrupt. */
 	/* Clear the freeze bit, and reenable the interrupt.
 	 * The counters won't actually start until the rfi clears
-	 * the PMM bit */
+	 * the PM/M bit */
 	pmc_start_ctrs();
 }
 

arch/powerpc/oprofile/op_model_fsl_booke.c

@@ -244,7 +244,7 @@ static void dump_pmcs(void)
 			mfpmr(PMRN_PMLCA3), mfpmr(PMRN_PMLCB3));
 }
 
-static void fsl_booke_cpu_setup(struct op_counter_config *ctr)
+static int fsl_booke_cpu_setup(struct op_counter_config *ctr)
 {
 	int i;
 
@@ -258,9 +258,11 @@ static void fsl_booke_cpu_setup(struct op_counter_config *ctr)
 
 		set_pmc_user_kernel(i, ctr[i].user, ctr[i].kernel);
 	}
+
+	return 0;
 }
 
-static void fsl_booke_reg_setup(struct op_counter_config *ctr,
+static int fsl_booke_reg_setup(struct op_counter_config *ctr,
 			     struct op_system_config *sys,
 			     int num_ctrs)
 {
@@ -276,9 +278,10 @@ static void fsl_booke_reg_setup(struct op_counter_config *ctr,
 	for (i = 0; i < num_counters; ++i)
 		reset_value[i] = 0x80000000UL - ctr[i].count;
 
+	return 0;
 }
 
-static void fsl_booke_start(struct op_counter_config *ctr)
+static int fsl_booke_start(struct op_counter_config *ctr)
 {
 	int i;
 
@@ -308,6 +311,8 @@ static void fsl_booke_start(struct op_counter_config *ctr)
 
 	pr_debug("start on cpu %d, pmgc0 %x\n", smp_processor_id(),
 			mfpmr(PMRN_PMGC0));
+
+	return 0;
 }
 
 static void fsl_booke_stop(void)

arch/powerpc/oprofile/op_model_pa6t.c

@@ -89,7 +89,7 @@ static inline void ctr_write(unsigned int i, u64 val)
 
 
 /* precompute the values to stuff in the hardware registers */
-static void pa6t_reg_setup(struct op_counter_config *ctr,
+static int pa6t_reg_setup(struct op_counter_config *ctr,
 			   struct op_system_config *sys,
 			   int num_ctrs)
 {
@@ -135,10 +135,12 @@ static void pa6t_reg_setup(struct op_counter_config *ctr,
 		pr_debug("reset_value for pmc%u inited to 0x%lx\n",
 				 pmc, reset_value[pmc]);
 	}
+
+	return 0;
 }
 
 /* configure registers on this cpu */
-static void pa6t_cpu_setup(struct op_counter_config *ctr)
+static int pa6t_cpu_setup(struct op_counter_config *ctr)
 {
 	u64 mmcr0 = mmcr0_val;
 	u64 mmcr1 = mmcr1_val;
@@ -154,9 +156,11 @@ static void pa6t_cpu_setup(struct op_counter_config *ctr)
 		mfspr(SPRN_PA6T_MMCR0));
 	pr_debug("setup on cpu %d, mmcr1 %016lx\n", smp_processor_id(),
 		mfspr(SPRN_PA6T_MMCR1));
+
+	return 0;
 }
 
-static void pa6t_start(struct op_counter_config *ctr)
+static int pa6t_start(struct op_counter_config *ctr)
 {
 	int i;
 
@@ -174,6 +178,8 @@ static void pa6t_start(struct op_counter_config *ctr)
 	oprofile_running = 1;
 
 	pr_debug("start on cpu %d, mmcr0 %lx\n", smp_processor_id(), mmcr0);
+
+	return 0;
 }
 
 static void pa6t_stop(void)

arch/powerpc/oprofile/op_model_power4.c

@@ -32,7 +32,7 @@ static u32 mmcr0_val;
 static u64 mmcr1_val;
 static u64 mmcra_val;
 
-static void power4_reg_setup(struct op_counter_config *ctr,
+static int power4_reg_setup(struct op_counter_config *ctr,
 			     struct op_system_config *sys,
 			     int num_ctrs)
 {
@@ -60,6 +60,8 @@ static void power4_reg_setup(struct op_counter_config *ctr,
 		mmcr0_val &= ~MMCR0_PROBLEM_DISABLE;
 	else
 		mmcr0_val |= MMCR0_PROBLEM_DISABLE;
+
+	return 0;
 }
 
 extern void ppc64_enable_pmcs(void);
@@ -84,7 +86,7 @@ static inline int mmcra_must_set_sample(void)
 	return 0;
 }
 
-static void power4_cpu_setup(struct op_counter_config *ctr)
+static int power4_cpu_setup(struct op_counter_config *ctr)
 {
 	unsigned int mmcr0 = mmcr0_val;
 	unsigned long mmcra = mmcra_val;
@@ -111,9 +113,11 @@ static void power4_cpu_setup(struct op_counter_config *ctr)
 	    mfspr(SPRN_MMCR1));
 	dbg("setup on cpu %d, mmcra %lx\n", smp_processor_id(),
 	    mfspr(SPRN_MMCRA));
+
+	return 0;
 }
 
-static void power4_start(struct op_counter_config *ctr)
+static int power4_start(struct op_counter_config *ctr)
 {
 	int i;
 	unsigned int mmcr0;
@@ -148,6 +152,7 @@ static void power4_start(struct op_counter_config *ctr)
 	oprofile_running = 1;
 
 	dbg("start on cpu %d, mmcr0 %x\n", smp_processor_id(), mmcr0);
+	return 0;
 }
 
 static void power4_stop(void)

arch/powerpc/oprofile/op_model_rs64.c

@@ -88,7 +88,7 @@ static unsigned long reset_value[OP_MAX_COUNTER];
 
 static int num_counters;
 
-static void rs64_reg_setup(struct op_counter_config *ctr,
+static int rs64_reg_setup(struct op_counter_config *ctr,
 			   struct op_system_config *sys,
 			   int num_ctrs)
 {
@@ -100,9 +100,10 @@ static void rs64_reg_setup(struct op_counter_config *ctr,
 		reset_value[i] = 0x80000000UL - ctr[i].count;
 
 	/* XXX setup user and kernel profiling */
+	return 0;
 }
 
-static void rs64_cpu_setup(struct op_counter_config *ctr)
+static int rs64_cpu_setup(struct op_counter_config *ctr)
 {
 	unsigned int mmcr0;
 
@@ -125,9 +126,11 @@ static void rs64_cpu_setup(struct op_counter_config *ctr)
 	    mfspr(SPRN_MMCR0));
 	dbg("setup on cpu %d, mmcr1 %lx\n", smp_processor_id(),
 	    mfspr(SPRN_MMCR1));
+
+	return 0;
 }
 
-static void rs64_start(struct op_counter_config *ctr)
+static int rs64_start(struct op_counter_config *ctr)
 {
 	int i;
 	unsigned int mmcr0;
@@ -155,6 +158,7 @@ static void rs64_start(struct op_counter_config *ctr)
 	mtspr(SPRN_MMCR0, mmcr0);
 
 	dbg("start on cpu %d, mmcr0 %x\n", smp_processor_id(), mmcr0);
+	return 0;
 }
 
 static void rs64_stop(void)

arch/powerpc/platforms/Kconfig

@@ -272,4 +272,14 @@ config CPM2
 	  you wish to build a kernel for a machine with a CPM2 coprocessor
 	  on it (826x, 827x, 8560).
 
+config AXON_RAM
+	tristate "Axon DDR2 memory device driver"
+	depends on PPC_IBM_CELL_BLADE
+	default m
+	help
+	  It registers one block device per Axon's DDR2 memory bank found
+	  on a system. Block devices are called axonram?, their major and
+	  minor numbers are available in /proc/devices, /proc/partitions or
+	  in /sys/block/axonram?/dev.
+
 endmenu

arch/powerpc/platforms/cell/Kconfig

@@ -73,4 +73,14 @@ config CBE_CPUFREQ
 	  For details, take a look at <file:Documentation/cpu-freq/>.
 	  If you don't have such processor, say N
 
+config CBE_CPUFREQ_PMI
+	tristate "CBE frequency scaling using PMI interface"
+	depends on CBE_CPUFREQ && PPC_PMI && EXPERIMENTAL
+	default n
+	help
+	  Select this, if you want to use the PMI interface
+	  to switch frequencies. Using PMI, the
+	  processor will not only be able to run at lower speed,
+	  but also at lower core voltage.
+
 endmenu

arch/powerpc/platforms/cell/Makefile

@@ -4,7 +4,9 @@ obj-$(CONFIG_PPC_CELL_NATIVE)		+= interrupt.o iommu.o setup.o \
 obj-$(CONFIG_CBE_RAS)			+= ras.o
 
 obj-$(CONFIG_CBE_THERM)			+= cbe_thermal.o
-obj-$(CONFIG_CBE_CPUFREQ)		+= cbe_cpufreq.o
+obj-$(CONFIG_CBE_CPUFREQ_PMI)		+= cbe_cpufreq_pmi.o
+obj-$(CONFIG_CBE_CPUFREQ)		+= cbe-cpufreq.o
+cbe-cpufreq-y				+= cbe_cpufreq_pervasive.o cbe_cpufreq.o
 
 ifeq ($(CONFIG_SMP),y)
 obj-$(CONFIG_PPC_CELL_NATIVE)		+= smp.o
@@ -23,3 +25,5 @@ obj-$(CONFIG_SPU_BASE)			+= spu_callbacks.o spu_base.o \
 					   $(spu-priv1-y) \
 					   $(spu-manage-y) \
 					   spufs/
+
+obj-$(CONFIG_PCI_MSI)			+= axon_msi.o

arch/powerpc/platforms/cell/cbe_cpufreq.c

 /*
  * cpufreq driver for the cell processor
  *
- * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005-2007
  *
  * Author: Christian Krafft <krafft@de.ibm.com>
  *
@@ -21,18 +21,11 @@
  */
 
 #include <linux/cpufreq.h>
-#include <linux/timer.h>
-
-#include <asm/hw_irq.h>
-#include <asm/io.h>
 #include <asm/machdep.h>
-#include <asm/processor.h>
-#include <asm/prom.h>
-#include <asm/time.h>
-#include <asm/pmi.h>
 #include <asm/of_platform.h>
-
+#include <asm/prom.h>
 #include "cbe_regs.h"
+#include "cbe_cpufreq.h"
 
 static DEFINE_MUTEX(cbe_switch_mutex);
 
@@ -50,159 +43,24 @@ static struct cpufreq_frequency_table cbe_freqs[] = {
 	{0,	CPUFREQ_TABLE_END},
 };
 
-/* to write to MIC register */
-static u64 MIC_Slow_Fast_Timer_table[] = {
-	[0 ... 7] = 0x007fc00000000000ull,
-};
-
-/* more values for the MIC */
-static u64 MIC_Slow_Next_Timer_table[] = {
-	0x0000240000000000ull,
-	0x0000268000000000ull,
-	0x000029C000000000ull,
-	0x00002D0000000000ull,
-	0x0000300000000000ull,
-	0x0000334000000000ull,
-	0x000039C000000000ull,
-	0x00003FC000000000ull,
-};
-
-static unsigned int pmi_frequency_limit = 0;
 /*
  * hardware specific functions
  */
 
-static struct of_device *pmi_dev;
-
-#ifdef CONFIG_PPC_PMI
-static int set_pmode_pmi(int cpu, unsigned int pmode)
-{
-	int ret;
-	pmi_message_t pmi_msg;
-#ifdef DEBUG
-	u64 time;
-#endif
-
-	pmi_msg.type = PMI_TYPE_FREQ_CHANGE;
-	pmi_msg.data1 =	cbe_cpu_to_node(cpu);
-	pmi_msg.data2 = pmode;
-
-#ifdef DEBUG
-	time = (u64) get_cycles();
-#endif
-
-	pmi_send_message(pmi_dev, pmi_msg);
-	ret = pmi_msg.data2;
-
-	pr_debug("PMI returned slow mode %d\n", ret);
-
-#ifdef DEBUG
-	time = (u64) get_cycles() - time; /* actual cycles (not cpu cycles!) */
-	time = 1000000000 * time / CLOCK_TICK_RATE; /* time in ns (10^-9) */
-	pr_debug("had to wait %lu ns for a transition\n", time);
-#endif
-	return ret;
-}
-#endif
-
-static int get_pmode(int cpu)
+static int set_pmode(unsigned int cpu, unsigned int slow_mode)
 {
-	int ret;
-	struct cbe_pmd_regs __iomem *pmd_regs;
-
-	pmd_regs = cbe_get_cpu_pmd_regs(cpu);
-	ret = in_be64(&pmd_regs->pmsr) & 0x07;
-
-	return ret;
-}
-
-static int set_pmode_reg(int cpu, unsigned int pmode)
-{
-	struct cbe_pmd_regs __iomem *pmd_regs;
-	struct cbe_mic_tm_regs __iomem *mic_tm_regs;
-	u64 flags;
-	u64 value;
-
-	local_irq_save(flags);
-
-	mic_tm_regs = cbe_get_cpu_mic_tm_regs(cpu);
-	pmd_regs = cbe_get_cpu_pmd_regs(cpu);
-
-	pr_debug("pm register is mapped at %p\n", &pmd_regs->pmcr);
-	pr_debug("mic register is mapped at %p\n", &mic_tm_regs->slow_fast_timer_0);
-
-	out_be64(&mic_tm_regs->slow_fast_timer_0, MIC_Slow_Fast_Timer_table[pmode]);
-	out_be64(&mic_tm_regs->slow_fast_timer_1, MIC_Slow_Fast_Timer_table[pmode]);
-
-	out_be64(&mic_tm_regs->slow_next_timer_0, MIC_Slow_Next_Timer_table[pmode]);
-	out_be64(&mic_tm_regs->slow_next_timer_1, MIC_Slow_Next_Timer_table[pmode]);
-
-	value = in_be64(&pmd_regs->pmcr);
-	/* set bits to zero */
-	value &= 0xFFFFFFFFFFFFFFF8ull;
-	/* set bits to next pmode */
-	value |= pmode;
-
-	out_be64(&pmd_regs->pmcr, value);
-
-	/* wait until new pmode appears in status register */
-	value = in_be64(&pmd_regs->pmsr) & 0x07;
-	while(value != pmode) {
-		cpu_relax();
-		value = in_be64(&pmd_regs->pmsr) & 0x07;
-	}
-
-	local_irq_restore(flags);
-
-	return 0;
-}
+	int rc;
 
-static int set_pmode(int cpu, unsigned int slow_mode) {
-#ifdef CONFIG_PPC_PMI
-	if (pmi_dev)
-		return set_pmode_pmi(cpu, slow_mode);
+	if (cbe_cpufreq_has_pmi)
+		rc = cbe_cpufreq_set_pmode_pmi(cpu, slow_mode);
 	else
-#endif
-		return set_pmode_reg(cpu, slow_mode);
-}
-
-static void cbe_cpufreq_handle_pmi(struct of_device *dev, pmi_message_t pmi_msg)
-{
-	u8 cpu;
-	u8 cbe_pmode_new;
-
-	BUG_ON(pmi_msg.type != PMI_TYPE_FREQ_CHANGE);
+		rc = cbe_cpufreq_set_pmode(cpu, slow_mode);
 
-	cpu = cbe_node_to_cpu(pmi_msg.data1);
-	cbe_pmode_new = pmi_msg.data2;
+	pr_debug("register contains slow mode %d\n", cbe_cpufreq_get_pmode(cpu));
 
-	pmi_frequency_limit = cbe_freqs[cbe_pmode_new].frequency;
-
-	pr_debug("cbe_handle_pmi: max freq=%d\n", pmi_frequency_limit);
-}
-
-static int pmi_notifier(struct notifier_block *nb,
-				       unsigned long event, void *data)
-{
-	struct cpufreq_policy *policy = data;
-
-	if (event != CPUFREQ_INCOMPATIBLE)
-		return 0;
-
-	cpufreq_verify_within_limits(policy, 0, pmi_frequency_limit);
-	return 0;
+	return rc;
 }
 
-static struct notifier_block pmi_notifier_block = {
-	.notifier_call = pmi_notifier,
-};
-
-static struct pmi_handler cbe_pmi_handler = {
-	.type			= PMI_TYPE_FREQ_CHANGE,
-	.handle_pmi_message	= cbe_cpufreq_handle_pmi,
-};
-
-
 /*
  * cpufreq functions
  */
@@ -221,8 +79,19 @@ static int cbe_cpufreq_cpu_init(struct cpufreq_policy *policy)
 
 	pr_debug("init cpufreq on CPU %d\n", policy->cpu);
 
+	/*
+	 * Let's check we can actually get to the CELL regs
+	 */
+	if (!cbe_get_cpu_pmd_regs(policy->cpu) ||
+	    !cbe_get_cpu_mic_tm_regs(policy->cpu)) {
+		pr_info("invalid CBE regs pointers for cpufreq\n");
+		return -EINVAL;
+	}
+
 	max_freqp = of_get_property(cpu, "clock-frequency", NULL);
 
+	of_node_put(cpu);
+
 	if (!max_freqp)
 		return -EINVAL;
 
@@ -239,10 +108,12 @@ static int cbe_cpufreq_cpu_init(struct cpufreq_policy *policy)
 	}
 
 	policy->governor = CPUFREQ_DEFAULT_GOVERNOR;
-	/* if DEBUG is enabled set_pmode() measures the correct latency of a transition */
+
+	/* if DEBUG is enabled set_pmode() measures the latency
+	 * of a transition */
 	policy->cpuinfo.transition_latency = 25000;
 
-	cur_pmode = get_pmode(policy->cpu);
+	cur_pmode = cbe_cpufreq_get_pmode(policy->cpu);
 	pr_debug("current pmode is at %d\n",cur_pmode);
 
 	policy->cur = cbe_freqs[cur_pmode].frequency;
@@ -253,21 +124,13 @@ static int cbe_cpufreq_cpu_init(struct cpufreq_policy *policy)
 
 	cpufreq_frequency_table_get_attr(cbe_freqs, policy->cpu);
 
-	if (pmi_dev) {
-		/* frequency might get limited later, initialize limit with max_freq */
-		pmi_frequency_limit = max_freq;
-		cpufreq_register_notifier(&pmi_notifier_block, CPUFREQ_POLICY_NOTIFIER);
-	}
-
-	/* this ensures that policy->cpuinfo_min and policy->cpuinfo_max are set correctly */
+	/* this ensures that policy->cpuinfo_min
+	 * and policy->cpuinfo_max are set correctly */
 	return cpufreq_frequency_table_cpuinfo(policy, cbe_freqs);
 }
 
 static int cbe_cpufreq_cpu_exit(struct cpufreq_policy *policy)
 {
-	if (pmi_dev)
-		cpufreq_unregister_notifier(&pmi_notifier_block, CPUFREQ_POLICY_NOTIFIER);
-
 	cpufreq_frequency_table_put_attr(policy->cpu);
 	return 0;
 }
@@ -277,13 +140,13 @@ static int cbe_cpufreq_verify(struct cpufreq_policy *policy)
 	return cpufreq_frequency_table_verify(policy, cbe_freqs);
 }
 
-
-static int cbe_cpufreq_target(struct cpufreq_policy *policy, unsigned int target_freq,
-			    unsigned int relation)
+static int cbe_cpufreq_target(struct cpufreq_policy *policy,
+			      unsigned int target_freq,
+			      unsigned int relation)
 {
 	int rc;
 	struct cpufreq_freqs freqs;
-	int cbe_pmode_new;
+	unsigned int cbe_pmode_new;
 
 	cpufreq_frequency_table_target(policy,
 				       cbe_freqs,
@@ -298,12 +161,14 @@ static int cbe_cpufreq_target(struct cpufreq_policy *policy, unsigned int target
 	mutex_lock(&cbe_switch_mutex);
 	cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);
 
-	pr_debug("setting frequency for cpu %d to %d kHz, 1/%d of max frequency\n",
+	pr_debug("setting frequency for cpu %d to %d kHz, " \
+		 "1/%d of max frequency\n",
 		 policy->cpu,
 		 cbe_freqs[cbe_pmode_new].frequency,
 		 cbe_freqs[cbe_pmode_new].index);
 
 	rc = set_pmode(policy->cpu, cbe_pmode_new);
+
 	cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE);
 	mutex_unlock(&cbe_switch_mutex);
 
@@ -326,28 +191,14 @@ static struct cpufreq_driver cbe_cpufreq_driver = {
 
 static int __init cbe_cpufreq_init(void)
 {
-#ifdef CONFIG_PPC_PMI
-	struct device_node *np;
-#endif
 	if (!machine_is(cell))
 		return -ENODEV;
-#ifdef CONFIG_PPC_PMI
-	np = of_find_node_by_type(NULL, "ibm,pmi");
-
-	pmi_dev = of_find_device_by_node(np);
 
-	if (pmi_dev)
-		pmi_register_handler(pmi_dev, &cbe_pmi_handler);
-#endif
 	return cpufreq_register_driver(&cbe_cpufreq_driver);
 }
 
 static void __exit cbe_cpufreq_exit(void)
 {
-#ifdef CONFIG_PPC_PMI
-	if (pmi_dev)
-		pmi_unregister_handler(pmi_dev, &cbe_pmi_handler);
-#endif
 	cpufreq_unregister_driver(&cbe_cpufreq_driver);
 }
 

arch/powerpc/platforms/cell/cbe_cpufreq.h

+/*
+ * cbe_cpufreq.h
+ *
+ * This file contains the definitions used by the cbe_cpufreq driver.
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005-2007
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ */
+
+#include <linux/cpufreq.h>
+#include <linux/types.h>
+
+int cbe_cpufreq_set_pmode(int cpu, unsigned int pmode);
+int cbe_cpufreq_get_pmode(int cpu);
+
+int cbe_cpufreq_set_pmode_pmi(int cpu, unsigned int pmode);
+
+#if defined(CONFIG_CBE_CPUFREQ_PMI) || defined(CONFIG_CBE_CPUFREQ_PMI_MODULE)
+extern bool cbe_cpufreq_has_pmi;
+#else
+#define cbe_cpufreq_has_pmi (0)
+#endif

arch/powerpc/platforms/cell/cbe_cpufreq_pervasive.c

+/*
+ * pervasive backend for the cbe_cpufreq driver
+ *
+ * This driver makes use of the pervasive unit to
+ * engage the desired frequency.
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005-2007
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/io.h>
+#include <linux/kernel.h>
+#include <linux/time.h>
+#include <asm/machdep.h>
+#include <asm/hw_irq.h>
+
+#include "cbe_regs.h"
+#include "cbe_cpufreq.h"
+
+/* to write to MIC register */
+static u64 MIC_Slow_Fast_Timer_table[] = {
+	[0 ... 7] = 0x007fc00000000000ull,
+};
+
+/* more values for the MIC */
+static u64 MIC_Slow_Next_Timer_table[] = {
+	0x0000240000000000ull,
+	0x0000268000000000ull,
+	0x000029C000000000ull,
+	0x00002D0000000000ull,
+	0x0000300000000000ull,
+	0x0000334000000000ull,
+	0x000039C000000000ull,
+	0x00003FC000000000ull,
+};
+
+
+int cbe_cpufreq_set_pmode(int cpu, unsigned int pmode)
+{
+	struct cbe_pmd_regs __iomem *pmd_regs;
+	struct cbe_mic_tm_regs __iomem *mic_tm_regs;
+	u64 flags;
+	u64 value;
+#ifdef DEBUG
+	long time;
+#endif
+
+	local_irq_save(flags);
+
+	mic_tm_regs = cbe_get_cpu_mic_tm_regs(cpu);
+	pmd_regs = cbe_get_cpu_pmd_regs(cpu);
+
+#ifdef DEBUG
+	time = jiffies;
+#endif
+
+	out_be64(&mic_tm_regs->slow_fast_timer_0, MIC_Slow_Fast_Timer_table[pmode]);
+	out_be64(&mic_tm_regs->slow_fast_timer_1, MIC_Slow_Fast_Timer_table[pmode]);
+
+	out_be64(&mic_tm_regs->slow_next_timer_0, MIC_Slow_Next_Timer_table[pmode]);
+	out_be64(&mic_tm_regs->slow_next_timer_1, MIC_Slow_Next_Timer_table[pmode]);
+
+	value = in_be64(&pmd_regs->pmcr);
+	/* set bits to zero */
+	value &= 0xFFFFFFFFFFFFFFF8ull;
+	/* set bits to next pmode */
+	value |= pmode;
+
+	out_be64(&pmd_regs->pmcr, value);
+
+#ifdef DEBUG
+	/* wait until new pmode appears in status register */
+	value = in_be64(&pmd_regs->pmsr) & 0x07;
+	while (value != pmode) {
+		cpu_relax();
+		value = in_be64(&pmd_regs->pmsr) & 0x07;
+	}
+
+	time = jiffies  - time;
+	time = jiffies_to_msecs(time);
+	pr_debug("had to wait %lu ms for a transition using " \
+		 "pervasive unit\n", time);
+#endif
+	local_irq_restore(flags);
+
+	return 0;
+}
+
+
+int cbe_cpufreq_get_pmode(int cpu)
+{
+	int ret;
+	struct cbe_pmd_regs __iomem *pmd_regs;
+
+	pmd_regs = cbe_get_cpu_pmd_regs(cpu);
+	ret = in_be64(&pmd_regs->pmsr) & 0x07;
+
+	return ret;
+}
+

arch/powerpc/platforms/cell/cbe_cpufreq_pmi.c

+/*
+ * pmi backend for the cbe_cpufreq driver
+ *
+ * (C) Copyright IBM Deutschland Entwicklung GmbH 2005-2007
+ *
+ * Author: Christian Krafft <krafft@de.ibm.com>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2, or (at your option)
+ * any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+
+#include <linux/kernel.h>
+#include <linux/types.h>
+#include <linux/timer.h>
+#include <asm/of_platform.h>
+#include <asm/processor.h>
+#include <asm/prom.h>
+#include <asm/pmi.h>
+
+#ifdef DEBUG
+#include <asm/time.h>
+#endif
+
+#include "cbe_regs.h"
+#include "cbe_cpufreq.h"
+
+static u8 pmi_slow_mode_limit[MAX_CBE];
+
+bool cbe_cpufreq_has_pmi = false;
+EXPORT_SYMBOL_GPL(cbe_cpufreq_has_pmi);
+
+/*
+ * hardware specific functions
+ */
+
+int cbe_cpufreq_set_pmode_pmi(int cpu, unsigned int pmode)
+{
+	int ret;
+	pmi_message_t pmi_msg;
+#ifdef DEBUG
+	long time;
+#endif
+	pmi_msg.type = PMI_TYPE_FREQ_CHANGE;
+	pmi_msg.data1 =	cbe_cpu_to_node(cpu);
+	pmi_msg.data2 = pmode;
+
+#ifdef DEBUG
+	time = jiffies;
+#endif
+	pmi_send_message(pmi_msg);
+
+#ifdef DEBUG
+	time = jiffies  - time;
+	time = jiffies_to_msecs(time);
+	pr_debug("had to wait %lu ms for a transition using " \
+		 "PMI\n", time);
+#endif
+	ret = pmi_msg.data2;
+	pr_debug("PMI returned slow mode %d\n", ret);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(cbe_cpufreq_set_pmode_pmi);
+
+
+static void cbe_cpufreq_handle_pmi(pmi_message_t pmi_msg)
+{
+	u8 node, slow_mode;
+
+	BUG_ON(pmi_msg.type != PMI_TYPE_FREQ_CHANGE);
+
+	node = pmi_msg.data1;
+	slow_mode = pmi_msg.data2;
+
+	pmi_slow_mode_limit[node] = slow_mode;
+
+	pr_debug("cbe_handle_pmi: node: %d max_freq: %d\n", node, slow_mode);
+}
+
+static int pmi_notifier(struct notifier_block *nb,
+				       unsigned long event, void *data)
+{
+	struct cpufreq_policy *policy = data;
+	struct cpufreq_frequency_table *cbe_freqs;
+	u8 node;
+
+	cbe_freqs = cpufreq_frequency_get_table(policy->cpu);
+	node = cbe_cpu_to_node(policy->cpu);
+
+	pr_debug("got notified, event=%lu, node=%u\n", event, node);
+
+	if (pmi_slow_mode_limit[node] != 0) {
+		pr_debug("limiting node %d to slow mode %d\n",
+			 node, pmi_slow_mode_limit[node]);
+
+		cpufreq_verify_within_limits(policy, 0,
+
+			cbe_freqs[pmi_slow_mode_limit[node]].frequency);
+	}
+
+	return 0;
+}
+
+static struct notifier_block pmi_notifier_block = {
+	.notifier_call = pmi_notifier,
+};
+
+static struct pmi_handler cbe_pmi_handler = {
+	.type			= PMI_TYPE_FREQ_CHANGE,
+	.handle_pmi_message	= cbe_cpufreq_handle_pmi,
+};
+
+
+
+static int __init cbe_cpufreq_pmi_init(void)
+{
+	cbe_cpufreq_has_pmi = pmi_register_handler(&cbe_pmi_handler) == 0;
+
+	if (!cbe_cpufreq_has_pmi)
+		return -ENODEV;
+
+	cpufreq_register_notifier(&pmi_notifier_block, CPUFREQ_POLICY_NOTIFIER);
+
+	return 0;
+}
+
+static void __exit cbe_cpufreq_pmi_exit(void)
+{
+	cpufreq_unregister_notifier(&pmi_notifier_block, CPUFREQ_POLICY_NOTIFIER);
+	pmi_unregister_handler(&cbe_pmi_handler);
+}
+
+module_init(cbe_cpufreq_pmi_init);
+module_exit(cbe_cpufreq_pmi_exit);
+
+MODULE_LICENSE("GPL");
+MODULE_AUTHOR("Christian Krafft <krafft@de.ibm.com>");

arch/powerpc/platforms/cell/cbe_regs.c

@@ -174,6 +174,13 @@ static struct device_node *cbe_get_be_node(int cpu_id)
 
 		cpu_handle = of_get_property(np, "cpus", &len);
 
+		/*
+		 * the CAB SLOF tree is non compliant, so we just assume
+		 * there is only one node
+		 */
+		if (WARN_ON_ONCE(!cpu_handle))
+			return np;
+
 		for (i=0; i<len; i++)
 			if (of_find_node_by_phandle(cpu_handle[i]) == of_get_cpu_node(cpu_id, NULL))
 				return np;

arch/powerpc/platforms/cell/cbe_thermal.c

@@ -292,7 +292,7 @@ static struct attribute_group ppe_attribute_group = {
 /*
  * initialize throttling with default values
  */
-static void __init init_default_values(void)
+static int __init init_default_values(void)
 {
 	int cpu;
 	struct cbe_pmd_regs __iomem *pmd_regs;
@@ -339,25 +339,40 @@ static void __init init_default_values(void)
 	for_each_possible_cpu (cpu) {
 		pr_debug("processing cpu %d\n", cpu);
 		sysdev = get_cpu_sysdev(cpu);
+
+		if (!sysdev) {
+			pr_info("invalid sysdev pointer for cbe_thermal\n");
+			return -EINVAL;
+		}
+
 		pmd_regs = cbe_get_cpu_pmd_regs(sysdev->id);
 
+		if (!pmd_regs) {
+			pr_info("invalid CBE regs pointer for cbe_thermal\n");
+			return -EINVAL;
+		}
+
 		out_be64(&pmd_regs->tm_str2, str2);
 		out_be64(&pmd_regs->tm_str1.val, str1.val);
 		out_be64(&pmd_regs->tm_tpr.val, tpr.val);
 		out_be64(&pmd_regs->tm_cr1.val, cr1.val);
 		out_be64(&pmd_regs->tm_cr2, cr2);
 	}
+
+	return 0;
 }
 
 
 static int __init thermal_init(void)
 {
-	init_default_values();
+	int rc = init_default_values();
 
-	spu_add_sysdev_attr_group(&spu_attribute_group);
-	cpu_add_sysdev_attr_group(&ppe_attribute_group);
+	if (rc == 0) {
+		spu_add_sysdev_attr_group(&spu_attribute_group);
+		cpu_add_sysdev_attr_group(&ppe_attribute_group);
+	}
 
-	return 0;
+	return rc;
 }
 module_init(thermal_init);
 

arch/powerpc/platforms/cell/spu_syscalls.c

@@ -34,14 +34,27 @@ struct spufs_calls spufs_calls = {
  * this file is not used and the syscalls directly enter the fs code */
 
 asmlinkage long sys_spu_create(const char __user *name,
-		unsigned int flags, mode_t mode)
+		unsigned int flags, mode_t mode, int neighbor_fd)
 {
 	long ret;
 	struct module *owner = spufs_calls.owner;
+	struct file *neighbor;
+	int fput_needed;
 
 	ret = -ENOSYS;
 	if (owner && try_module_get(owner)) {
-		ret = spufs_calls.create_thread(name, flags, mode);
+		if (flags & SPU_CREATE_AFFINITY_SPU) {
+			neighbor = fget_light(neighbor_fd, &fput_needed);
+			if (neighbor) {
+				ret = spufs_calls.create_thread(name, flags,
+								mode, neighbor);
+				fput_light(neighbor, fput_needed);
+			}
+		}
+		else {
+			ret = spufs_calls.create_thread(name, flags,
+							mode, NULL);
+		}
 		module_put(owner);
 	}
 	return ret;

arch/powerpc/platforms/cell/spufs/context.c

@@ -22,6 +22,7 @@
 
 #include <linux/fs.h>
 #include <linux/mm.h>
+#include <linux/module.h>
 #include <linux/slab.h>
 #include <asm/atomic.h>
 #include <asm/spu.h>
@@ -55,12 +56,12 @@ struct spu_context *alloc_spu_context(struct spu_gang *gang)
 	ctx->ops = &spu_backing_ops;
 	ctx->owner = get_task_mm(current);
 	INIT_LIST_HEAD(&ctx->rq);
+	INIT_LIST_HEAD(&ctx->aff_list);
 	if (gang)
 		spu_gang_add_ctx(gang, ctx);
 	ctx->cpus_allowed = current->cpus_allowed;
 	spu_set_timeslice(ctx);
-	ctx->stats.execution_state = SPUCTX_UTIL_USER;
-	ctx->stats.tstamp = jiffies;
+	ctx->stats.util_state = SPU_UTIL_IDLE_LOADED;
 
 	atomic_inc(&nr_spu_contexts);
 	goto out;
@@ -81,6 +82,8 @@ void destroy_spu_context(struct kref *kref)
 	spu_fini_csa(&ctx->csa);
 	if (ctx->gang)
 		spu_gang_remove_ctx(ctx->gang, ctx);
+	if (ctx->prof_priv_kref)
+		kref_put(ctx->prof_priv_kref, ctx->prof_priv_release);
 	BUG_ON(!list_empty(&ctx->rq));
 	atomic_dec(&nr_spu_contexts);
 	kfree(ctx);
@@ -166,6 +169,39 @@ int spu_acquire_runnable(struct spu_context *ctx, unsigned long flags)
 void spu_acquire_saved(struct spu_context *ctx)
 {
 	spu_acquire(ctx);
-	if (ctx->state != SPU_STATE_SAVED)
+	if (ctx->state != SPU_STATE_SAVED) {
+		set_bit(SPU_SCHED_WAS_ACTIVE, &ctx->sched_flags);
 		spu_deactivate(ctx);
+	}
+}
+
+/**
+ * spu_release_saved - unlock spu context and return it to the runqueue
+ * @ctx:	context to unlock
+ */
+void spu_release_saved(struct spu_context *ctx)
+{
+	BUG_ON(ctx->state != SPU_STATE_SAVED);
+
+	if (test_and_clear_bit(SPU_SCHED_WAS_ACTIVE, &ctx->sched_flags))
+		spu_activate(ctx, 0);
+
+	spu_release(ctx);
 }
+
+void spu_set_profile_private_kref(struct spu_context *ctx,
+				  struct kref *prof_info_kref,
+				  void ( * prof_info_release) (struct kref *kref))
+{
+	ctx->prof_priv_kref = prof_info_kref;
+	ctx->prof_priv_release = prof_info_release;
+}
+EXPORT_SYMBOL_GPL(spu_set_profile_private_kref);
+
+void *spu_get_profile_private_kref(struct spu_context *ctx)
+{
+	return ctx->prof_priv_kref;
+}
+EXPORT_SYMBOL_GPL(spu_get_profile_private_kref);
+
+

arch/powerpc/platforms/cell/spufs/coredump.c

@@ -226,7 +226,7 @@ static void spufs_arch_write_notes(struct file *file)
 		spu_acquire_saved(ctx_info->ctx);
 		for (j = 0; j < spufs_coredump_num_notes; j++)
 			spufs_arch_write_note(ctx_info, j, file);
-		spu_release(ctx_info->ctx);
+		spu_release_saved(ctx_info->ctx);
 		list_del(&ctx_info->list);
 		kfree(ctx_info);
 	}

arch/powerpc/platforms/cell/spufs/fault.c

@@ -179,16 +179,14 @@ int spufs_handle_class1(struct spu_context *ctx)
 	if (!(dsisr & (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED)))
 		return 0;
 
-	spuctx_switch_state(ctx, SPUCTX_UTIL_IOWAIT);
+	spuctx_switch_state(ctx, SPU_UTIL_IOWAIT);
 
 	pr_debug("ctx %p: ea %016lx, dsisr %016lx state %d\n", ctx, ea,
 		dsisr, ctx->state);
 
 	ctx->stats.hash_flt++;
-	if (ctx->state == SPU_STATE_RUNNABLE) {
+	if (ctx->state == SPU_STATE_RUNNABLE)
 		ctx->spu->stats.hash_flt++;
-		spu_switch_state(ctx->spu, SPU_UTIL_IOWAIT);
-	}
 
 	/* we must not hold the lock when entering spu_handle_mm_fault */
 	spu_release(ctx);
@@ -226,7 +224,7 @@ int spufs_handle_class1(struct spu_context *ctx)
 	} else
 		spufs_handle_dma_error(ctx, ea, SPE_EVENT_SPE_DATA_STORAGE);
 
-	spuctx_switch_state(ctx, SPUCTX_UTIL_SYSTEM);
+	spuctx_switch_state(ctx, SPU_UTIL_SYSTEM);
 	return ret;
 }
 EXPORT_SYMBOL_GPL(spufs_handle_class1);

arch/powerpc/platforms/cell/spufs/file.c

@@ -370,7 +370,7 @@ spufs_regs_read(struct file *file, char __user *buffer,
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_regs_read(ctx, buffer, size, pos);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 
@@ -392,7 +392,7 @@ spufs_regs_write(struct file *file, const char __user *buffer,
 	ret = copy_from_user(lscsa->gprs + *pos - size,
 			     buffer, size) ? -EFAULT : size;
 
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 
@@ -421,7 +421,7 @@ spufs_fpcr_read(struct file *file, char __user * buffer,
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_fpcr_read(ctx, buffer, size, pos);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 
@@ -443,7 +443,7 @@ spufs_fpcr_write(struct file *file, const char __user * buffer,
 	ret = copy_from_user((char *)&lscsa->fpcr + *pos - size,
 			     buffer, size) ? -EFAULT : size;
 
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 
@@ -868,7 +868,7 @@ static ssize_t spufs_signal1_read(struct file *file, char __user *buf,
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_signal1_read(ctx, buf, len, pos);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -934,6 +934,13 @@ static const struct file_operations spufs_signal1_fops = {
 	.mmap = spufs_signal1_mmap,
 };
 
+static const struct file_operations spufs_signal1_nosched_fops = {
+	.open = spufs_signal1_open,
+	.release = spufs_signal1_release,
+	.write = spufs_signal1_write,
+	.mmap = spufs_signal1_mmap,
+};
+
 static int spufs_signal2_open(struct inode *inode, struct file *file)
 {
 	struct spufs_inode_info *i = SPUFS_I(inode);
@@ -992,7 +999,7 @@ static ssize_t spufs_signal2_read(struct file *file, char __user *buf,
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_signal2_read(ctx, buf, len, pos);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -1062,6 +1069,13 @@ static const struct file_operations spufs_signal2_fops = {
 	.mmap = spufs_signal2_mmap,
 };
 
+static const struct file_operations spufs_signal2_nosched_fops = {
+	.open = spufs_signal2_open,
+	.release = spufs_signal2_release,
+	.write = spufs_signal2_write,
+	.mmap = spufs_signal2_mmap,
+};
+
 static void spufs_signal1_type_set(void *data, u64 val)
 {
 	struct spu_context *ctx = data;
@@ -1612,7 +1626,7 @@ static void spufs_decr_set(void *data, u64 val)
 	struct spu_lscsa *lscsa = ctx->csa.lscsa;
 	spu_acquire_saved(ctx);
 	lscsa->decr.slot[0] = (u32) val;
-	spu_release(ctx);
+	spu_release_saved(ctx);
 }
 
 static u64 __spufs_decr_get(void *data)
@@ -1628,7 +1642,7 @@ static u64 spufs_decr_get(void *data)
 	u64 ret;
 	spu_acquire_saved(ctx);
 	ret = __spufs_decr_get(data);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 DEFINE_SIMPLE_ATTRIBUTE(spufs_decr_ops, spufs_decr_get, spufs_decr_set,
@@ -1637,17 +1651,21 @@ DEFINE_SIMPLE_ATTRIBUTE(spufs_decr_ops, spufs_decr_get, spufs_decr_set,
 static void spufs_decr_status_set(void *data, u64 val)
 {
 	struct spu_context *ctx = data;
-	struct spu_lscsa *lscsa = ctx->csa.lscsa;
 	spu_acquire_saved(ctx);
-	lscsa->decr_status.slot[0] = (u32) val;
-	spu_release(ctx);
+	if (val)
+		ctx->csa.priv2.mfc_control_RW |= MFC_CNTL_DECREMENTER_RUNNING;
+	else
+		ctx->csa.priv2.mfc_control_RW &= ~MFC_CNTL_DECREMENTER_RUNNING;
+	spu_release_saved(ctx);
 }
 
 static u64 __spufs_decr_status_get(void *data)
 {
 	struct spu_context *ctx = data;
-	struct spu_lscsa *lscsa = ctx->csa.lscsa;
-	return lscsa->decr_status.slot[0];
+	if (ctx->csa.priv2.mfc_control_RW & MFC_CNTL_DECREMENTER_RUNNING)
+		return SPU_DECR_STATUS_RUNNING;
+	else
+		return 0;
 }
 
 static u64 spufs_decr_status_get(void *data)
@@ -1656,7 +1674,7 @@ static u64 spufs_decr_status_get(void *data)
 	u64 ret;
 	spu_acquire_saved(ctx);
 	ret = __spufs_decr_status_get(data);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 DEFINE_SIMPLE_ATTRIBUTE(spufs_decr_status_ops, spufs_decr_status_get,
@@ -1668,7 +1686,7 @@ static void spufs_event_mask_set(void *data, u64 val)
 	struct spu_lscsa *lscsa = ctx->csa.lscsa;
 	spu_acquire_saved(ctx);
 	lscsa->event_mask.slot[0] = (u32) val;
-	spu_release(ctx);
+	spu_release_saved(ctx);
 }
 
 static u64 __spufs_event_mask_get(void *data)
@@ -1684,7 +1702,7 @@ static u64 spufs_event_mask_get(void *data)
 	u64 ret;
 	spu_acquire_saved(ctx);
 	ret = __spufs_event_mask_get(data);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 DEFINE_SIMPLE_ATTRIBUTE(spufs_event_mask_ops, spufs_event_mask_get,
@@ -1708,7 +1726,7 @@ static u64 spufs_event_status_get(void *data)
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_event_status_get(data);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 DEFINE_SIMPLE_ATTRIBUTE(spufs_event_status_ops, spufs_event_status_get,
@@ -1720,7 +1738,7 @@ static void spufs_srr0_set(void *data, u64 val)
 	struct spu_lscsa *lscsa = ctx->csa.lscsa;
 	spu_acquire_saved(ctx);
 	lscsa->srr0.slot[0] = (u32) val;
-	spu_release(ctx);
+	spu_release_saved(ctx);
 }
 
 static u64 spufs_srr0_get(void *data)
@@ -1730,7 +1748,7 @@ static u64 spufs_srr0_get(void *data)
 	u64 ret;
 	spu_acquire_saved(ctx);
 	ret = lscsa->srr0.slot[0];
-	spu_release(ctx);
+	spu_release_saved(ctx);
 	return ret;
 }
 DEFINE_SIMPLE_ATTRIBUTE(spufs_srr0_ops, spufs_srr0_get, spufs_srr0_set,
@@ -1786,7 +1804,7 @@ static u64 spufs_lslr_get(void *data)
 
 	spu_acquire_saved(ctx);
 	ret = __spufs_lslr_get(data);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -1850,7 +1868,7 @@ static ssize_t spufs_mbox_info_read(struct file *file, char __user *buf,
 	spin_lock(&ctx->csa.register_lock);
 	ret = __spufs_mbox_info_read(ctx, buf, len, pos);
 	spin_unlock(&ctx->csa.register_lock);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -1888,7 +1906,7 @@ static ssize_t spufs_ibox_info_read(struct file *file, char __user *buf,
 	spin_lock(&ctx->csa.register_lock);
 	ret = __spufs_ibox_info_read(ctx, buf, len, pos);
 	spin_unlock(&ctx->csa.register_lock);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -1929,7 +1947,7 @@ static ssize_t spufs_wbox_info_read(struct file *file, char __user *buf,
 	spin_lock(&ctx->csa.register_lock);
 	ret = __spufs_wbox_info_read(ctx, buf, len, pos);
 	spin_unlock(&ctx->csa.register_lock);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -1979,7 +1997,7 @@ static ssize_t spufs_dma_info_read(struct file *file, char __user *buf,
 	spin_lock(&ctx->csa.register_lock);
 	ret = __spufs_dma_info_read(ctx, buf, len, pos);
 	spin_unlock(&ctx->csa.register_lock);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -2030,7 +2048,7 @@ static ssize_t spufs_proxydma_info_read(struct file *file, char __user *buf,
 	spin_lock(&ctx->csa.register_lock);
 	ret = __spufs_proxydma_info_read(ctx, buf, len, pos);
 	spin_unlock(&ctx->csa.register_lock);
-	spu_release(ctx);
+	spu_release_saved(ctx);
 
 	return ret;
 }
@@ -2065,14 +2083,26 @@ static const char *ctx_state_names[] = {
 };
 
 static unsigned long long spufs_acct_time(struct spu_context *ctx,
-		enum spuctx_execution_state state)
+		enum spu_utilization_state state)
 {
-	unsigned long time = ctx->stats.times[state];
+	struct timespec ts;
+	unsigned long long time = ctx->stats.times[state];
 
-	if (ctx->stats.execution_state == state)
-		time += jiffies - ctx->stats.tstamp;
+	/*
+	 * In general, utilization statistics are updated by the controlling
+	 * thread as the spu context moves through various well defined
+	 * state transitions, but if the context is lazily loaded its
+	 * utilization statistics are not updated as the controlling thread
+	 * is not tightly coupled with the execution of the spu context.  We
+	 * calculate and apply the time delta from the last recorded state
+	 * of the spu context.
+	 */
+	if (ctx->spu && ctx->stats.util_state == state) {
+		ktime_get_ts(&ts);
+		time += timespec_to_ns(&ts) - ctx->stats.tstamp;
+	}
 
-	return jiffies_to_msecs(time);
+	return time / NSEC_PER_MSEC;
 }
 
 static unsigned long long spufs_slb_flts(struct spu_context *ctx)
@@ -2107,11 +2137,11 @@ static int spufs_show_stat(struct seq_file *s, void *private)
 	spu_acquire(ctx);
 	seq_printf(s, "%s %llu %llu %llu %llu "
 		      "%llu %llu %llu %llu %llu %llu %llu %llu\n",
-		ctx_state_names[ctx->stats.execution_state],
-		spufs_acct_time(ctx, SPUCTX_UTIL_USER),
-		spufs_acct_time(ctx, SPUCTX_UTIL_SYSTEM),
-		spufs_acct_time(ctx, SPUCTX_UTIL_IOWAIT),
-		spufs_acct_time(ctx, SPUCTX_UTIL_LOADED),
+		ctx_state_names[ctx->stats.util_state],
+		spufs_acct_time(ctx, SPU_UTIL_USER),
+		spufs_acct_time(ctx, SPU_UTIL_SYSTEM),
+		spufs_acct_time(ctx, SPU_UTIL_IOWAIT),
+		spufs_acct_time(ctx, SPU_UTIL_IDLE_LOADED),
 		ctx->stats.vol_ctx_switch,
 		ctx->stats.invol_ctx_switch,
 		spufs_slb_flts(ctx),
@@ -2184,8 +2214,8 @@ struct tree_descr spufs_dir_nosched_contents[] = {
 	{ "mbox_stat", &spufs_mbox_stat_fops, 0444, },
 	{ "ibox_stat", &spufs_ibox_stat_fops, 0444, },
 	{ "wbox_stat", &spufs_wbox_stat_fops, 0444, },
-	{ "signal1", &spufs_signal1_fops, 0666, },
-	{ "signal2", &spufs_signal2_fops, 0666, },
+	{ "signal1", &spufs_signal1_nosched_fops, 0222, },
+	{ "signal2", &spufs_signal2_nosched_fops, 0222, },
 	{ "signal1_type", &spufs_signal1_type, 0666, },
 	{ "signal2_type", &spufs_signal2_type, 0666, },
 	{ "mss", &spufs_mss_fops, 0666, },

arch/powerpc/platforms/cell/spufs/gang.c

@@ -35,7 +35,9 @@ struct spu_gang *alloc_spu_gang(void)
 
 	kref_init(&gang->kref);
 	mutex_init(&gang->mutex);
+	mutex_init(&gang->aff_mutex);
 	INIT_LIST_HEAD(&gang->list);
+	INIT_LIST_HEAD(&gang->aff_list_head);
 
 out:
 	return gang;
@@ -73,6 +75,10 @@ void spu_gang_remove_ctx(struct spu_gang *gang, struct spu_context *ctx)
 {
 	mutex_lock(&gang->mutex);
 	WARN_ON(ctx->gang != gang);
+	if (!list_empty(&ctx->aff_list)) {
+		list_del_init(&ctx->aff_list);
+		gang->aff_flags &= ~AFF_OFFSETS_SET;
+	}
 	list_del_init(&ctx->gang_list);
 	gang->contexts--;
 	mutex_unlock(&gang->mutex);

arch/powerpc/platforms/cell/spufs/inode.c

@@ -316,11 +316,107 @@ out:
 	return ret;
 }
 
-static int spufs_create_context(struct inode *inode,
-			struct dentry *dentry,
-			struct vfsmount *mnt, int flags, int mode)
+static struct spu_context *
+spufs_assert_affinity(unsigned int flags, struct spu_gang *gang,
+						struct file *filp)
+{
+	struct spu_context *tmp, *neighbor;
+	int count, node;
+	int aff_supp;
+
+	aff_supp = !list_empty(&(list_entry(cbe_spu_info[0].spus.next,
+					struct spu, cbe_list))->aff_list);
+
+	if (!aff_supp)
+		return ERR_PTR(-EINVAL);
+
+	if (flags & SPU_CREATE_GANG)
+		return ERR_PTR(-EINVAL);
+
+	if (flags & SPU_CREATE_AFFINITY_MEM &&
+	    gang->aff_ref_ctx &&
+	    gang->aff_ref_ctx->flags & SPU_CREATE_AFFINITY_MEM)
+		return ERR_PTR(-EEXIST);
+
+	if (gang->aff_flags & AFF_MERGED)
+		return ERR_PTR(-EBUSY);
+
+	neighbor = NULL;
+	if (flags & SPU_CREATE_AFFINITY_SPU) {
+		if (!filp || filp->f_op != &spufs_context_fops)
+			return ERR_PTR(-EINVAL);
+
+		neighbor = get_spu_context(
+				SPUFS_I(filp->f_dentry->d_inode)->i_ctx);
+
+		if (!list_empty(&neighbor->aff_list) && !(neighbor->aff_head) &&
+		    !list_is_last(&neighbor->aff_list, &gang->aff_list_head) &&
+		    !list_entry(neighbor->aff_list.next, struct spu_context,
+		    aff_list)->aff_head)
+			return ERR_PTR(-EEXIST);
+
+		if (gang != neighbor->gang)
+			return ERR_PTR(-EINVAL);
+
+		count = 1;
+		list_for_each_entry(tmp, &gang->aff_list_head, aff_list)
+			count++;
+		if (list_empty(&neighbor->aff_list))
+			count++;
+
+		for (node = 0; node < MAX_NUMNODES; node++) {
+			if ((cbe_spu_info[node].n_spus - atomic_read(
+				&cbe_spu_info[node].reserved_spus)) >= count)
+				break;
+		}
+
+		if (node == MAX_NUMNODES)
+			return ERR_PTR(-EEXIST);
+	}
+
+	return neighbor;
+}
+
+static void
+spufs_set_affinity(unsigned int flags, struct spu_context *ctx,
+					struct spu_context *neighbor)
+{
+	if (flags & SPU_CREATE_AFFINITY_MEM)
+		ctx->gang->aff_ref_ctx = ctx;
+
+	if (flags & SPU_CREATE_AFFINITY_SPU) {
+		if (list_empty(&neighbor->aff_list)) {
+			list_add_tail(&neighbor->aff_list,
+				&ctx->gang->aff_list_head);
+			neighbor->aff_head = 1;
+		}
+
+		if (list_is_last(&neighbor->aff_list, &ctx->gang->aff_list_head)
+		    || list_entry(neighbor->aff_list.next, struct spu_context,
+							aff_list)->aff_head) {
+			list_add(&ctx->aff_list, &neighbor->aff_list);
+		} else  {
+			list_add_tail(&ctx->aff_list, &neighbor->aff_list);
+			if (neighbor->aff_head) {
+				neighbor->aff_head = 0;
+				ctx->aff_head = 1;
+			}
+		}
+
+		if (!ctx->gang->aff_ref_ctx)
+			ctx->gang->aff_ref_ctx = ctx;
+	}
+}
+
+static int
+spufs_create_context(struct inode *inode, struct dentry *dentry,
+			struct vfsmount *mnt, int flags, int mode,
+			struct file *aff_filp)
 {
 	int ret;
+	int affinity;
+	struct spu_gang *gang;
+	struct spu_context *neighbor;
 
 	ret = -EPERM;
 	if ((flags & SPU_CREATE_NOSCHED) &&
@@ -336,9 +432,29 @@ static int spufs_create_context(struct inode *inode,
 	if ((flags & SPU_CREATE_ISOLATE) && !isolated_loader)
 		goto out_unlock;
 
+	gang = NULL;
+	neighbor = NULL;
+	affinity = flags & (SPU_CREATE_AFFINITY_MEM | SPU_CREATE_AFFINITY_SPU);
+	if (affinity) {
+		gang = SPUFS_I(inode)->i_gang;
+		ret = -EINVAL;
+		if (!gang)
+			goto out_unlock;
+		mutex_lock(&gang->aff_mutex);
+		neighbor = spufs_assert_affinity(flags, gang, aff_filp);
+		if (IS_ERR(neighbor)) {
+			ret = PTR_ERR(neighbor);
+			goto out_aff_unlock;
+		}
+	}
+
 	ret = spufs_mkdir(inode, dentry, flags, mode & S_IRWXUGO);
 	if (ret)
-		goto out_unlock;
+		goto out_aff_unlock;
+
+	if (affinity)
+		spufs_set_affinity(flags, SPUFS_I(dentry->d_inode)->i_ctx,
+								neighbor);
 
 	/*
 	 * get references for dget and mntget, will be released
@@ -352,6 +468,9 @@ static int spufs_create_context(struct inode *inode,
 		goto out;
 	}
 
+out_aff_unlock:
+	if (affinity)
+		mutex_unlock(&gang->aff_mutex);
 out_unlock:
 	mutex_unlock(&inode->i_mutex);
 out:
@@ -450,7 +569,8 @@ out:
 
 static struct file_system_type spufs_type;
 
-long spufs_create(struct nameidata *nd, unsigned int flags, mode_t mode)
+long spufs_create(struct nameidata *nd, unsigned int flags, mode_t mode,
+							struct file *filp)
 {
 	struct dentry *dentry;
 	int ret;
@@ -487,7 +607,7 @@ long spufs_create(struct nameidata *nd, unsigned int flags, mode_t mode)
 					dentry, nd->mnt, mode);
 	else
 		return spufs_create_context(nd->dentry->d_inode,
-					dentry, nd->mnt, flags, mode);
+					dentry, nd->mnt, flags, mode, filp);
 
 out_dput:
 	dput(dentry);

arch/powerpc/platforms/cell/spufs/run.c

@@ -18,15 +18,17 @@ void spufs_stop_callback(struct spu *spu)
 	wake_up_all(&ctx->stop_wq);
 }
 
-static inline int spu_stopped(struct spu_context *ctx, u32 * stat)
+static inline int spu_stopped(struct spu_context *ctx, u32 *stat)
 {
 	struct spu *spu;
 	u64 pte_fault;
 
 	*stat = ctx->ops->status_read(ctx);
-	if (ctx->state != SPU_STATE_RUNNABLE)
-		return 1;
+
 	spu = ctx->spu;
+	if (ctx->state != SPU_STATE_RUNNABLE ||
+	    test_bit(SPU_SCHED_NOTIFY_ACTIVE, &ctx->sched_flags))
+		return 1;
 	pte_fault = spu->dsisr &
 	    (MFC_DSISR_PTE_NOT_FOUND | MFC_DSISR_ACCESS_DENIED);
 	return (!(*stat & SPU_STATUS_RUNNING) || pte_fault || spu->class_0_pending) ?
@@ -124,8 +126,10 @@ out:
 	return ret;
 }
 
-static int spu_run_init(struct spu_context *ctx, u32 * npc)
+static int spu_run_init(struct spu_context *ctx, u32 *npc)
 {
+	spuctx_switch_state(ctx, SPU_UTIL_SYSTEM);
+
 	if (ctx->flags & SPU_CREATE_ISOLATE) {
 		unsigned long runcntl;
 
@@ -151,16 +155,20 @@ static int spu_run_init(struct spu_context *ctx, u32 * npc)
 		ctx->ops->runcntl_write(ctx, SPU_RUNCNTL_RUNNABLE);
 	}
 
+	spuctx_switch_state(ctx, SPU_UTIL_USER);
+
 	return 0;
 }
 
-static int spu_run_fini(struct spu_context *ctx, u32 * npc,
-			       u32 * status)
+static int spu_run_fini(struct spu_context *ctx, u32 *npc,
+			       u32 *status)
 {
 	int ret = 0;
 
 	*status = ctx->ops->status_read(ctx);
 	*npc = ctx->ops->npc_read(ctx);
+
+	spuctx_switch_state(ctx, SPU_UTIL_IDLE_LOADED);
 	spu_release(ctx);
 
 	if (signal_pending(current))
@@ -289,10 +297,10 @@ static inline int spu_process_events(struct spu_context *ctx)
 	return ret;
 }
 
-long spufs_run_spu(struct file *file, struct spu_context *ctx,
-		   u32 *npc, u32 *event)
+long spufs_run_spu(struct spu_context *ctx, u32 *npc, u32 *event)
 {
 	int ret;
+	struct spu *spu;
 	u32 status;
 
 	if (mutex_lock_interruptible(&ctx->run_mutex))
@@ -328,6 +336,17 @@ long spufs_run_spu(struct file *file, struct spu_context *ctx,
 		ret = spufs_wait(ctx->stop_wq, spu_stopped(ctx, &status));
 		if (unlikely(ret))
 			break;
+		spu = ctx->spu;
+		if (unlikely(test_and_clear_bit(SPU_SCHED_NOTIFY_ACTIVE,
+						&ctx->sched_flags))) {
+			if (!(status & SPU_STATUS_STOPPED_BY_STOP)) {
+				spu_switch_notify(spu, ctx);
+				continue;
+			}
+		}
+
+		spuctx_switch_state(ctx, SPU_UTIL_SYSTEM);
+
 		if ((status & SPU_STATUS_STOPPED_BY_STOP) &&
 		    (status >> SPU_STOP_STATUS_SHIFT == 0x2104)) {
 			ret = spu_process_callback(ctx);
@@ -356,6 +375,7 @@ long spufs_run_spu(struct file *file, struct spu_context *ctx,
 	    (ctx->state == SPU_STATE_RUNNABLE))
 		ctx->stats.libassist++;
 
+
 	ctx->ops->master_stop(ctx);
 	ret = spu_run_fini(ctx, npc, &status);
 	spu_yield(ctx);

arch/powerpc/platforms/cell/spufs/spu_restore.c

@@ -84,13 +84,13 @@ static inline void restore_decr(void)
 	unsigned int decr_running;
 	unsigned int decr;
 
-	/* Restore, Step 6:
+	/* Restore, Step 6(moved):
 	 *    If the LSCSA "decrementer running" flag is set
 	 *    then write the SPU_WrDec channel with the
 	 *    decrementer value from LSCSA.
 	 */
 	offset = LSCSA_QW_OFFSET(decr_status);
-	decr_running = regs_spill[offset].slot[0];
+	decr_running = regs_spill[offset].slot[0] & SPU_DECR_STATUS_RUNNING;
 	if (decr_running) {
 		offset = LSCSA_QW_OFFSET(decr);
 		decr = regs_spill[offset].slot[0];
@@ -318,10 +318,10 @@ int main()
 	build_dma_list(lscsa_ea);	/* Step 3.  */
 	restore_upper_240kb(lscsa_ea);	/* Step 4.  */
 					/* Step 5: done by 'exit'. */
-	restore_decr();			/* Step 6. */
 	enqueue_putllc(lscsa_ea);	/* Step 7. */
 	set_tag_update();		/* Step 8. */
 	read_tag_status();		/* Step 9. */
+	restore_decr();			/* moved Step 6. */
 	read_llar_status();		/* Step 10. */
 	write_ppu_mb();			/* Step 11. */
 	write_ppuint_mb();		/* Step 12. */

arch/powerpc/platforms/cell/spufs/spufs.h

@@ -40,17 +40,13 @@ enum {
 struct spu_context_ops;
 struct spu_gang;
 
-/*
- * This is the state for spu utilization reporting to userspace.
- * Because this state is visible to userspace it must never change and needs
- * to be kept strictly separate from any internal state kept by the kernel.
- */
-enum spuctx_execution_state {
-	SPUCTX_UTIL_USER = 0,
-	SPUCTX_UTIL_SYSTEM,
-	SPUCTX_UTIL_IOWAIT,
-	SPUCTX_UTIL_LOADED,
-	SPUCTX_UTIL_MAX
+enum {
+	SPU_SCHED_WAS_ACTIVE,	/* was active upon spu_acquire_saved()  */
+};
+
+/* ctx->sched_flags */
+enum {
+	SPU_SCHED_NOTIFY_ACTIVE,
 };
 
 struct spu_context {
@@ -89,6 +85,8 @@ struct spu_context {
 
 	struct list_head gang_list;
 	struct spu_gang *gang;
+	struct kref *prof_priv_kref;
+	void ( * prof_priv_release) (struct kref *kref);
 
 	/* owner thread */
 	pid_t tid;
@@ -104,9 +102,9 @@ struct spu_context {
 	/* statistics */
 	struct {
 		/* updates protected by ctx->state_mutex */
-		enum spuctx_execution_state execution_state;
-		unsigned long tstamp;		/* time of last ctx switch */
-		unsigned long times[SPUCTX_UTIL_MAX];
+		enum spu_utilization_state util_state;
+		unsigned long long tstamp;	/* time of last state switch */
+		unsigned long long times[SPU_UTIL_MAX];
 		unsigned long long vol_ctx_switch;
 		unsigned long long invol_ctx_switch;
 		unsigned long long min_flt;
@@ -118,6 +116,10 @@ struct spu_context {
 		unsigned long long class2_intr_base; /* # at last ctx switch */
 		unsigned long long libassist;
 	} stats;
+
+	struct list_head aff_list;
+	int aff_head;
+	int aff_offset;
 };
 
 struct spu_gang {
@@ -125,8 +127,19 @@ struct spu_gang {
 	struct mutex mutex;
 	struct kref kref;
 	int contexts;
+
+	struct spu_context *aff_ref_ctx;
+	struct list_head aff_list_head;
+	struct mutex aff_mutex;
+	int aff_flags;
+	struct spu *aff_ref_spu;
+	atomic_t aff_sched_count;
 };
 
+/* Flag bits for spu_gang aff_flags */
+#define AFF_OFFSETS_SET		1
+#define AFF_MERGED		2
+
 struct mfc_dma_command {
 	int32_t pad;	/* reserved */
 	uint32_t lsa;	/* local storage address */
@@ -190,10 +203,9 @@ extern struct tree_descr spufs_dir_contents[];
 extern struct tree_descr spufs_dir_nosched_contents[];
 
 /* system call implementation */
-long spufs_run_spu(struct file *file,
-		   struct spu_context *ctx, u32 *npc, u32 *status);
-long spufs_create(struct nameidata *nd,
-			 unsigned int flags, mode_t mode);
+long spufs_run_spu(struct spu_context *ctx, u32 *npc, u32 *status);
+long spufs_create(struct nameidata *nd, unsigned int flags,
+			mode_t mode, struct file *filp);
 extern const struct file_operations spufs_context_fops;
 
 /* gang management */
@@ -206,6 +218,9 @@ void spu_gang_add_ctx(struct spu_gang *gang, struct spu_context *ctx);
 /* fault handling */
 int spufs_handle_class1(struct spu_context *ctx);
 
+/* affinity */
+struct spu *affinity_check(struct spu_context *ctx);
+
 /* context management */
 extern atomic_t nr_spu_contexts;
 static inline void spu_acquire(struct spu_context *ctx)
@@ -227,15 +242,17 @@ void spu_unmap_mappings(struct spu_context *ctx);
 void spu_forget(struct spu_context *ctx);
 int spu_acquire_runnable(struct spu_context *ctx, unsigned long flags);
 void spu_acquire_saved(struct spu_context *ctx);
+void spu_release_saved(struct spu_context *ctx);
 
 int spu_activate(struct spu_context *ctx, unsigned long flags);
 void spu_deactivate(struct spu_context *ctx);
 void spu_yield(struct spu_context *ctx);
+void spu_switch_notify(struct spu *spu, struct spu_context *ctx);
 void spu_set_timeslice(struct spu_context *ctx);
 void spu_update_sched_info(struct spu_context *ctx);
 void __spu_update_sched_info(struct spu_context *ctx);
 int __init spu_sched_init(void);
-void __exit spu_sched_exit(void);
+void spu_sched_exit(void);
 
 extern char *isolated_loader;
 
@@ -293,30 +310,34 @@ extern int spufs_coredump_num_notes;
  * line.
  */
 static inline void spuctx_switch_state(struct spu_context *ctx,
-		enum spuctx_execution_state new_state)
+		enum spu_utilization_state new_state)
 {
-	WARN_ON(!mutex_is_locked(&ctx->state_mutex));
-
-	if (ctx->stats.execution_state != new_state) {
-		unsigned long curtime = jiffies;
-
-		ctx->stats.times[ctx->stats.execution_state] +=
-				 curtime - ctx->stats.tstamp;
-		ctx->stats.tstamp = curtime;
-		ctx->stats.execution_state = new_state;
-	}
-}
+	unsigned long long curtime;
+	signed long long delta;
+	struct timespec ts;
+	struct spu *spu;
+	enum spu_utilization_state old_state;
 
-static inline void spu_switch_state(struct spu *spu,
-		enum spuctx_execution_state new_state)
-{
-	if (spu->stats.utilization_state != new_state) {
-		unsigned long curtime = jiffies;
+	ktime_get_ts(&ts);
+	curtime = timespec_to_ns(&ts);
+	delta = curtime - ctx->stats.tstamp;
 
-		spu->stats.times[spu->stats.utilization_state] +=
-				 curtime - spu->stats.tstamp;
+	WARN_ON(!mutex_is_locked(&ctx->state_mutex));
+	WARN_ON(delta < 0);
+
+	spu = ctx->spu;
+	old_state = ctx->stats.util_state;
+	ctx->stats.util_state = new_state;
+	ctx->stats.tstamp = curtime;
+
+	/*
+	 * Update the physical SPU utilization statistics.
+	 */
+	if (spu) {
+		ctx->stats.times[old_state] += delta;
+		spu->stats.times[old_state] += delta;
+		spu->stats.util_state = new_state;
 		spu->stats.tstamp = curtime;
-		spu->stats.utilization_state = new_state;
 	}
 }
 

arch/powerpc/platforms/cell/spufs/syscalls.c

@@ -47,7 +47,7 @@ static long do_spu_run(struct file *filp,
 		goto out;
 
 	i = SPUFS_I(filp->f_path.dentry->d_inode);
-	ret = spufs_run_spu(filp, i->i_ctx, &npc, &status);
+	ret = spufs_run_spu(i->i_ctx, &npc, &status);
 
 	if (put_user(npc, unpc))
 		ret = -EFAULT;
@@ -76,8 +76,8 @@ asmlinkage long sys_spu_run(int fd, __u32 __user *unpc, __u32 __user *ustatus)
 }
 #endif
 
-asmlinkage long sys_spu_create(const char __user *pathname,
-					unsigned int flags, mode_t mode)
+asmlinkage long do_spu_create(const char __user *pathname, unsigned int flags,
+				mode_t mode, struct file *neighbor)
 {
 	char *tmp;
 	int ret;
@@ -90,7 +90,7 @@ asmlinkage long sys_spu_create(const char __user *pathname,
 		ret = path_lookup(tmp, LOOKUP_PARENT|
 				LOOKUP_OPEN|LOOKUP_CREATE, &nd);
 		if (!ret) {
-			ret = spufs_create(&nd, flags, mode);
+			ret = spufs_create(&nd, flags, mode, neighbor);
 			path_release(&nd);
 		}
 		putname(tmp);
@@ -99,8 +99,32 @@ asmlinkage long sys_spu_create(const char __user *pathname,
 	return ret;
 }
 
+#ifndef MODULE
+asmlinkage long sys_spu_create(const char __user *pathname, unsigned int flags,
+				mode_t mode, int neighbor_fd)
+{
+	int fput_needed;
+	struct file *neighbor;
+	long ret;
+
+	if (flags & SPU_CREATE_AFFINITY_SPU) {
+		ret = -EBADF;
+		neighbor = fget_light(neighbor_fd, &fput_needed);
+		if (neighbor) {
+			ret = do_spu_create(pathname, flags, mode, neighbor);
+			fput_light(neighbor, fput_needed);
+		}
+	}
+	else {
+		ret = do_spu_create(pathname, flags, mode, NULL);
+	}
+
+	return ret;
+}
+#endif
+
 struct spufs_calls spufs_calls = {
-	.create_thread = sys_spu_create,
+	.create_thread = do_spu_create,
 	.spu_run = do_spu_run,
 	.owner = THIS_MODULE,
 };

arch/powerpc/sysdev/Makefile

@@ -17,6 +17,7 @@ obj-$(CONFIG_QUICC_ENGINE)	+= qe_lib/
 mv64x60-$(CONFIG_PCI)		+= mv64x60_pci.o
 obj-$(CONFIG_MV64X60)		+= $(mv64x60-y) mv64x60_pic.o mv64x60_dev.o
 obj-$(CONFIG_RTC_DRV_CMOS)	+= rtc_cmos_setup.o
+obj-$(CONFIG_AXON_RAM)		+= axonram.o
 
 # contains only the suspend handler for time
 ifeq ($(CONFIG_RTC_CLASS),)

drivers/oprofile/buffer_sync.c

@@ -26,8 +26,9 @@
 #include <linux/profile.h>
 #include <linux/module.h>
 #include <linux/fs.h>
+#include <linux/oprofile.h>
 #include <linux/sched.h>
- 
+
 #include "oprofile_stats.h"
 #include "event_buffer.h"
 #include "cpu_buffer.h"

drivers/oprofile/event_buffer.h

@@ -19,28 +19,10 @@ void free_event_buffer(void);
  
 /* wake up the process sleeping on the event file */
 void wake_up_buffer_waiter(void);
- 
-/* Each escaped entry is prefixed by ESCAPE_CODE
- * then one of the following codes, then the
- * relevant data.
- */
-#define ESCAPE_CODE			~0UL
-#define CTX_SWITCH_CODE 		1
-#define CPU_SWITCH_CODE 		2
-#define COOKIE_SWITCH_CODE 		3
-#define KERNEL_ENTER_SWITCH_CODE	4
-#define KERNEL_EXIT_SWITCH_CODE		5
-#define MODULE_LOADED_CODE		6
-#define CTX_TGID_CODE			7
-#define TRACE_BEGIN_CODE		8
-#define TRACE_END_CODE			9
- 
+
 #define INVALID_COOKIE ~0UL
 #define NO_COOKIE 0UL
 
-/* add data to the event buffer */
-void add_event_entry(unsigned long data);
- 
 extern const struct file_operations event_buffer_fops;
  
 /* mutex between sync_cpu_buffers() and the

include/asm-powerpc/oprofile_impl.h

@@ -39,14 +39,16 @@ struct op_system_config {
 
 /* Per-arch configuration */
 struct op_powerpc_model {
-	void (*reg_setup) (struct op_counter_config *,
+	int (*reg_setup) (struct op_counter_config *,
 			   struct op_system_config *,
 			   int num_counters);
-	void (*cpu_setup) (struct op_counter_config *);
-	void (*start) (struct op_counter_config *);
-        void (*global_start) (struct op_counter_config *);
+	int  (*cpu_setup) (struct op_counter_config *);
+	int  (*start) (struct op_counter_config *);
+	int  (*global_start) (struct op_counter_config *);
 	void (*stop) (void);
 	void (*global_stop) (void);
+	int (*sync_start)(void);
+	int (*sync_stop)(void);
 	void (*handle_interrupt) (struct pt_regs *,
 				  struct op_counter_config *);
 	int num_counters;

include/linux/dcookies.h

@@ -12,6 +12,7 @@
 
 #ifdef CONFIG_PROFILING
  
+#include <linux/dcache.h>
 #include <linux/types.h>
  
 struct dcookie_user;