xpc_partition.c

linux 内核源代码

		}

		nasid_mask = xpc_IPI_receive(&act_amos[word]);
		if (nasid_mask == 0) {
			/* no IRQs from nasids in this variable */
			continue;
		}

		dev_dbg(xpc_part, "AMO[%d] gave back 0x%lx\n", word,
			nasid_mask);


		/*
		 * If this nasid has been added to the machine since
		 * our partition was reset, this will retain the
		 * remote nasid in our reserved pages machine mask.
		 * This is used in the event of module reload.
		 */
		xpc_mach_nasids[word] |= nasid_mask;


		/* locate the nasid(s) which sent interrupts */

		for (bit = 0; bit < (8 * sizeof(u64)); bit++) {
			if (nasid_mask & (1UL << bit)) {
				n_IRQs_detected++;
				nasid = XPC_NASID_FROM_W_B(word, bit);
				dev_dbg(xpc_part, "interrupt from nasid %ld\n",
					nasid);
				xpc_identify_act_IRQ_req(nasid);
			}
		}
	}
	return n_IRQs_detected;
}


/*
 * See if the other side has responded to a partition disengage request
 * from us.
 */
int
xpc_partition_disengaged(struct xpc_partition *part)
{
	partid_t partid = XPC_PARTID(part);
	int disengaged;


	disengaged = (xpc_partition_engaged(1UL << partid) == 0);
	if (part->disengage_request_timeout) {
		if (!disengaged) {
			if (jiffies < part->disengage_request_timeout) {
				/* timelimit hasn't been reached yet */
				return 0;
			}

			/*
			 * Other side hasn't responded to our disengage
			 * request in a timely fashion, so assume it's dead.
			 */

			dev_info(xpc_part, "disengage from remote partition %d "
				"timed out\n", partid);
			xpc_disengage_request_timedout = 1;
			xpc_clear_partition_engaged(1UL << partid);
			disengaged = 1;
		}
		part->disengage_request_timeout = 0;

		/* cancel the timer function, provided it's not us */
		if (!in_interrupt()) {
			del_singleshot_timer_sync(&part->
						      disengage_request_timer);
		}

		DBUG_ON(part->act_state != XPC_P_DEACTIVATING &&
					part->act_state != XPC_P_INACTIVE);
		if (part->act_state != XPC_P_INACTIVE) {
			xpc_wakeup_channel_mgr(part);
		}

		if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
			xpc_cancel_partition_disengage_request(part);
		}
	}
	return disengaged;
}


/*
 * Mark specified partition as active.
 */
enum xpc_retval
xpc_mark_partition_active(struct xpc_partition *part)
{
	unsigned long irq_flags;
	enum xpc_retval ret;


	dev_dbg(xpc_part, "setting partition %d to ACTIVE\n", XPC_PARTID(part));

	spin_lock_irqsave(&part->act_lock, irq_flags);
	if (part->act_state == XPC_P_ACTIVATING) {
		part->act_state = XPC_P_ACTIVE;
		ret = xpcSuccess;
	} else {
		DBUG_ON(part->reason == xpcSuccess);
		ret = part->reason;
	}
	spin_unlock_irqrestore(&part->act_lock, irq_flags);

	return ret;
}


/*
 * Notify XPC that the partition is down.
 */
void
xpc_deactivate_partition(const int line, struct xpc_partition *part,
				enum xpc_retval reason)
{
	unsigned long irq_flags;


	spin_lock_irqsave(&part->act_lock, irq_flags);

	if (part->act_state == XPC_P_INACTIVE) {
		XPC_SET_REASON(part, reason, line);
		spin_unlock_irqrestore(&part->act_lock, irq_flags);
		if (reason == xpcReactivating) {
			/* we interrupt ourselves to reactivate partition */
			xpc_IPI_send_reactivate(part);
		}
		return;
	}
	if (part->act_state == XPC_P_DEACTIVATING) {
		if ((part->reason == xpcUnloading && reason != xpcUnloading) ||
					reason == xpcReactivating) {
			XPC_SET_REASON(part, reason, line);
		}
		spin_unlock_irqrestore(&part->act_lock, irq_flags);
		return;
	}

	part->act_state = XPC_P_DEACTIVATING;
	XPC_SET_REASON(part, reason, line);

	spin_unlock_irqrestore(&part->act_lock, irq_flags);

	if (XPC_SUPPORTS_DISENGAGE_REQUEST(part->remote_vars_version)) {
		xpc_request_partition_disengage(part);
		xpc_IPI_send_disengage(part);

		/* set a timelimit on the disengage request */
		part->disengage_request_timeout = jiffies +
					(xpc_disengage_request_timelimit * HZ);
		part->disengage_request_timer.expires =
					part->disengage_request_timeout;
		add_timer(&part->disengage_request_timer);
	}

	dev_dbg(xpc_part, "bringing partition %d down, reason = %d\n",
		XPC_PARTID(part), reason);

	xpc_partition_going_down(part, reason);
}


/*
 * Mark specified partition as inactive.
 */
void
xpc_mark_partition_inactive(struct xpc_partition *part)
{
	unsigned long irq_flags;


	dev_dbg(xpc_part, "setting partition %d to INACTIVE\n",
		XPC_PARTID(part));

	spin_lock_irqsave(&part->act_lock, irq_flags);
	part->act_state = XPC_P_INACTIVE;
	spin_unlock_irqrestore(&part->act_lock, irq_flags);
	part->remote_rp_pa = 0;
}


/*
 * SAL has provided a partition and machine mask.  The partition mask
 * contains a bit for each even nasid in our partition.  The machine
 * mask contains a bit for each even nasid in the entire machine.
 *
 * Using those two bit arrays, we can determine which nasids are
 * known in the machine.  Each should also have a reserved page
 * initialized if they are available for partitioning.
 */
void
xpc_discovery(void)
{
	void *remote_rp_base;
	struct xpc_rsvd_page *remote_rp;
	struct xpc_vars *remote_vars;
	u64 remote_rp_pa;
	u64 remote_vars_pa;
	int region;
	int region_size;
	int max_regions;
	int nasid;
	struct xpc_rsvd_page *rp;
	partid_t partid;
	struct xpc_partition *part;
	u64 *discovered_nasids;
	enum xpc_retval ret;


	remote_rp = xpc_kmalloc_cacheline_aligned(XPC_RP_HEADER_SIZE +
						xp_nasid_mask_bytes,
						GFP_KERNEL, &remote_rp_base);
	if (remote_rp == NULL) {
		return;
	}
	remote_vars = (struct xpc_vars *) remote_rp;


	discovered_nasids = kzalloc(sizeof(u64) * xp_nasid_mask_words,
							GFP_KERNEL);
	if (discovered_nasids == NULL) {
		kfree(remote_rp_base);
		return;
	}

	rp = (struct xpc_rsvd_page *) xpc_rsvd_page;

	/*
	 * The term 'region' in this context refers to the minimum number of
	 * nodes that can comprise an access protection grouping. The access
	 * protection is in regards to memory, IOI and IPI.
	 */
	max_regions = 64;
	region_size = sn_region_size;

	switch (region_size) {
	case 128:
		max_regions *= 2;
	case 64:
		max_regions *= 2;
	case 32:
		max_regions *= 2;
		region_size = 16;
		DBUG_ON(!is_shub2());
	}

	for (region = 0; region < max_regions; region++) {

		if ((volatile int) xpc_exiting) {
			break;
		}

		dev_dbg(xpc_part, "searching region %d\n", region);

		for (nasid = (region * region_size * 2);
		     nasid < ((region + 1) * region_size * 2);
		     nasid += 2) {

			if ((volatile int) xpc_exiting) {
				break;
			}

			dev_dbg(xpc_part, "checking nasid %d\n", nasid);


			if (XPC_NASID_IN_ARRAY(nasid, xpc_part_nasids)) {
				dev_dbg(xpc_part, "PROM indicates Nasid %d is "
					"part of the local partition; skipping "
					"region\n", nasid);
				break;
			}

			if (!(XPC_NASID_IN_ARRAY(nasid, xpc_mach_nasids))) {
				dev_dbg(xpc_part, "PROM indicates Nasid %d was "
					"not on Numa-Link network at reset\n",
					nasid);
				continue;
			}

			if (XPC_NASID_IN_ARRAY(nasid, discovered_nasids)) {
				dev_dbg(xpc_part, "Nasid %d is part of a "
					"partition which was previously "
					"discovered\n", nasid);
				continue;
			}


			/* pull over the reserved page structure */

			ret = xpc_get_remote_rp(nasid, discovered_nasids,
					      remote_rp, &remote_rp_pa);
			if (ret != xpcSuccess) {
				dev_dbg(xpc_part, "unable to get reserved page "
					"from nasid %d, reason=%d\n", nasid,
					ret);

				if (ret == xpcLocalPartid) {
					break;
				}
				continue;
			}

			remote_vars_pa = remote_rp->vars_pa;

			partid = remote_rp->partid;
			part = &xpc_partitions[partid];


			/* pull over the cross partition variables */

			ret = xpc_get_remote_vars(remote_vars_pa, remote_vars);
			if (ret != xpcSuccess) {
				dev_dbg(xpc_part, "unable to get XPC variables "
					"from nasid %d, reason=%d\n", nasid,
					ret);

				XPC_DEACTIVATE_PARTITION(part, ret);
				continue;
			}

			if (part->act_state != XPC_P_INACTIVE) {
				dev_dbg(xpc_part, "partition %d on nasid %d is "
					"already activating\n", partid, nasid);
				break;
			}

			/*
			 * Register the remote partition's AMOs with SAL so it
			 * can handle and cleanup errors within that address
			 * range should the remote partition go down. We don't
			 * unregister this range because it is difficult to
			 * tell when outstanding writes to the remote partition
			 * are finished and thus when it is thus safe to
			 * unregister. This should not result in wasted space
			 * in the SAL xp_addr_region table because we should
			 * get the same page for remote_act_amos_pa after
			 * module reloads and system reboots.
			 */
			if (sn_register_xp_addr_region(
					    remote_vars->amos_page_pa,
							PAGE_SIZE, 1) < 0) {
				dev_dbg(xpc_part, "partition %d failed to "
					"register xp_addr region 0x%016lx\n",
					partid, remote_vars->amos_page_pa);

				XPC_SET_REASON(part, xpcPhysAddrRegFailed,
						__LINE__);
				break;
			}

			/*
			 * The remote nasid is valid and available.
			 * Send an interrupt to that nasid to notify
			 * it that we are ready to begin activation.
			 */
			dev_dbg(xpc_part, "sending an interrupt to AMO 0x%lx, "
				"nasid %d, phys_cpuid 0x%x\n",
				remote_vars->amos_page_pa,
				remote_vars->act_nasid,
				remote_vars->act_phys_cpuid);

			if (XPC_SUPPORTS_DISENGAGE_REQUEST(remote_vars->
								version)) {
				part->remote_amos_page_pa =
						remote_vars->amos_page_pa;
				xpc_mark_partition_disengaged(part);
				xpc_cancel_partition_disengage_request(part);
			}
			xpc_IPI_send_activate(remote_vars);
		}
	}

	kfree(discovered_nasids);
	kfree(remote_rp_base);
}


/*
 * Given a partid, get the nasids owned by that partition from the
 * remote partition's reserved page.
 */
enum xpc_retval
xpc_initiate_partid_to_nasids(partid_t partid, void *nasid_mask)
{
	struct xpc_partition *part;
	u64 part_nasid_pa;
	int bte_res;


	part = &xpc_partitions[partid];
	if (part->remote_rp_pa == 0) {
		return xpcPartitionDown;
	}

	memset(nasid_mask, 0, XP_NASID_MASK_BYTES);

	part_nasid_pa = (u64) XPC_RP_PART_NASIDS(part->remote_rp_pa);

	bte_res = xp_bte_copy(part_nasid_pa, (u64) nasid_mask,
			xp_nasid_mask_bytes, (BTE_NOTIFY | BTE_WACQUIRE), NULL);

	return xpc_map_bte_errors(bte_res);
}