xpc_partition.c

linux 内核源代码

/*
 * This file is subject to the terms and conditions of the GNU General Public
 * License.  See the file "COPYING" in the main directory of this archive
 * for more details.
 *
 * Copyright (c) 2004-2006 Silicon Graphics, Inc.  All Rights Reserved.
 */


/*
 * Cross Partition Communication (XPC) partition support.
 *
 *	This is the part of XPC that detects the presence/absence of
 *	other partitions. It provides a heartbeat and monitors the
 *	heartbeats of other partitions.
 *
 */


#include <linux/kernel.h>
#include <linux/sysctl.h>
#include <linux/cache.h>
#include <linux/mmzone.h>
#include <linux/nodemask.h>
#include <asm/uncached.h>
#include <asm/sn/bte.h>
#include <asm/sn/intr.h>
#include <asm/sn/sn_sal.h>
#include <asm/sn/nodepda.h>
#include <asm/sn/addrs.h>
#include <asm/sn/xpc.h>


/* XPC is exiting flag */
int xpc_exiting;


/* SH_IPI_ACCESS shub register value on startup */
static u64 xpc_sh1_IPI_access;
static u64 xpc_sh2_IPI_access0;
static u64 xpc_sh2_IPI_access1;
static u64 xpc_sh2_IPI_access2;
static u64 xpc_sh2_IPI_access3;


/* original protection values for each node */
u64 xpc_prot_vec[MAX_NUMNODES];


/* this partition's reserved page pointers */
struct xpc_rsvd_page *xpc_rsvd_page;
static u64 *xpc_part_nasids;
static u64 *xpc_mach_nasids;
struct xpc_vars *xpc_vars;
struct xpc_vars_part *xpc_vars_part;

static int xp_nasid_mask_bytes;	/* actual size in bytes of nasid mask */
static int xp_nasid_mask_words;	/* actual size in words of nasid mask */


/*
 * For performance reasons, each entry of xpc_partitions[] is cacheline
 * aligned. And xpc_partitions[] is padded with an additional entry at the
 * end so that the last legitimate entry doesn't share its cacheline with
 * another variable.
 */
struct xpc_partition xpc_partitions[XP_MAX_PARTITIONS + 1];


/*
 * Generic buffer used to store a local copy of portions of a remote
 * partition's reserved page (either its header and part_nasids mask,
 * or its vars).
 */
char *xpc_remote_copy_buffer;
void *xpc_remote_copy_buffer_base;


/*
 * Guarantee that the kmalloc'd memory is cacheline aligned.
 */
void *
xpc_kmalloc_cacheline_aligned(size_t size, gfp_t flags, void **base)
{
	/* see if kmalloc will give us cachline aligned memory by default */
	*base = kmalloc(size, flags);
	if (*base == NULL) {
		return NULL;
	}
	if ((u64) *base == L1_CACHE_ALIGN((u64) *base)) {
		return *base;
	}
	kfree(*base);

	/* nope, we'll have to do it ourselves */
	*base = kmalloc(size + L1_CACHE_BYTES, flags);
	if (*base == NULL) {
		return NULL;
	}
	return (void *) L1_CACHE_ALIGN((u64) *base);
}


/*
 * Given a nasid, get the physical address of the  partition's reserved page
 * for that nasid. This function returns 0 on any error.
 */
static u64
xpc_get_rsvd_page_pa(int nasid)
{
	bte_result_t bte_res;
	s64 status;
	u64 cookie = 0;
	u64 rp_pa = nasid;	/* seed with nasid */
	u64 len = 0;
	u64 buf = buf;
	u64 buf_len = 0;
	void *buf_base = NULL;


	while (1) {

		status = sn_partition_reserved_page_pa(buf, &cookie, &rp_pa,
								&len);

		dev_dbg(xpc_part, "SAL returned with status=%li, cookie="
			"0x%016lx, address=0x%016lx, len=0x%016lx\n",
			status, cookie, rp_pa, len);

		if (status != SALRET_MORE_PASSES) {
			break;
		}

		if (L1_CACHE_ALIGN(len) > buf_len) {
			kfree(buf_base);
			buf_len = L1_CACHE_ALIGN(len);
			buf = (u64) xpc_kmalloc_cacheline_aligned(buf_len,
							GFP_KERNEL, &buf_base);
			if (buf_base == NULL) {
				dev_err(xpc_part, "unable to kmalloc "
					"len=0x%016lx\n", buf_len);
				status = SALRET_ERROR;
				break;
			}
		}

		bte_res = xp_bte_copy(rp_pa, buf, buf_len,
					(BTE_NOTIFY | BTE_WACQUIRE), NULL);
		if (bte_res != BTE_SUCCESS) {
			dev_dbg(xpc_part, "xp_bte_copy failed %i\n", bte_res);
			status = SALRET_ERROR;
			break;
		}
	}

	kfree(buf_base);

	if (status != SALRET_OK) {
		rp_pa = 0;
	}
	dev_dbg(xpc_part, "reserved page at phys address 0x%016lx\n", rp_pa);
	return rp_pa;
}


/*
 * Fill the partition reserved page with the information needed by
 * other partitions to discover we are alive and establish initial
 * communications.
 */
struct xpc_rsvd_page *
xpc_rsvd_page_init(void)
{
	struct xpc_rsvd_page *rp;
	AMO_t *amos_page;
	u64 rp_pa, nasid_array = 0;
	int i, ret;


	/* get the local reserved page's address */

	preempt_disable();
	rp_pa = xpc_get_rsvd_page_pa(cpuid_to_nasid(smp_processor_id()));
	preempt_enable();
	if (rp_pa == 0) {
		dev_err(xpc_part, "SAL failed to locate the reserved page\n");
		return NULL;
	}
	rp = (struct xpc_rsvd_page *) __va(rp_pa);

	if (rp->partid != sn_partition_id) {
		dev_err(xpc_part, "the reserved page's partid of %d should be "
			"%d\n", rp->partid, sn_partition_id);
		return NULL;
	}

	rp->version = XPC_RP_VERSION;

	/* establish the actual sizes of the nasid masks */
	if (rp->SAL_version == 1) {
		/* SAL_version 1 didn't set the nasids_size field */
		rp->nasids_size = 128;
	}
	xp_nasid_mask_bytes = rp->nasids_size;
	xp_nasid_mask_words = xp_nasid_mask_bytes / 8;

	/* setup the pointers to the various items in the reserved page */
	xpc_part_nasids = XPC_RP_PART_NASIDS(rp);
	xpc_mach_nasids = XPC_RP_MACH_NASIDS(rp);
	xpc_vars = XPC_RP_VARS(rp);
	xpc_vars_part = XPC_RP_VARS_PART(rp);

	/*
	 * Before clearing xpc_vars, see if a page of AMOs had been previously
	 * allocated. If not we'll need to allocate one and set permissions
	 * so that cross-partition AMOs are allowed.
	 *
	 * The allocated AMO page needs MCA reporting to remain disabled after
	 * XPC has unloaded.  To make this work, we keep a copy of the pointer
	 * to this page (i.e., amos_page) in the struct xpc_vars structure,
	 * which is pointed to by the reserved page, and re-use that saved copy
	 * on subsequent loads of XPC. This AMO page is never freed, and its
	 * memory protections are never restricted.
	 */
	if ((amos_page = xpc_vars->amos_page) == NULL) {
		amos_page = (AMO_t *) TO_AMO(uncached_alloc_page(0));
		if (amos_page == NULL) {
			dev_err(xpc_part, "can't allocate page of AMOs\n");
			return NULL;
		}

		/*
		 * Open up AMO-R/W to cpu.  This is done for Shub 1.1 systems
		 * when xpc_allow_IPI_ops() is called via xpc_hb_init().
		 */
		if (!enable_shub_wars_1_1()) {
			ret = sn_change_memprotect(ia64_tpa((u64) amos_page),
					PAGE_SIZE, SN_MEMPROT_ACCESS_CLASS_1,
					&nasid_array);
			if (ret != 0) {
				dev_err(xpc_part, "can't change memory "
					"protections\n");
				uncached_free_page(__IA64_UNCACHED_OFFSET |
						   TO_PHYS((u64) amos_page));
				return NULL;
			}
		}
	} else if (!IS_AMO_ADDRESS((u64) amos_page)) {
		/*
		 * EFI's XPBOOT can also set amos_page in the reserved page,
		 * but it happens to leave it as an uncached physical address
		 * and we need it to be an uncached virtual, so we'll have to
		 * convert it.
		 */
		if (!IS_AMO_PHYS_ADDRESS((u64) amos_page)) {
			dev_err(xpc_part, "previously used amos_page address "
				"is bad = 0x%p\n", (void *) amos_page);
			return NULL;
		}
		amos_page = (AMO_t *) TO_AMO((u64) amos_page);
	}

	/* clear xpc_vars */
	memset(xpc_vars, 0, sizeof(struct xpc_vars));

	xpc_vars->version = XPC_V_VERSION;
	xpc_vars->act_nasid = cpuid_to_nasid(0);
	xpc_vars->act_phys_cpuid = cpu_physical_id(0);
	xpc_vars->vars_part_pa = __pa(xpc_vars_part);
	xpc_vars->amos_page_pa = ia64_tpa((u64) amos_page);
	xpc_vars->amos_page = amos_page;  /* save for next load of XPC */


	/* clear xpc_vars_part */
	memset((u64 *) xpc_vars_part, 0, sizeof(struct xpc_vars_part) *
							XP_MAX_PARTITIONS);

	/* initialize the activate IRQ related AMO variables */
	for (i = 0; i < xp_nasid_mask_words; i++) {
		(void) xpc_IPI_init(XPC_ACTIVATE_IRQ_AMOS + i);
	}

	/* initialize the engaged remote partitions related AMO variables */
	(void) xpc_IPI_init(XPC_ENGAGED_PARTITIONS_AMO);
	(void) xpc_IPI_init(XPC_DISENGAGE_REQUEST_AMO);

	/* timestamp of when reserved page was setup by XPC */
	rp->stamp = CURRENT_TIME;

	/*
	 * This signifies to the remote partition that our reserved
	 * page is initialized.
	 */
	rp->vars_pa = __pa(xpc_vars);

	return rp;
}


/*
 * Change protections to allow IPI operations (and AMO operations on
 * Shub 1.1 systems).
 */
void
xpc_allow_IPI_ops(void)
{
	int node;
	int nasid;


	// >>> Change SH_IPI_ACCESS code to use SAL call once it is available.

	if (is_shub2()) {
		xpc_sh2_IPI_access0 =
			(u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS0));
		xpc_sh2_IPI_access1 =
			(u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS1));
		xpc_sh2_IPI_access2 =
			(u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS2));
		xpc_sh2_IPI_access3 =
			(u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH2_IPI_ACCESS3));

		for_each_online_node(node) {
			nasid = cnodeid_to_nasid(node);
			HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
								-1UL);
			HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
								-1UL);
			HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
								-1UL);
			HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
								-1UL);
		}

	} else {
		xpc_sh1_IPI_access =
			(u64) HUB_L((u64 *) LOCAL_MMR_ADDR(SH1_IPI_ACCESS));

		for_each_online_node(node) {
			nasid = cnodeid_to_nasid(node);
			HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
								-1UL);

			/*
			 * Since the BIST collides with memory operations on
			 * SHUB 1.1 sn_change_memprotect() cannot be used.
			 */
			if (enable_shub_wars_1_1()) {
				/* open up everything */
				xpc_prot_vec[node] = (u64) HUB_L((u64 *)
						GLOBAL_MMR_ADDR(nasid,
						SH1_MD_DQLP_MMR_DIR_PRIVEC0));
				HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
						SH1_MD_DQLP_MMR_DIR_PRIVEC0),
								-1UL);
				HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
						SH1_MD_DQRP_MMR_DIR_PRIVEC0),
								-1UL);
			}
		}
	}
}


/*
 * Restrict protections to disallow IPI operations (and AMO operations on
 * Shub 1.1 systems).
 */
void
xpc_restrict_IPI_ops(void)
{
	int node;
	int nasid;


	// >>> Change SH_IPI_ACCESS code to use SAL call once it is available.

	if (is_shub2()) {

		for_each_online_node(node) {
			nasid = cnodeid_to_nasid(node);
			HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS0),
							xpc_sh2_IPI_access0);
			HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS1),
							xpc_sh2_IPI_access1);
			HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS2),
							xpc_sh2_IPI_access2);
			HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH2_IPI_ACCESS3),
							xpc_sh2_IPI_access3);
		}

	} else {

		for_each_online_node(node) {
			nasid = cnodeid_to_nasid(node);
			HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid, SH1_IPI_ACCESS),
							xpc_sh1_IPI_access);

			if (enable_shub_wars_1_1()) {
				HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
						SH1_MD_DQLP_MMR_DIR_PRIVEC0),
							xpc_prot_vec[node]);
				HUB_S((u64 *) GLOBAL_MMR_ADDR(nasid,
						SH1_MD_DQRP_MMR_DIR_PRIVEC0),
							xpc_prot_vec[node]);
			}
		}
	}
}


/*
 * At periodic intervals, scan through all active partitions and ensure
 * their heartbeat is still active.  If not, the partition is deactivated.