sn_hwperf.c

linux-2.6.15.6

			return -EINVAL;
		for (ordinal=0, p=objs; p != obj; p++) {
			if (!SN_HWPERF_FOREIGN(p))
				ordinal += p->ports;
		}
		for (pt = 0; pt < obj->ports; pt++) {
			for (p = objs, i = 0; i < sn_hwperf_obj_cnt; i++, p++) {
				if (ptdata[pt].conn_id == p->id) {
					break;
				}
			}
			seq_printf(s, "numalink %d %s-%d",
			    ordinal+pt, obj->location, ptdata[pt].port);

			if (i >= sn_hwperf_obj_cnt) {
				/* no connection */
				seq_puts(s, " local endpoint disconnected"
					    ", protocol unknown\n");
				continue;
			}

			if (obj->sn_hwp_this_part && p->sn_hwp_this_part)
				/* both ends local to this partition */
				seq_puts(s, " local");
			else if (!obj->sn_hwp_this_part && !p->sn_hwp_this_part)
				/* both ends of the link in foreign partiton */
				seq_puts(s, " foreign");
			else
				/* link straddles a partition */
				seq_puts(s, " shared");

			/*
			 * Unlikely, but strictly should query the LLP config
			 * registers because an NL4R can be configured to run
			 * NL3 protocol, even when not talking to an NL3 router.
			 * Ditto for node-node.
			 */
			seq_printf(s, " endpoint %s-%d, protocol %s\n",
				p->location, ptdata[pt].conn_port,
				(SN_HWPERF_IS_NL3ROUTER(obj) ||
				SN_HWPERF_IS_NL3ROUTER(p)) ?  "LLP3" : "LLP4");
		}
		vfree(ptdata);
	}

	return 0;
}

static void *sn_topology_start(struct seq_file *s, loff_t * pos)
{
	struct sn_hwperf_object_info *objs = s->private;

	if (*pos < sn_hwperf_obj_cnt)
		return (void *)(objs + *pos);

	return NULL;
}

static void *sn_topology_next(struct seq_file *s, void *v, loff_t * pos)
{
	++*pos;
	return sn_topology_start(s, pos);
}

static void sn_topology_stop(struct seq_file *m, void *v)
{
	return;
}

/*
 * /proc/sgi_sn/sn_topology, read-only using seq_file
 */
static struct seq_operations sn_topology_seq_ops = {
	.start = sn_topology_start,
	.next = sn_topology_next,
	.stop = sn_topology_stop,
	.show = sn_topology_show
};

struct sn_hwperf_op_info {
	u64 op;
	struct sn_hwperf_ioctl_args *a;
	void *p;
	int *v0;
	int ret;
};

static void sn_hwperf_call_sal(void *info)
{
	struct sn_hwperf_op_info *op_info = info;
	int r;

	r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op_info->op,
		      op_info->a->arg, op_info->a->sz,
		      (u64) op_info->p, 0, 0, op_info->v0);
	op_info->ret = r;
}

static int sn_hwperf_op_cpu(struct sn_hwperf_op_info *op_info)
{
	u32 cpu;
	u32 use_ipi;
	int r = 0;
	cpumask_t save_allowed;
	
	cpu = (op_info->a->arg & SN_HWPERF_ARG_CPU_MASK) >> 32;
	use_ipi = op_info->a->arg & SN_HWPERF_ARG_USE_IPI_MASK;
	op_info->a->arg &= SN_HWPERF_ARG_OBJID_MASK;

	if (cpu != SN_HWPERF_ARG_ANY_CPU) {
		if (cpu >= num_online_cpus() || !cpu_online(cpu)) {
			r = -EINVAL;
			goto out;
		}
	}

	if (cpu == SN_HWPERF_ARG_ANY_CPU || cpu == get_cpu()) {
		/* don't care, or already on correct cpu */
		sn_hwperf_call_sal(op_info);
	}
	else {
		if (use_ipi) {
			/* use an interprocessor interrupt to call SAL */
			smp_call_function_single(cpu, sn_hwperf_call_sal,
				op_info, 1, 1);
		}
		else {
			/* migrate the task before calling SAL */ 
			save_allowed = current->cpus_allowed;
			set_cpus_allowed(current, cpumask_of_cpu(cpu));
			sn_hwperf_call_sal(op_info);
			set_cpus_allowed(current, save_allowed);
		}
	}
	r = op_info->ret;

out:
	return r;
}

/* map SAL hwperf error code to system error code */
static int sn_hwperf_map_err(int hwperf_err)
{
	int e;

	switch(hwperf_err) {
	case SN_HWPERF_OP_OK:
		e = 0;
		break;

	case SN_HWPERF_OP_NOMEM:
		e = -ENOMEM;
		break;

	case SN_HWPERF_OP_NO_PERM:
		e = -EPERM;
		break;

	case SN_HWPERF_OP_IO_ERROR:
		e = -EIO;
		break;

	case SN_HWPERF_OP_BUSY:
		e = -EBUSY;
		break;

	case SN_HWPERF_OP_RECONFIGURE:
		e = -EAGAIN;
		break;

	case SN_HWPERF_OP_INVAL:
	default:
		e = -EINVAL;
		break;
	}

	return e;
}

/*
 * ioctl for "sn_hwperf" misc device
 */
static int
sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
{
	struct sn_hwperf_ioctl_args a;
	struct cpuinfo_ia64 *cdata;
	struct sn_hwperf_object_info *objs;
	struct sn_hwperf_object_info *cpuobj;
	struct sn_hwperf_op_info op_info;
	void *p = NULL;
	int nobj;
	char slice;
	int node;
	int r;
	int v0;
	int i;
	int j;

	unlock_kernel();

	/* only user requests are allowed here */
	if ((op & SN_HWPERF_OP_MASK) < 10) {
		r = -EINVAL;
		goto error;
	}
	r = copy_from_user(&a, (const void __user *)arg,
		sizeof(struct sn_hwperf_ioctl_args));
	if (r != 0) {
		r = -EFAULT;
		goto error;
	}

	/*
	 * Allocate memory to hold a kernel copy of the user buffer. The
	 * buffer contents are either copied in or out (or both) of user
	 * space depending on the flags encoded in the requested operation.
	 */
	if (a.ptr) {
		p = vmalloc(a.sz);
		if (!p) {
			r = -ENOMEM;
			goto error;
		}
	}

	if (op & SN_HWPERF_OP_MEM_COPYIN) {
		r = copy_from_user(p, (const void __user *)a.ptr, a.sz);
		if (r != 0) {
			r = -EFAULT;
			goto error;
		}
	}

	switch (op) {
	case SN_HWPERF_GET_CPU_INFO:
		if (a.sz == sizeof(u64)) {
			/* special case to get size needed */
			*(u64 *) p = (u64) num_online_cpus() *
				sizeof(struct sn_hwperf_object_info);
		} else
		if (a.sz < num_online_cpus() * sizeof(struct sn_hwperf_object_info)) {
			r = -ENOMEM;
			goto error;
		} else
		if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
			memset(p, 0, a.sz);
			for (i = 0; i < nobj; i++) {
				int cpuobj_index = 0;
				if (!SN_HWPERF_IS_NODE(objs + i))
					continue;
				node = sn_hwperf_obj_to_cnode(objs + i);
				for_each_online_cpu(j) {
					if (node != cpu_to_node(j))
						continue;
					cpuobj = (struct sn_hwperf_object_info *) p + cpuobj_index++;
					slice = 'a' + cpuid_to_slice(j);
					cdata = cpu_data(j);
					cpuobj->id = j;
					snprintf(cpuobj->name,
						 sizeof(cpuobj->name),
						 "CPU %luMHz %s",
						 cdata->proc_freq / 1000000,
						 cdata->vendor);
					snprintf(cpuobj->location,
						 sizeof(cpuobj->location),
						 "%s%c", objs[i].location,
						 slice);
				}
			}

			vfree(objs);
		}
		break;

	case SN_HWPERF_GET_NODE_NASID:
		if (a.sz != sizeof(u64) ||
		   (node = a.arg) < 0 || !node_possible(node)) {
			r = -EINVAL;
			goto error;
		}
		*(u64 *)p = (u64)cnodeid_to_nasid(node);
		break;

	case SN_HWPERF_GET_OBJ_NODE:
		if (a.sz != sizeof(u64) || a.arg < 0) {
			r = -EINVAL;
			goto error;
		}
		if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
			if (a.arg >= nobj) {
				r = -EINVAL;
				vfree(objs);
				goto error;
			}
			if (objs[(i = a.arg)].id != a.arg) {
				for (i = 0; i < nobj; i++) {
					if (objs[i].id == a.arg)
						break;
				}
			}
			if (i == nobj) {
				r = -EINVAL;
				vfree(objs);
				goto error;
			}

			if (!SN_HWPERF_IS_NODE(objs + i) &&
			    !SN_HWPERF_IS_IONODE(objs + i)) {
			    	r = -ENOENT;
				vfree(objs);
				goto error;
			}

			*(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i);
			vfree(objs);
		}
		break;

	case SN_HWPERF_GET_MMRS:
	case SN_HWPERF_SET_MMRS:
	case SN_HWPERF_OBJECT_DISTANCE:
		op_info.p = p;
		op_info.a = &a;
		op_info.v0 = &v0;
		op_info.op = op;
		r = sn_hwperf_op_cpu(&op_info);
		if (r) {
			r = sn_hwperf_map_err(r);
			a.v0 = v0;
			goto error;
		}
		break;

	default:
		/* all other ops are a direct SAL call */
		r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op,
			      a.arg, a.sz, (u64) p, 0, 0, &v0);
		if (r) {
			r = sn_hwperf_map_err(r);
			goto error;
		}
		a.v0 = v0;
		break;
	}

	if (op & SN_HWPERF_OP_MEM_COPYOUT) {
		r = copy_to_user((void __user *)a.ptr, p, a.sz);
		if (r != 0) {
			r = -EFAULT;
			goto error;
		}
	}

error:
	vfree(p);

	lock_kernel();
	return r;
}

static struct file_operations sn_hwperf_fops = {
	.ioctl = sn_hwperf_ioctl,
};

static struct miscdevice sn_hwperf_dev = {
	MISC_DYNAMIC_MINOR,
	"sn_hwperf",
	&sn_hwperf_fops
};

static int sn_hwperf_init(void)
{
	u64 v;
	int salr;
	int e = 0;

	/* single threaded, once-only initialization */
	down(&sn_hwperf_init_mutex);

	if (sn_hwperf_salheap) {
		up(&sn_hwperf_init_mutex);
		return e;
	}

	/*
	 * The PROM code needs a fixed reference node. For convenience the
	 * same node as the console I/O is used.
	 */
	sn_hwperf_master_nasid = (nasid_t) ia64_sn_get_console_nasid();

	/*
	 * Request the needed size and install the PROM scratch area.
	 * The PROM keeps various tracking bits in this memory area.
	 */
	salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
				 (u64) SN_HWPERF_GET_HEAPSIZE, 0,
				 (u64) sizeof(u64), (u64) &v, 0, 0, NULL);
	if (salr != SN_HWPERF_OP_OK) {
		e = -EINVAL;
		goto out;
	}

	if ((sn_hwperf_salheap = vmalloc(v)) == NULL) {
		e = -ENOMEM;
		goto out;
	}
	salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
				 SN_HWPERF_INSTALL_HEAP, 0, v,
				 (u64) sn_hwperf_salheap, 0, 0, NULL);
	if (salr != SN_HWPERF_OP_OK) {
		e = -EINVAL;
		goto out;
	}

	salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
				 SN_HWPERF_OBJECT_COUNT, 0,
				 sizeof(u64), (u64) &v, 0, 0, NULL);
	if (salr != SN_HWPERF_OP_OK) {
		e = -EINVAL;
		goto out;
	}
	sn_hwperf_obj_cnt = (int)v;

out:
	if (e < 0 && sn_hwperf_salheap) {
		vfree(sn_hwperf_salheap);
		sn_hwperf_salheap = NULL;
		sn_hwperf_obj_cnt = 0;
	}
	up(&sn_hwperf_init_mutex);
	return e;
}

int sn_topology_open(struct inode *inode, struct file *file)
{
	int e;
	struct seq_file *seq;
	struct sn_hwperf_object_info *objbuf;
	int nobj;

	if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
		e = seq_open(file, &sn_topology_seq_ops);
		seq = file->private_data;
		seq->private = objbuf;
	}

	return e;
}

int sn_topology_release(struct inode *inode, struct file *file)
{
	struct seq_file *seq = file->private_data;

	vfree(seq->private);
	return seq_release(inode, file);
}

int sn_hwperf_get_nearest_node(cnodeid_t node,
	cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
{
	int e;
	int nobj;
	struct sn_hwperf_object_info *objbuf;

	if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
		e = sn_hwperf_get_nearest_node_objdata(objbuf, nobj,
			node, near_mem_node, near_cpu_node);
		vfree(objbuf);
	}

	return e;
}

static int __devinit sn_hwperf_misc_register_init(void)
{
	int e;

	sn_hwperf_init();

	/*
	 * Register a dynamic misc device for hwperf ioctls. Platforms
	 * supporting hotplug will create /dev/sn_hwperf, else user
	 * can to look up the minor number in /proc/misc.
	 */
	if ((e = misc_register(&sn_hwperf_dev)) != 0) {
		printk(KERN_ERR "sn_hwperf_misc_register_init: failed to "
		"register misc device for \"%s\"\n", sn_hwperf_dev.name);
	}

	return e;
}

device_initcall(sn_hwperf_misc_register_init); /* after misc_init() */
EXPORT_SYMBOL(sn_hwperf_get_nearest_node);