Skip to contentSkip to conversation

arch/x86/platform/uv: add UV4 support to BAU

PATCH Superseded Andrew Banman 4 participants 34 messages Sep 20, 2016 – Sep 22, 2016

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

cover arch/x86/platform/uv: add UV4 support to BAU

The following patch set adds support for UV4 architecture to the Broadcast Assist Unit (BAU). Major hardware changes to the BAU require these fixes to ensure correct operation and to avoid illegal MMR writes.

The patch set can be thought of in three logical groups:

1) General cleanup.

[PATCH 1/9] arch/x86/platform/uv: BAU cleanup: update printks [PATCH 2/9] arch/x86/platform/uv: BAU cleanup: pq_init [PATCH 3/9] arch/x86/platform/uv: BAU replace uv_physnodeaddr

These housekeeping patches make the subsequent UV4 patches clearer, and they should be done in any case.

2) Implement a new scheme to abstract UV version-specific functions.

[PATCH 4/9] arch/x86/platform/uv: BAU add generic function pointers [PATCH 5/9] arch/x86/platform/uv: BAU use generic function pointers

We add a struct of function pointers to define version-specific BAU operations. The philosophy is to abstract functions that perform the same operation on all UV versions but have different implementations. This will simplify their use in the body of the driver code and greatly simplify the UV4 patches to follow.

3) Add UV4 functionality.

[PATCH 6/9] arch/x86/platform/uv: BAU UV4 populate uvhub_version [PATCH 7/9] arch/x86/platform/uv: BAU UV4 disable software timeout [PATCH 8/9] arch/x86/platform/uv: BAU UV4 fix payload queue setup [PATCH 9/9] arch/x86/platform/uv: BAU UV4 add version-specific

These patches feature a minimal set of changes to make the BAU on UV4 operational.

This patch set has been tested for regressions on pre-UV4 architectures and for correct functionality on UV4. The patches apply cleanly to 4.8-rc7. Fine-tuned performance tweaking for UV4 will come in a future patch set.

Thank you,

Andrew Banman

#1 lore raw

PATCH 1/9 arch/x86/platform/uv: BAU cleanup: update printks

Replace all uses of printk with the appropriate pr_* function.

---
 arch/x86/platform/uv/tlb_uv.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)
arch/x86/platform/uv/tlb_uv.c +13 -13
index fdb4d42..7baea46 100644
@@ -1497,16 +1497,16 @@ static ssize_t ptc_proc_write(struct file *file, const char __user *user,
 	}
 
 	if (kstrtol(optstr, 10, &input_arg) < 0) {
-		printk(KERN_DEBUG "%s is invalid\n", optstr);
+		pr_debug("%s is invalid\n", optstr);
 		return -EINVAL;
 	}
 
 	if (input_arg == 0) {
 		elements = ARRAY_SIZE(stat_description);
-		printk(KERN_DEBUG "# cpu:      cpu number\n");
-		printk(KERN_DEBUG "Sender statistics:\n");
+		pr_debug("# cpu:      cpu number\n");
+		pr_debug("Sender statistics:\n");
 		for (i = 0; i < elements; i++)
-			printk(KERN_DEBUG "%s\n", stat_description[i]);
+			pr_debug("%s\n", stat_description[i]);
 	} else if (input_arg == -1) {
 		for_each_present_cpu(cpu) {
 			stat = &per_cpu(ptcstats, cpu);
@@ -1554,7 +1554,7 @@ static int parse_tunables_write(struct bau_control *bcp, char *instr,
 			break;
 	}
 	if (cnt != e) {
-		printk(KERN_INFO "bau tunable error: should be %d values\n", e);
+		pr_info("bau tunable error: should be %d values\n", e);
 		return -EINVAL;
 	}
 
@@ -1571,7 +1571,7 @@ static int parse_tunables_write(struct bau_control *bcp, char *instr,
 				continue;
 			}
 			if (val < 1 || val > bcp->cpus_in_uvhub) {
-				printk(KERN_DEBUG
+				pr_debug(
 				"Error: BAU max concurrent %d is invalid\n",
 				val);
 				return -EINVAL;
@@ -1676,21 +1676,21 @@ static int __init uv_ptc_init(void)
 	proc_uv_ptc = proc_create(UV_PTC_BASENAME, 0444, NULL,
 				  &proc_uv_ptc_operations);
 	if (!proc_uv_ptc) {
-		printk(KERN_ERR "unable to create %s proc entry\n",
+		pr_err("unable to create %s proc entry\n",
 		       UV_PTC_BASENAME);
 		return -EINVAL;
 	}
 
 	tunables_dir = debugfs_create_dir(UV_BAU_TUNABLES_DIR, NULL);
 	if (!tunables_dir) {
-		printk(KERN_ERR "unable to create debugfs directory %s\n",
+		pr_err("unable to create debugfs directory %s\n",
 		       UV_BAU_TUNABLES_DIR);
 		return -EINVAL;
 	}
 	tunables_file = debugfs_create_file(UV_BAU_TUNABLES_FILE, 0600,
 					tunables_dir, NULL, &tunables_fops);
 	if (!tunables_file) {
-		printk(KERN_ERR "unable to create debugfs file %s\n",
+		pr_err("unable to create debugfs file %s\n",
 		       UV_BAU_TUNABLES_FILE);
 		return -EINVAL;
 	}
@@ -1944,7 +1944,7 @@ static int __init get_cpu_topology(int base_pnode,
 
 		pnode = uv_cpu_hub_info(cpu)->pnode;
 		if ((pnode - base_pnode) >= UV_DISTRIBUTION_SIZE) {
-			printk(KERN_EMERG
+			pr_emerg(
 				"cpu %d pnode %d-%d beyond %d; BAU disabled\n",
 				cpu, pnode, base_pnode, UV_DISTRIBUTION_SIZE);
 			return 1;
@@ -1969,7 +1969,7 @@ static int __init get_cpu_topology(int base_pnode,
 		sdp->cpu_number[sdp->num_cpus] = cpu;
 		sdp->num_cpus++;
 		if (sdp->num_cpus > MAX_CPUS_PER_SOCKET) {
-			printk(KERN_EMERG "%d cpus per socket invalid\n",
+			pr_emerg("%d cpus per socket invalid\n",
 				sdp->num_cpus);
 			return 1;
 		}
@@ -2036,14 +2036,14 @@ static int scan_sock(struct socket_desc *sdp, struct uvhub_desc *bdp,
 		else if (is_uv3_hub())
 			bcp->uvhub_version = 3;
 		else {
-			printk(KERN_EMERG "uvhub version not 1, 2 or 3\n");
+			pr_emerg("uvhub version not 1, 2, or 3\n");
 			return 1;
 		}
 		bcp->uvhub_master = *hmasterp;
 		bcp->uvhub_cpu = uv_cpu_blade_processor_id(cpu);
 
 		if (bcp->uvhub_cpu >= MAX_CPUS_PER_UVHUB) {
-			printk(KERN_EMERG "%d cpus per uvhub invalid\n",
+			pr_emerg("%d cpus per uvhub invalid\n",
 				bcp->uvhub_cpu);
 			return 1;
 		}
Signature · 1 line
1.8.5.6
#2 lore raw

PATCH 2/9 arch/x86/platform/uv: BAU cleanup: pq_init

In reply to Andrew Banman

The payload queue first MMR requires the physical memory address and hub GNODE where the payload queue, but the associated variables are named as if the PNODE were used. Rename gnode-related variables and clarify the definitions of the payload queue head, last, and tail pointers.

---
 arch/x86/include/asm/uv/uv_bau.h |  2 +-
 arch/x86/platform/uv/tlb_uv.c    | 19 ++++++++-----------
 2 files changed, 9 insertions(+), 12 deletions(-)
arch/x86/include/asm/uv/uv_bau.h +1 -1
index cc44d92..cc058c6 100644
@@ -49,7 +49,7 @@
 #define UV_NET_ENDPOINT_INTD		(is_uv1_hub() ?			\
 			UV1_NET_ENDPOINT_INTD : UV2_NET_ENDPOINT_INTD)
 #define UV_DESC_PSHIFT			49
-#define UV_PAYLOADQ_PNODE_SHIFT		49
+#define UV_PAYLOADQ_GNODE_SHIFT		49
 #define UV_PTC_BASENAME			"sgi_uv/ptc_statistics"
 #define UV_BAU_BASENAME			"sgi_uv/bau_tunables"
 #define UV_BAU_TUNABLES_DIR		"sgi_uv"
arch/x86/platform/uv/tlb_uv.c +8 -11
index 7baea46..e8c2d2f 100644
@@ -1790,10 +1790,7 @@ static void pq_init(int node, int pnode)
 	size_t plsize;
 	char *cp;
 	void *vp;
-	unsigned long pn;
-	unsigned long first;
-	unsigned long pn_first;
-	unsigned long last;
+	unsigned long gnode, first, last, tail;
 	struct bau_pq_entry *pqp;
 	struct bau_control *bcp;
 
@@ -1814,16 +1811,16 @@ static void pq_init(int node, int pnode)
 		bcp->bau_msg_head	= pqp;
 		bcp->queue_last		= pqp + (DEST_Q_SIZE - 1);
 	}
-	/*
-	 * need the gnode of where the memory was really allocated
-	 */
-	pn = uv_gpa_to_gnode(uv_gpa(pqp));
+
 	first = uv_physnodeaddr(pqp);
-	pn_first = ((unsigned long)pn << UV_PAYLOADQ_PNODE_SHIFT) | first;
 	last = uv_physnodeaddr(pqp + (DEST_Q_SIZE - 1));
-	write_mmr_payload_first(pnode, pn_first);
-	write_mmr_payload_tail(pnode, first);
+	tail = first;
+	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
+	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
+
+	write_mmr_payload_first(pnode, first);
 	write_mmr_payload_last(pnode, last);
+	write_mmr_payload_tail(pnode, tail);
 	write_gmmr_sw_ack(pnode, 0xffffUL);
 
 	/* in effect, all msg_type's are set to MSG_NOOP */
Signature · 1 line
1.8.5.6
#3 lore raw

PATCH 3/9 arch/x86/platform/uv: BAU replace uv_physnodeaddr

In reply to Andrew Banman

The BAU driver should use the functions provided by uv_hub.h rather than its own implementations. uv_physnodeaddr converts vaddrs to paddrs for BAU MMR fields, but this is done better by uv_gpa_to_offset.

---
 arch/x86/include/asm/uv/uv_bau.h | 2 --
 arch/x86/platform/uv/tlb_uv.c    | 4 ++--
 2 files changed, 2 insertions(+), 4 deletions(-)
arch/x86/include/asm/uv/uv_bau.h +0 -2
index cc058c6..a46f270 100644
@@ -55,8 +55,6 @@
 #define UV_BAU_TUNABLES_DIR		"sgi_uv"
 #define UV_BAU_TUNABLES_FILE		"bau_tunables"
 #define WHITESPACE			" \t\n"
-#define uv_mmask			((1UL << uv_hub_info->m_val) - 1)
-#define uv_physnodeaddr(x)		((__pa((unsigned long)(x)) & uv_mmask))
 #define cpubit_isset(cpu, bau_local_cpumask) \
 	test_bit((cpu), (bau_local_cpumask).bits)
 
arch/x86/platform/uv/tlb_uv.c +2 -2
index e8c2d2f..98f1d43 100644
@@ -1812,8 +1812,8 @@ static void pq_init(int node, int pnode)
 		bcp->queue_last		= pqp + (DEST_Q_SIZE - 1);
 	}
 
-	first = uv_physnodeaddr(pqp);
-	last = uv_physnodeaddr(pqp + (DEST_Q_SIZE - 1));
+	first = uv_gpa_to_offset(uv_gpa(pqp));
+	last = uv_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
 	tail = first;
 	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
 	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
Signature · 1 line
1.8.5.6
#4 lore raw

PATCH 4/9 arch/x86/platform/uv: BAU add generic function pointers

In reply to Andrew Banman

Many BAU functions have different implementations depending on the UV version. Rather than switching on the uvhub_version throughout the driver, we can define a set of operations for each version. This is especially beneficial for UV4, which will require many new MMR read/write functions.

Currently, the set of abstracted functions are the same for UV1, UV2, and UV3. The functions were chosen because each one will have a different implementation for UV4. Other functions will be added as needed to handle new implementations or to cleanup the existing differences between UV1, UV2, and UV3, i.e. read_status and wait_completion.

---
 arch/x86/include/asm/uv/uv_bau.h | 11 +++++++++++
 arch/x86/platform/uv/tlb_uv.c    | 19 +++++++++++++++++++
 2 files changed, 30 insertions(+)
arch/x86/include/asm/uv/uv_bau.h +11 -0
index a46f270..a7a93a5 100644
@@ -385,6 +385,17 @@ struct uv2_3_bau_msg_header {
 	/* bits 127:120 */
 };
 
+/* Abstracted BAU functions */
+struct bau_operations {
+	unsigned long (*read_l_sw_ack)(void);
+	unsigned long (*read_g_sw_ack)(int pnode);
+	unsigned long (*bau_gpa_to_offset)(unsigned long vaddr);
+	void (*write_l_sw_ack)(unsigned long mmr);
+	void (*write_g_sw_ack)(int pnode, unsigned long mmr);
+	void (*write_payload_first)(int pnode, unsigned long mmr);
+	void (*write_payload_last)(int pnode, unsigned long mmr);
+};
+
 /*
  * The activation descriptor:
  * The format of the message to send, plus all accompanying control
arch/x86/platform/uv/tlb_uv.c +19 -0
index 98f1d43..045da97 100644
@@ -24,6 +24,18 @@
 #include <asm/irq_vectors.h>
 #include <asm/timer.h>
 
+static struct bau_operations ops;
+
+static struct bau_operations uv123_bau_ops = {
+	.bau_gpa_to_offset = uv_gpa_to_offset,
+	.read_l_sw_ack = read_mmr_sw_ack,
+	.read_g_sw_ack = read_gmmr_sw_ack,
+	.write_l_sw_ack = write_mmr_sw_ack,
+	.write_g_sw_ack = write_gmmr_sw_ack,
+	.write_payload_first = write_mmr_payload_first,
+	.write_payload_last = write_mmr_payload_last,
+};
+
 /* timeouts in nanoseconds (indexed by UVH_AGING_PRESCALE_SEL urgency7 30:28) */
 static int timeout_base_ns[] = {
 		20,
@@ -2135,6 +2147,13 @@ static int __init uv_bau_init(void)
 	if (!is_uv_system())
 		return 0;
 
+	if (is_uv3_hub())
+		ops = uv123_bau_ops;
+	else if (is_uv2_hub())
+		ops = uv123_bau_ops;
+	else if (is_uv1_hub())
+		ops = uv123_bau_ops;
+
 	for_each_possible_cpu(cur_cpu) {
 		mask = &per_cpu(uv_flush_tlb_mask, cur_cpu);
 		zalloc_cpumask_var_node(mask, GFP_KERNEL, cpu_to_node(cur_cpu));
Signature · 1 line
1.8.5.6
#5 lore raw

PATCH 5/9 arch/x86/platform/uv: BAU use generic function pointers

In reply to Andrew Banman

Convert the use of UV version-specific functions to their abstracted counterparts.

---
 arch/x86/platform/uv/tlb_uv.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)
arch/x86/platform/uv/tlb_uv.c +13 -13
index 045da97..bfeb382 100644
@@ -228,7 +228,7 @@ static void reply_to_message(struct msg_desc *mdp, struct bau_control *bcp,
 	msg = mdp->msg;
 	if (!msg->canceled && do_acknowledge) {
 		dw = (msg->swack_vec << UV_SW_ACK_NPENDING) | msg->swack_vec;
-		write_mmr_sw_ack(dw);
+		ops.write_l_sw_ack(dw);
 	}
 	msg->replied_to = 1;
 	msg->swack_vec = 0;
@@ -264,7 +264,7 @@ static void bau_process_retry_msg(struct msg_desc *mdp,
 			msg->swack_vec) == 0) &&
 		    (msg2->sending_cpu == msg->sending_cpu) &&
 		    (msg2->msg_type != MSG_NOOP)) {
-			mmr = read_mmr_sw_ack();
+			mmr = ops.read_l_sw_ack();
 			msg_res = msg2->swack_vec;
 			/*
 			 * This is a message retry; clear the resources held
@@ -282,7 +282,7 @@ static void bau_process_retry_msg(struct msg_desc *mdp,
 				stat->d_canceled++;
 				cancel_count++;
 				mr = (msg_res << UV_SW_ACK_NPENDING) | msg_res;
-				write_mmr_sw_ack(mr);
+				ops.write_l_sw_ack(mr);
 			}
 		}
 	}
@@ -415,12 +415,12 @@ static void do_reset(void *ptr)
 			/*
 			 * only reset the resource if it is still pending
 			 */
-			mmr = read_mmr_sw_ack();
+			mmr = ops.read_l_sw_ack();
 			msg_res = msg->swack_vec;
 			mr = (msg_res << UV_SW_ACK_NPENDING) | msg_res;
 			if (mmr & msg_res) {
 				stat->d_rcanceled++;
-				write_mmr_sw_ack(mr);
+				ops.write_l_sw_ack(mr);
 			}
 		}
 	}
@@ -1214,7 +1214,7 @@ void process_uv2_message(struct msg_desc *mdp, struct bau_control *bcp)
 	struct bau_pq_entry *msg = mdp->msg;
 	struct bau_pq_entry *other_msg;
 
-	mmr_image = read_mmr_sw_ack();
+	mmr_image = ops.read_l_sw_ack();
 	swack_vec = msg->swack_vec;
 
 	if ((swack_vec & mmr_image) == 0) {
@@ -1443,7 +1443,7 @@ static int ptc_seq_show(struct seq_file *file, void *data)
 		/* destination side statistics */
 		seq_printf(file,
 			"%lx %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld\n",
-			   read_gmmr_sw_ack(uv_cpu_to_pnode(cpu)),
+			   ops.read_g_sw_ack(uv_cpu_to_pnode(cpu)),
 			   stat->d_requestee, cycles_2_us(stat->d_time),
 			   stat->d_alltlb, stat->d_onetlb, stat->d_multmsg,
 			   stat->d_nomsg, stat->d_retries, stat->d_canceled,
@@ -1737,7 +1737,7 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode)
 
 	gpa = uv_gpa(bau_desc);
 	n = uv_gpa_to_gnode(gpa);
-	m = uv_gpa_to_offset(gpa);
+	m = ops.bau_gpa_to_offset(gpa);
 	if (is_uv1_hub())
 		uv1 = 1;
 
@@ -1824,16 +1824,16 @@ static void pq_init(int node, int pnode)
 		bcp->queue_last		= pqp + (DEST_Q_SIZE - 1);
 	}
 
-	first = uv_gpa_to_offset(uv_gpa(pqp));
-	last = uv_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
+	first = ops.bau_gpa_to_offset(uv_gpa(pqp));
+	last = ops.bau_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
 	tail = first;
 	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
 	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
 
-	write_mmr_payload_first(pnode, first);
-	write_mmr_payload_last(pnode, last);
 	write_mmr_payload_tail(pnode, tail);
-	write_gmmr_sw_ack(pnode, 0xffffUL);
+	ops.write_payload_first(pnode, first);
+	ops.write_payload_last(pnode, last);
+	ops.write_g_sw_ack(pnode, 0xffffUL);
 
 	/* in effect, all msg_type's are set to MSG_NOOP */
 	memset(pqp, 0, sizeof(struct bau_pq_entry) * DEST_Q_SIZE);
Signature · 1 line
1.8.5.6
#6 lore raw

PATCH 6/9 arch/x86/platform/uv: BAU UV4 populate uvhub_version

In reply to Andrew Banman
---
 arch/x86/platform/uv/tlb_uv.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/platform/uv/tlb_uv.c +3 -1
index bfeb382..67a31d0 100644
@@ -2044,8 +2044,10 @@ static int scan_sock(struct socket_desc *sdp, struct uvhub_desc *bdp,
 			bcp->uvhub_version = 2;
 		else if (is_uv3_hub())
 			bcp->uvhub_version = 3;
+		else if (is_uv4_hub())
+			bcp->uvhub_version = 4;
 		else {
-			pr_emerg("uvhub version not 1, 2, or 3\n");
+			pr_emerg("uvhub version not 1, 2, 3, or 4\n");
 			return 1;
 		}
 		bcp->uvhub_master = *hmasterp;
Signature · 1 line
1.8.5.6
#7 lore raw

PATCH 7/9 arch/x86/platform/uv: BAU UV4 disable software timeout

In reply to Andrew Banman

Software timeouts are not currently supported on BAU for UV4. Instead, the BAU will rely on hardware-level fairness protocols to determine broadcast timeouts.

Do not call enable_timeouts or calculate_destination_timeout on UV4. These functions write to pre-UV4 MMRs so they generate error messages on UV4.

---
 arch/x86/platform/uv/tlb_uv.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
arch/x86/platform/uv/tlb_uv.c +5 -2
index 67a31d0..bfd03db 100644
@@ -2109,7 +2109,8 @@ static int __init init_per_cpu(int nuvhubs, int base_part_pnode)
 	void *vp;
 	struct uvhub_desc *uvhub_descs;
 
-	timeout_us = calculate_destination_timeout();
+	if (is_uv3_hub() || is_uv2_hub() || is_uv1_hub())
+		timeout_us = calculate_destination_timeout();
 
 	vp = kmalloc(nuvhubs * sizeof(struct uvhub_desc), GFP_KERNEL);
 	uvhub_descs = (struct uvhub_desc *)vp;
@@ -2171,7 +2172,9 @@ static int __init uv_bau_init(void)
 			uv_base_pnode = uv_blade_to_pnode(uvhub);
 	}
 
-	enable_timeouts();
+	/* software timeouts are not supported on UV4 */
+	if (is_uv3_hub() || is_uv2_hub() || is_uv1_hub())
+		enable_timeouts();
 
 	if (init_per_cpu(nuvhubs, uv_base_pnode)) {
 		set_bau_off();
Signature · 1 line
1.8.5.6
#8 lore raw

PATCH 8/9 arch/x86/platform/uv: BAU UV4 fix payload queue setup

In reply to Andrew Banman

The BAU on UV4 does not need to maintain the payload queue tail pointer. Do not initialize the tail pointer MMR on UV4.

Note that write_payload_tail is not an abstracted BAU function since it is an operation specific to pre-UV4 versions. Then we must switch on the UV version to control its usage, for which we use uvhub_version rather than is_uv*_hub because it is quicker/more concise.

---
 arch/x86/platform/uv/tlb_uv.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
arch/x86/platform/uv/tlb_uv.c +12 -4
index bfd03db..56d12eb 100644
@@ -1826,11 +1826,19 @@ static void pq_init(int node, int pnode)
 
 	first = ops.bau_gpa_to_offset(uv_gpa(pqp));
 	last = ops.bau_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
-	tail = first;
-	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
-	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
 
-	write_mmr_payload_tail(pnode, tail);
+	/*
+	 * Pre UV4, the gnode is required to locate the payload queue
+	 * and the payload queue tail must be maintained by the kernel.
+	 */
+	bcp = &per_cpu(bau_control, smp_processor_id());
+	if (bcp->uvhub_version <= 3) {
+		tail = first;
+		gnode = uv_gpa_to_gnode(uv_gpa(pqp));
+		first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
+		write_mmr_payload_tail(pnode, tail);
+	}
+
 	ops.write_payload_first(pnode, first);
 	ops.write_payload_last(pnode, last);
 	ops.write_g_sw_ack(pnode, 0xffffUL);
Signature · 1 line
1.8.5.6
#9 lore raw

PATCH 9/9 arch/x86/platform/uv: BAU UV4 add version-specific functions

In reply to Andrew Banman

Add the UV4-specific function definitions and define an operations struct to implement them in the BAU driver.

Many BAU MMRs, although functionally the same, have new addresses on UV4 due to hardware changes. Each MMR requires new read/write functions, but their implementation in the driver does not change. Thus, it is enough to enumerate them in the operations struct for the changes to take effect.

---
 arch/x86/include/asm/uv/uv_bau.h | 30 ++++++++++++++++++++++++++++++
 arch/x86/platform/uv/tlb_uv.c    | 15 ++++++++++++++-
 2 files changed, 44 insertions(+), 1 deletion(-)
arch/x86/include/asm/uv/uv_bau.h +30 -0
index a7a93a5..57ab86d 100644
@@ -664,6 +664,16 @@ static inline void write_gmmr_activation(int pnode, unsigned long mmr_image)
 	write_gmmr(pnode, UVH_LB_BAU_SB_ACTIVATION_CONTROL, mmr_image);
 }
 
+static inline void write_mmr_proc_payload_first(int pnode, unsigned long mmr_image)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_FIRST, mmr_image);
+}
+
+static inline void write_mmr_proc_payload_last(int pnode, unsigned long mmr_image)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_LAST, mmr_image);
+}
+
 static inline void write_mmr_payload_first(int pnode, unsigned long mmr_image)
 {
 	write_gmmr(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST, mmr_image);
@@ -709,6 +719,26 @@ static inline unsigned long read_gmmr_sw_ack(int pnode)
 	return read_gmmr(pnode, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
 }
 
+static inline void write_mmr_proc_sw_ack(unsigned long mr)
+{
+	uv_write_local_mmr(UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
+}
+
+static inline void write_gmmr_proc_sw_ack(int pnode, unsigned long mr)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
+}
+
+static inline unsigned long read_mmr_proc_sw_ack(void)
+{
+	return read_lmmr(UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
+}
+
+static inline unsigned long read_gmmr_proc_sw_ack(int pnode)
+{
+	return read_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
+}
+
 static inline void write_mmr_data_config(int pnode, unsigned long mr)
 {
 	uv_write_global_mmr64(pnode, UVH_BAU_DATA_CONFIG, mr);
arch/x86/platform/uv/tlb_uv.c +14 -1
index 56d12eb..470d73c 100644
@@ -36,6 +36,17 @@ static struct bau_operations uv123_bau_ops = {
 	.write_payload_last = write_mmr_payload_last,
 };
 
+static struct bau_operations uv4_bau_ops = {
+	.bau_gpa_to_offset = uv_gpa_to_soc_phys_ram,
+	.read_l_sw_ack = read_mmr_proc_sw_ack,
+	.read_g_sw_ack = read_gmmr_proc_sw_ack,
+	.write_l_sw_ack = write_mmr_proc_sw_ack,
+	.write_g_sw_ack = write_gmmr_proc_sw_ack,
+	.write_payload_first = write_mmr_proc_payload_first,
+	.write_payload_last = write_mmr_proc_payload_last,
+};
+
+
 /* timeouts in nanoseconds (indexed by UVH_AGING_PRESCALE_SEL urgency7 30:28) */
 static int timeout_base_ns[] = {
 		20,
@@ -2158,7 +2169,9 @@ static int __init uv_bau_init(void)
 	if (!is_uv_system())
 		return 0;
 
-	if (is_uv3_hub())
+	if (is_uv4_hub())
+		ops = uv4_bau_ops;
+	else if (is_uv3_hub())
 		ops = uv123_bau_ops;
 	else if (is_uv2_hub())
 		ops = uv123_bau_ops;
Signature · 1 line
1.8.5.6
#10 lore raw

Re: [PATCH 9/9] arch/x86/platform/uv: BAU UV4 add version-specific functions

Quoted 71 lines · Andrew Banman

Add the UV4-specific function definitions and define an operations struct to implement them in the BAU driver.

Many BAU MMRs, although functionally the same, have new addresses on UV4 due to hardware changes. Each MMR requires new read/write functions, but their implementation in the driver does not change. Thus, it is enough to enumerate them in the operations struct for the changes to take effect.

---
 arch/x86/include/asm/uv/uv_bau.h | 30 ++++++++++++++++++++++++++++++
 arch/x86/platform/uv/tlb_uv.c    | 15 ++++++++++++++-
 2 files changed, 44 insertions(+), 1 deletion(-)
arch/x86/include/asm/uv/uv_bau.h +30 -0
index a7a93a5..57ab86d 100644
@@ -664,6 +664,16 @@ static inline void write_gmmr_activation(int pnode, unsigned long mmr_image)
 	write_gmmr(pnode, UVH_LB_BAU_SB_ACTIVATION_CONTROL, mmr_image);
 }
 
+static inline void write_mmr_proc_payload_first(int pnode, unsigned long mmr_image)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_FIRST, mmr_image);
+}
+
+static inline void write_mmr_proc_payload_last(int pnode, unsigned long mmr_image)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_LAST, mmr_image);
+}
+
 static inline void write_mmr_payload_first(int pnode, unsigned long mmr_image)
 {
 	write_gmmr(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST, mmr_image);
@@ -709,6 +719,26 @@ static inline unsigned long read_gmmr_sw_ack(int pnode)
 	return read_gmmr(pnode, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
 }
 
+static inline void write_mmr_proc_sw_ack(unsigned long mr)
+{
+	uv_write_local_mmr(UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
+}
+
+static inline void write_gmmr_proc_sw_ack(int pnode, unsigned long mr)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
+}
+
+static inline unsigned long read_mmr_proc_sw_ack(void)
+{
+	return read_lmmr(UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
+}
+
+static inline unsigned long read_gmmr_proc_sw_ack(int pnode)
+{
+	return read_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
+}
+
 static inline void write_mmr_data_config(int pnode, unsigned long mr)
 {
 	uv_write_global_mmr64(pnode, UVH_BAU_DATA_CONFIG, mr);
arch/x86/platform/uv/tlb_uv.c +0 -0
index 56d12eb..470d73c 100644
@@ -36,6 +36,17 @@ static struct bau_operations uv123_bau_ops = {
 	.write_payload_last = write_mmr_payload_last,
 };

arch/x86/platform/uv/tlb_uv.c static struct bau_operations uv123_bau_ops = { · line 47

Quoted 10 lines
arch/x86/platform/uv/tlb_uv.c +9 -0
 
+static struct bau_operations uv4_bau_ops = {
+	.bau_gpa_to_offset = uv_gpa_to_soc_phys_ram,
+	.read_l_sw_ack = read_mmr_proc_sw_ack,
+	.read_g_sw_ack = read_gmmr_proc_sw_ack,
+	.write_l_sw_ack = write_mmr_proc_sw_ack,
+	.write_g_sw_ack = write_gmmr_proc_sw_ack,
+	.write_payload_first = write_mmr_proc_payload_first,
+	.write_payload_last = write_mmr_proc_payload_last,
+};

The series looks mostly good to me, only a minor nit: could you please organize such initializations vertically?

Something like:

	.bau_gpa_to_offset	= uv_gpa_to_soc_phys_ram,
	.read_l_sw_ack		= read_mmr_proc_sw_ack,
	.read_g_sw_ack		= read_gmmr_proc_sw_ack,
	.write_l_sw_ack		= write_mmr_proc_sw_ack,
	.write_g_sw_ack		= write_gmmr_proc_sw_ack,
	.write_payload_first	= write_mmr_proc_payload_first,
	.write_payload_last	= write_mmr_proc_payload_last,

would make it more readable. Same might apply to other patches too in the series.

Plus it might make sense to do the same to the existing tunables[] structure as well.

Thanks,

Ingo

#11 lore raw

Re: [PATCH 9/9] arch/x86/platform/uv: BAU UV4 add version-specific functions

Quoted 98 lines · Ingo Molnar
Quoted 71 lines · Andrew Banman

Add the UV4-specific function definitions and define an operations struct to implement them in the BAU driver.

Many BAU MMRs, although functionally the same, have new addresses on UV4 due to hardware changes. Each MMR requires new read/write functions, but their implementation in the driver does not change. Thus, it is enough to enumerate them in the operations struct for the changes to take effect.

---
 arch/x86/include/asm/uv/uv_bau.h | 30 ++++++++++++++++++++++++++++++
 arch/x86/platform/uv/tlb_uv.c    | 15 ++++++++++++++-
 2 files changed, 44 insertions(+), 1 deletion(-)
arch/x86/include/asm/uv/uv_bau.h +30 -0
index a7a93a5..57ab86d 100644
@@ -664,6 +664,16 @@ static inline void write_gmmr_activation(int pnode, unsigned long mmr_image)
 	write_gmmr(pnode, UVH_LB_BAU_SB_ACTIVATION_CONTROL, mmr_image);
 }
 
+static inline void write_mmr_proc_payload_first(int pnode, unsigned long mmr_image)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_FIRST, mmr_image);
+}
+
+static inline void write_mmr_proc_payload_last(int pnode, unsigned long mmr_image)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_LAST, mmr_image);
+}
+
 static inline void write_mmr_payload_first(int pnode, unsigned long mmr_image)
 {
 	write_gmmr(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST, mmr_image);
@@ -709,6 +719,26 @@ static inline unsigned long read_gmmr_sw_ack(int pnode)
 	return read_gmmr(pnode, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
 }
 
+static inline void write_mmr_proc_sw_ack(unsigned long mr)
+{
+	uv_write_local_mmr(UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
+}
+
+static inline void write_gmmr_proc_sw_ack(int pnode, unsigned long mr)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
+}
+
+static inline unsigned long read_mmr_proc_sw_ack(void)
+{
+	return read_lmmr(UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
+}
+
+static inline unsigned long read_gmmr_proc_sw_ack(int pnode)
+{
+	return read_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
+}
+
 static inline void write_mmr_data_config(int pnode, unsigned long mr)
 {
 	uv_write_global_mmr64(pnode, UVH_BAU_DATA_CONFIG, mr);
arch/x86/platform/uv/tlb_uv.c +0 -0
index 56d12eb..470d73c 100644
@@ -36,6 +36,17 @@ static struct bau_operations uv123_bau_ops = {
 	.write_payload_last = write_mmr_payload_last,
 };

arch/x86/platform/uv/tlb_uv.c static struct bau_operations uv123_bau_ops = { · line 47

Quoted 10 lines
arch/x86/platform/uv/tlb_uv.c +9 -0
 
+static struct bau_operations uv4_bau_ops = {
+	.bau_gpa_to_offset = uv_gpa_to_soc_phys_ram,
+	.read_l_sw_ack = read_mmr_proc_sw_ack,
+	.read_g_sw_ack = read_gmmr_proc_sw_ack,
+	.write_l_sw_ack = write_mmr_proc_sw_ack,
+	.write_g_sw_ack = write_gmmr_proc_sw_ack,
+	.write_payload_first = write_mmr_proc_payload_first,
+	.write_payload_last = write_mmr_proc_payload_last,
+};

The series looks mostly good to me, only a minor nit: could you please organize such initializations vertically?

Something like:

	.bau_gpa_to_offset	= uv_gpa_to_soc_phys_ram,
	.read_l_sw_ack		= read_mmr_proc_sw_ack,
	.read_g_sw_ack		= read_gmmr_proc_sw_ack,
	.write_l_sw_ack		= write_mmr_proc_sw_ack,
	.write_g_sw_ack		= write_gmmr_proc_sw_ack,
	.write_payload_first	= write_mmr_proc_payload_first,
	.write_payload_last	= write_mmr_proc_payload_last,
Quoted 8 lines

would make it more readable. Same might apply to other patches too in the series.

Plus it might make sense to do the same to the existing tunables[] structure as well.

Thanks,

Ingo

I will add another cleanup patch to fix the alignment of those code blocks and resubmit. Thank you for the review, Ingo!

Andrew

Acked-by on the v1 series
#12 lore raw
In reply to Andrew Banman
Quoted 3 lines · Andrew Banman

The following patch set adds support for UV4 architecture to the Broadcast Assist Unit (BAU). Major hardware changes to the BAU require these fixes to ensure correct operation and to avoid illegal MMR writes.

Nice work and aside of the nitpicks Ingo had:

#13 lore raw

v2 cover arch/x86/platform/uv: add UV4 support to BAU

In reply to Andrew Banman

This is version 2 of a patch set to add UV4 support to the BAU. The only changes from the the original version are whitespace sytle fixes (PATCH 1).

The following patch set adds support for UV4 architecture to the Broadcast Assist Unit (BAU). Major hardware changes to the BAU require these fixes to ensure correct operation and to avoid illegal MMR writes.

The patch set can be thought of in three logical groups:

1) General cleanup.

[PATCH 01/10] arch/x86/platform/uv: BAU cleanup: vertical alignment [PATCH 02/10] arch/x86/platform/uv: BAU cleanup: update printks [PATCH 03/10] arch/x86/platform/uv: BAU cleanup: pq_init [PATCH 04/10] arch/x86/platform/uv: BAU replace uv_physnodeaddr

These housekeeping patches make the subsequent UV4 patches clearer, and they should be done in any case.

2) Implement a new scheme to abstract UV version-specific functions.

[PATCH 05/10] arch/x86/platform/uv: BAU add generic function pointers [PATCH 06/10] arch/x86/platform/uv: BAU use generic function pointers

We add a struct of function pointers to define version-specific BAU operations. The philosophy is to abstract functions that perform the same operation on all UV versions but have different implementations. This will simplify their use in the body of the driver code and greatly simplify the UV4 patches to follow.

3) Add UV4 functionality.

[PATCH 07/10] arch/x86/platform/uv: BAU UV4 populate uvhub_version [PATCH 08/10] arch/x86/platform/uv: BAU UV4 disable software timeout [PATCH 09/10] arch/x86/platform/uv: BAU UV4 fix payload queue setup [PATCH 10/10] arch/x86/platform/uv: BAU UV4 add version-specific

These patches feature a minimal set of changes to make the BAU on UV4 operational.

This patch set has been tested for regressions on pre-UV4 architectures and for correct functionality on UV4. The patches apply cleanly to 4.8-rc7. Fine-tuned performance tweaking for UV4 will come in a future patch set.

Thank you,

Andrew Banman

#14 lore raw

PATCH v2 1/10 arch/x86/platform/uv: BAU cleanup: vertical alignment

Fix whitespace on blocks of code to be vertically aligned.

---
 arch/x86/platform/uv/tlb_uv.c | 64 +++++++++++++++++++++----------------------
 1 file changed, 32 insertions(+), 32 deletions(-)
arch/x86/platform/uv/tlb_uv.c +32 -32
index fdb4d42..b84c2a2 100644
@@ -55,16 +55,16 @@ static int congested_reps	= CONGESTED_REPS;
 static int disabled_period	= DISABLED_PERIOD;
 
 static struct tunables tunables[] = {
-	{&max_concurr, MAX_BAU_CONCURRENT}, /* must be [0] */
-	{&plugged_delay, PLUGGED_DELAY},
-	{&plugsb4reset, PLUGSB4RESET},
-	{&timeoutsb4reset, TIMEOUTSB4RESET},
-	{&ipi_reset_limit, IPI_RESET_LIMIT},
-	{&complete_threshold, COMPLETE_THRESHOLD},
-	{&congested_respns_us, CONGESTED_RESPONSE_US},
-	{&congested_reps, CONGESTED_REPS},
-	{&disabled_period, DISABLED_PERIOD},
-	{&giveup_limit, GIVEUP_LIMIT}
+	{&max_concurr,           MAX_BAU_CONCURRENT}, /* must be [0] */
+	{&plugged_delay,         PLUGGED_DELAY},
+	{&plugsb4reset,          PLUGSB4RESET},
+	{&timeoutsb4reset,       TIMEOUTSB4RESET},
+	{&ipi_reset_limit,       IPI_RESET_LIMIT},
+	{&complete_threshold,    COMPLETE_THRESHOLD},
+	{&congested_respns_us,   CONGESTED_RESPONSE_US},
+	{&congested_reps,        CONGESTED_REPS},
+	{&disabled_period,       DISABLED_PERIOD},
+	{&giveup_limit,          GIVEUP_LIMIT}
 };
 
 static struct dentry *tunables_dir;
@@ -1619,17 +1619,17 @@ static ssize_t tunables_write(struct file *file, const char __user *user,
 
 	for_each_present_cpu(cpu) {
 		bcp = &per_cpu(bau_control, cpu);
-		bcp->max_concurr =		max_concurr;
-		bcp->max_concurr_const =	max_concurr;
-		bcp->plugged_delay =		plugged_delay;
-		bcp->plugsb4reset =		plugsb4reset;
-		bcp->timeoutsb4reset =		timeoutsb4reset;
-		bcp->ipi_reset_limit =		ipi_reset_limit;
-		bcp->complete_threshold =	complete_threshold;
-		bcp->cong_response_us =		congested_respns_us;
-		bcp->cong_reps =		congested_reps;
-		bcp->disabled_period =		sec_2_cycles(disabled_period);
-		bcp->giveup_limit =		giveup_limit;
+		bcp->max_concurr         = max_concurr;
+		bcp->max_concurr_const   = max_concurr;
+		bcp->plugged_delay       = plugged_delay;
+		bcp->plugsb4reset        = plugsb4reset;
+		bcp->timeoutsb4reset     = timeoutsb4reset;
+		bcp->ipi_reset_limit     = ipi_reset_limit;
+		bcp->complete_threshold  = complete_threshold;
+		bcp->cong_response_us    = congested_respns_us;
+		bcp->cong_reps           = congested_reps;
+		bcp->disabled_period     = sec_2_cycles(disabled_period);
+		bcp->giveup_limit        = giveup_limit;
 	}
 	return count;
 }
@@ -1740,7 +1740,7 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode)
 		memset(bd2, 0, sizeof(struct bau_desc));
 		if (uv1) {
 			uv1_hdr = &bd2->header.uv1_hdr;
-			uv1_hdr->swack_flag =	1;
+			uv1_hdr->swack_flag = 1;
 			/*
 			 * The base_dest_nasid set in the message header
 			 * is the nasid of the first uvhub in the partition.
@@ -1749,10 +1749,10 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode)
 			 * if nasid striding is being used.
 			 */
 			uv1_hdr->base_dest_nasid =
-						UV_PNODE_TO_NASID(base_pnode);
-			uv1_hdr->dest_subnodeid =	UV_LB_SUBNODEID;
-			uv1_hdr->command =		UV_NET_ENDPOINT_INTD;
-			uv1_hdr->int_both =		1;
+			                          UV_PNODE_TO_NASID(base_pnode);
+			uv1_hdr->dest_subnodeid  = UV_LB_SUBNODEID;
+			uv1_hdr->command         = UV_NET_ENDPOINT_INTD;
+			uv1_hdr->int_both        = 1;
 			/*
 			 * all others need to be set to zero:
 			 *   fairness chaining multilevel count replied_to
@@ -1763,11 +1763,11 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode)
 			 * uses native mode for selective broadcasts.
 			 */
 			uv2_3_hdr = &bd2->header.uv2_3_hdr;
-			uv2_3_hdr->swack_flag =	1;
+			uv2_3_hdr->swack_flag      = 1;
 			uv2_3_hdr->base_dest_nasid =
-						UV_PNODE_TO_NASID(base_pnode);
-			uv2_3_hdr->dest_subnodeid =	UV_LB_SUBNODEID;
-			uv2_3_hdr->command =		UV_NET_ENDPOINT_INTD;
+			                          UV_PNODE_TO_NASID(base_pnode);
+			uv2_3_hdr->dest_subnodeid  = UV_LB_SUBNODEID;
+			uv2_3_hdr->command         = UV_NET_ENDPOINT_INTD;
 		}
 	}
 	for_each_present_cpu(cpu) {
@@ -1914,8 +1914,8 @@ static void __init init_per_cpu_tunables(void)
 		bcp->complete_threshold		= complete_threshold;
 		bcp->cong_response_us		= congested_respns_us;
 		bcp->cong_reps			= congested_reps;
-		bcp->disabled_period =		sec_2_cycles(disabled_period);
-		bcp->giveup_limit =		giveup_limit;
+		bcp->disabled_period		= sec_2_cycles(disabled_period);
+		bcp->giveup_limit		= giveup_limit;
 		spin_lock_init(&bcp->queue_lock);
 		spin_lock_init(&bcp->uvhub_lock);
 		spin_lock_init(&bcp->disable_lock);
Signature · 1 line
1.8.5.6
#15 lore raw

[tip:x86/platform] x86/platform/uv/BAU: Clean up vertical alignment

x86/platform/uv/BAU: Clean up vertical alignment
Commit-ID:  67492c86b33db0a8a056c72293d4802b37ac8ac6
Gitweb:     http://git.kernel.org/tip/67492c86b33db0a8a056c72293d4802b37ac8ac6
Author:     Andrew Banman <[email protected]>
AuthorDate: Wed, 21 Sep 2016 11:09:12 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 22 Sep 2016 11:16:12 +0200

x86/platform/uv/BAU: Clean up vertical alignment

Fix whitespace on blocks of code to be vertically aligned.

---
 arch/x86/platform/uv/tlb_uv.c | 64 +++++++++++++++++++++----------------------
 1 file changed, 32 insertions(+), 32 deletions(-)
arch/x86/platform/uv/tlb_uv.c +32 -32
index fdb4d42..b84c2a2 100644
@@ -55,16 +55,16 @@ static int congested_reps	= CONGESTED_REPS;
 static int disabled_period	= DISABLED_PERIOD;
 
 static struct tunables tunables[] = {
-	{&max_concurr, MAX_BAU_CONCURRENT}, /* must be [0] */
-	{&plugged_delay, PLUGGED_DELAY},
-	{&plugsb4reset, PLUGSB4RESET},
-	{&timeoutsb4reset, TIMEOUTSB4RESET},
-	{&ipi_reset_limit, IPI_RESET_LIMIT},
-	{&complete_threshold, COMPLETE_THRESHOLD},
-	{&congested_respns_us, CONGESTED_RESPONSE_US},
-	{&congested_reps, CONGESTED_REPS},
-	{&disabled_period, DISABLED_PERIOD},
-	{&giveup_limit, GIVEUP_LIMIT}
+	{&max_concurr,           MAX_BAU_CONCURRENT}, /* must be [0] */
+	{&plugged_delay,         PLUGGED_DELAY},
+	{&plugsb4reset,          PLUGSB4RESET},
+	{&timeoutsb4reset,       TIMEOUTSB4RESET},
+	{&ipi_reset_limit,       IPI_RESET_LIMIT},
+	{&complete_threshold,    COMPLETE_THRESHOLD},
+	{&congested_respns_us,   CONGESTED_RESPONSE_US},
+	{&congested_reps,        CONGESTED_REPS},
+	{&disabled_period,       DISABLED_PERIOD},
+	{&giveup_limit,          GIVEUP_LIMIT}
 };
 
 static struct dentry *tunables_dir;
@@ -1619,17 +1619,17 @@ static ssize_t tunables_write(struct file *file, const char __user *user,
 
 	for_each_present_cpu(cpu) {
 		bcp = &per_cpu(bau_control, cpu);
-		bcp->max_concurr =		max_concurr;
-		bcp->max_concurr_const =	max_concurr;
-		bcp->plugged_delay =		plugged_delay;
-		bcp->plugsb4reset =		plugsb4reset;
-		bcp->timeoutsb4reset =		timeoutsb4reset;
-		bcp->ipi_reset_limit =		ipi_reset_limit;
-		bcp->complete_threshold =	complete_threshold;
-		bcp->cong_response_us =		congested_respns_us;
-		bcp->cong_reps =		congested_reps;
-		bcp->disabled_period =		sec_2_cycles(disabled_period);
-		bcp->giveup_limit =		giveup_limit;
+		bcp->max_concurr         = max_concurr;
+		bcp->max_concurr_const   = max_concurr;
+		bcp->plugged_delay       = plugged_delay;
+		bcp->plugsb4reset        = plugsb4reset;
+		bcp->timeoutsb4reset     = timeoutsb4reset;
+		bcp->ipi_reset_limit     = ipi_reset_limit;
+		bcp->complete_threshold  = complete_threshold;
+		bcp->cong_response_us    = congested_respns_us;
+		bcp->cong_reps           = congested_reps;
+		bcp->disabled_period     = sec_2_cycles(disabled_period);
+		bcp->giveup_limit        = giveup_limit;
 	}
 	return count;
 }
@@ -1740,7 +1740,7 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode)
 		memset(bd2, 0, sizeof(struct bau_desc));
 		if (uv1) {
 			uv1_hdr = &bd2->header.uv1_hdr;
-			uv1_hdr->swack_flag =	1;
+			uv1_hdr->swack_flag = 1;
 			/*
 			 * The base_dest_nasid set in the message header
 			 * is the nasid of the first uvhub in the partition.
@@ -1749,10 +1749,10 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode)
 			 * if nasid striding is being used.
 			 */
 			uv1_hdr->base_dest_nasid =
-						UV_PNODE_TO_NASID(base_pnode);
-			uv1_hdr->dest_subnodeid =	UV_LB_SUBNODEID;
-			uv1_hdr->command =		UV_NET_ENDPOINT_INTD;
-			uv1_hdr->int_both =		1;
+			                          UV_PNODE_TO_NASID(base_pnode);
+			uv1_hdr->dest_subnodeid  = UV_LB_SUBNODEID;
+			uv1_hdr->command         = UV_NET_ENDPOINT_INTD;
+			uv1_hdr->int_both        = 1;
 			/*
 			 * all others need to be set to zero:
 			 *   fairness chaining multilevel count replied_to
@@ -1763,11 +1763,11 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode)
 			 * uses native mode for selective broadcasts.
 			 */
 			uv2_3_hdr = &bd2->header.uv2_3_hdr;
-			uv2_3_hdr->swack_flag =	1;
+			uv2_3_hdr->swack_flag      = 1;
 			uv2_3_hdr->base_dest_nasid =
-						UV_PNODE_TO_NASID(base_pnode);
-			uv2_3_hdr->dest_subnodeid =	UV_LB_SUBNODEID;
-			uv2_3_hdr->command =		UV_NET_ENDPOINT_INTD;
+			                          UV_PNODE_TO_NASID(base_pnode);
+			uv2_3_hdr->dest_subnodeid  = UV_LB_SUBNODEID;
+			uv2_3_hdr->command         = UV_NET_ENDPOINT_INTD;
 		}
 	}
 	for_each_present_cpu(cpu) {
@@ -1914,8 +1914,8 @@ static void __init init_per_cpu_tunables(void)
 		bcp->complete_threshold		= complete_threshold;
 		bcp->cong_response_us		= congested_respns_us;
 		bcp->cong_reps			= congested_reps;
-		bcp->disabled_period =		sec_2_cycles(disabled_period);
-		bcp->giveup_limit =		giveup_limit;
+		bcp->disabled_period		= sec_2_cycles(disabled_period);
+		bcp->giveup_limit		= giveup_limit;
 		spin_lock_init(&bcp->queue_lock);
 		spin_lock_init(&bcp->uvhub_lock);
 		spin_lock_init(&bcp->disable_lock);
#16 lore raw

PATCH v2 2/10 arch/x86/platform/uv: BAU cleanup: update printks

In reply to Andrew Banman

Replace all uses of printk with the appropriate pr_* function.

---
 arch/x86/platform/uv/tlb_uv.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)
arch/x86/platform/uv/tlb_uv.c +13 -13
index b84c2a2..8462fd1 100644
@@ -1497,16 +1497,16 @@ static ssize_t ptc_proc_write(struct file *file, const char __user *user,
 	}
 
 	if (kstrtol(optstr, 10, &input_arg) < 0) {
-		printk(KERN_DEBUG "%s is invalid\n", optstr);
+		pr_debug("%s is invalid\n", optstr);
 		return -EINVAL;
 	}
 
 	if (input_arg == 0) {
 		elements = ARRAY_SIZE(stat_description);
-		printk(KERN_DEBUG "# cpu:      cpu number\n");
-		printk(KERN_DEBUG "Sender statistics:\n");
+		pr_debug("# cpu:      cpu number\n");
+		pr_debug("Sender statistics:\n");
 		for (i = 0; i < elements; i++)
-			printk(KERN_DEBUG "%s\n", stat_description[i]);
+			pr_debug("%s\n", stat_description[i]);
 	} else if (input_arg == -1) {
 		for_each_present_cpu(cpu) {
 			stat = &per_cpu(ptcstats, cpu);
@@ -1554,7 +1554,7 @@ static int parse_tunables_write(struct bau_control *bcp, char *instr,
 			break;
 	}
 	if (cnt != e) {
-		printk(KERN_INFO "bau tunable error: should be %d values\n", e);
+		pr_info("bau tunable error: should be %d values\n", e);
 		return -EINVAL;
 	}
 
@@ -1571,7 +1571,7 @@ static int parse_tunables_write(struct bau_control *bcp, char *instr,
 				continue;
 			}
 			if (val < 1 || val > bcp->cpus_in_uvhub) {
-				printk(KERN_DEBUG
+				pr_debug(
 				"Error: BAU max concurrent %d is invalid\n",
 				val);
 				return -EINVAL;
@@ -1676,21 +1676,21 @@ static int __init uv_ptc_init(void)
 	proc_uv_ptc = proc_create(UV_PTC_BASENAME, 0444, NULL,
 				  &proc_uv_ptc_operations);
 	if (!proc_uv_ptc) {
-		printk(KERN_ERR "unable to create %s proc entry\n",
+		pr_err("unable to create %s proc entry\n",
 		       UV_PTC_BASENAME);
 		return -EINVAL;
 	}
 
 	tunables_dir = debugfs_create_dir(UV_BAU_TUNABLES_DIR, NULL);
 	if (!tunables_dir) {
-		printk(KERN_ERR "unable to create debugfs directory %s\n",
+		pr_err("unable to create debugfs directory %s\n",
 		       UV_BAU_TUNABLES_DIR);
 		return -EINVAL;
 	}
 	tunables_file = debugfs_create_file(UV_BAU_TUNABLES_FILE, 0600,
 					tunables_dir, NULL, &tunables_fops);
 	if (!tunables_file) {
-		printk(KERN_ERR "unable to create debugfs file %s\n",
+		pr_err("unable to create debugfs file %s\n",
 		       UV_BAU_TUNABLES_FILE);
 		return -EINVAL;
 	}
@@ -1944,7 +1944,7 @@ static int __init get_cpu_topology(int base_pnode,
 
 		pnode = uv_cpu_hub_info(cpu)->pnode;
 		if ((pnode - base_pnode) >= UV_DISTRIBUTION_SIZE) {
-			printk(KERN_EMERG
+			pr_emerg(
 				"cpu %d pnode %d-%d beyond %d; BAU disabled\n",
 				cpu, pnode, base_pnode, UV_DISTRIBUTION_SIZE);
 			return 1;
@@ -1969,7 +1969,7 @@ static int __init get_cpu_topology(int base_pnode,
 		sdp->cpu_number[sdp->num_cpus] = cpu;
 		sdp->num_cpus++;
 		if (sdp->num_cpus > MAX_CPUS_PER_SOCKET) {
-			printk(KERN_EMERG "%d cpus per socket invalid\n",
+			pr_emerg("%d cpus per socket invalid\n",
 				sdp->num_cpus);
 			return 1;
 		}
@@ -2036,14 +2036,14 @@ static int scan_sock(struct socket_desc *sdp, struct uvhub_desc *bdp,
 		else if (is_uv3_hub())
 			bcp->uvhub_version = 3;
 		else {
-			printk(KERN_EMERG "uvhub version not 1, 2 or 3\n");
+			pr_emerg("uvhub version not 1, 2, or 3\n");
 			return 1;
 		}
 		bcp->uvhub_master = *hmasterp;
 		bcp->uvhub_cpu = uv_cpu_blade_processor_id(cpu);
 
 		if (bcp->uvhub_cpu >= MAX_CPUS_PER_UVHUB) {
-			printk(KERN_EMERG "%d cpus per uvhub invalid\n",
+			pr_emerg("%d cpus per uvhub invalid\n",
 				bcp->uvhub_cpu);
 			return 1;
 		}
Signature · 1 line
1.8.5.6
#17 lore raw

[tip:x86/platform] x86/platform/uv/BAU: Clean up and update printks

x86/platform/uv/BAU: Clean up and update printks
Commit-ID:  efa59ab3e7526650265f0fd9696ef8be8d88ec13
Gitweb:     http://git.kernel.org/tip/efa59ab3e7526650265f0fd9696ef8be8d88ec13
Author:     Andrew Banman <[email protected]>
AuthorDate: Wed, 21 Sep 2016 11:09:13 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 22 Sep 2016 11:16:12 +0200

x86/platform/uv/BAU: Clean up and update printks

Replace all uses of printk with the appropriate pr_*() function.

---
 arch/x86/platform/uv/tlb_uv.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)
arch/x86/platform/uv/tlb_uv.c +13 -13
index b84c2a2..8462fd1 100644
@@ -1497,16 +1497,16 @@ static ssize_t ptc_proc_write(struct file *file, const char __user *user,
 	}
 
 	if (kstrtol(optstr, 10, &input_arg) < 0) {
-		printk(KERN_DEBUG "%s is invalid\n", optstr);
+		pr_debug("%s is invalid\n", optstr);
 		return -EINVAL;
 	}
 
 	if (input_arg == 0) {
 		elements = ARRAY_SIZE(stat_description);
-		printk(KERN_DEBUG "# cpu:      cpu number\n");
-		printk(KERN_DEBUG "Sender statistics:\n");
+		pr_debug("# cpu:      cpu number\n");
+		pr_debug("Sender statistics:\n");
 		for (i = 0; i < elements; i++)
-			printk(KERN_DEBUG "%s\n", stat_description[i]);
+			pr_debug("%s\n", stat_description[i]);
 	} else if (input_arg == -1) {
 		for_each_present_cpu(cpu) {
 			stat = &per_cpu(ptcstats, cpu);
@@ -1554,7 +1554,7 @@ static int parse_tunables_write(struct bau_control *bcp, char *instr,
 			break;
 	}
 	if (cnt != e) {
-		printk(KERN_INFO "bau tunable error: should be %d values\n", e);
+		pr_info("bau tunable error: should be %d values\n", e);
 		return -EINVAL;
 	}
 
@@ -1571,7 +1571,7 @@ static int parse_tunables_write(struct bau_control *bcp, char *instr,
 				continue;
 			}
 			if (val < 1 || val > bcp->cpus_in_uvhub) {
-				printk(KERN_DEBUG
+				pr_debug(
 				"Error: BAU max concurrent %d is invalid\n",
 				val);
 				return -EINVAL;
@@ -1676,21 +1676,21 @@ static int __init uv_ptc_init(void)
 	proc_uv_ptc = proc_create(UV_PTC_BASENAME, 0444, NULL,
 				  &proc_uv_ptc_operations);
 	if (!proc_uv_ptc) {
-		printk(KERN_ERR "unable to create %s proc entry\n",
+		pr_err("unable to create %s proc entry\n",
 		       UV_PTC_BASENAME);
 		return -EINVAL;
 	}
 
 	tunables_dir = debugfs_create_dir(UV_BAU_TUNABLES_DIR, NULL);
 	if (!tunables_dir) {
-		printk(KERN_ERR "unable to create debugfs directory %s\n",
+		pr_err("unable to create debugfs directory %s\n",
 		       UV_BAU_TUNABLES_DIR);
 		return -EINVAL;
 	}
 	tunables_file = debugfs_create_file(UV_BAU_TUNABLES_FILE, 0600,
 					tunables_dir, NULL, &tunables_fops);
 	if (!tunables_file) {
-		printk(KERN_ERR "unable to create debugfs file %s\n",
+		pr_err("unable to create debugfs file %s\n",
 		       UV_BAU_TUNABLES_FILE);
 		return -EINVAL;
 	}
@@ -1944,7 +1944,7 @@ static int __init get_cpu_topology(int base_pnode,
 
 		pnode = uv_cpu_hub_info(cpu)->pnode;
 		if ((pnode - base_pnode) >= UV_DISTRIBUTION_SIZE) {
-			printk(KERN_EMERG
+			pr_emerg(
 				"cpu %d pnode %d-%d beyond %d; BAU disabled\n",
 				cpu, pnode, base_pnode, UV_DISTRIBUTION_SIZE);
 			return 1;
@@ -1969,7 +1969,7 @@ static int __init get_cpu_topology(int base_pnode,
 		sdp->cpu_number[sdp->num_cpus] = cpu;
 		sdp->num_cpus++;
 		if (sdp->num_cpus > MAX_CPUS_PER_SOCKET) {
-			printk(KERN_EMERG "%d cpus per socket invalid\n",
+			pr_emerg("%d cpus per socket invalid\n",
 				sdp->num_cpus);
 			return 1;
 		}
@@ -2036,14 +2036,14 @@ static int scan_sock(struct socket_desc *sdp, struct uvhub_desc *bdp,
 		else if (is_uv3_hub())
 			bcp->uvhub_version = 3;
 		else {
-			printk(KERN_EMERG "uvhub version not 1, 2 or 3\n");
+			pr_emerg("uvhub version not 1, 2, or 3\n");
 			return 1;
 		}
 		bcp->uvhub_master = *hmasterp;
 		bcp->uvhub_cpu = uv_cpu_blade_processor_id(cpu);
 
 		if (bcp->uvhub_cpu >= MAX_CPUS_PER_UVHUB) {
-			printk(KERN_EMERG "%d cpus per uvhub invalid\n",
+			pr_emerg("%d cpus per uvhub invalid\n",
 				bcp->uvhub_cpu);
 			return 1;
 		}
#18 lore raw

PATCH v2 3/10 arch/x86/platform/uv: BAU cleanup: pq_init

In reply to Andrew Banman

The payload queue first MMR requires the physical memory address and hub GNODE of where the payload queue resides in memory, but the associated variables are named as if the PNODE were used. Rename gnode-related variables and clarify the definitions of the payload queue head, last, and tail pointers.

---
 arch/x86/include/asm/uv/uv_bau.h |  2 +-
 arch/x86/platform/uv/tlb_uv.c    | 19 ++++++++-----------
 2 files changed, 9 insertions(+), 12 deletions(-)
arch/x86/include/asm/uv/uv_bau.h +1 -1
index cc44d92..cc058c6 100644
@@ -49,7 +49,7 @@
 #define UV_NET_ENDPOINT_INTD		(is_uv1_hub() ?			\
 			UV1_NET_ENDPOINT_INTD : UV2_NET_ENDPOINT_INTD)
 #define UV_DESC_PSHIFT			49
-#define UV_PAYLOADQ_PNODE_SHIFT		49
+#define UV_PAYLOADQ_GNODE_SHIFT		49
 #define UV_PTC_BASENAME			"sgi_uv/ptc_statistics"
 #define UV_BAU_BASENAME			"sgi_uv/bau_tunables"
 #define UV_BAU_TUNABLES_DIR		"sgi_uv"
arch/x86/platform/uv/tlb_uv.c +8 -11
index 8462fd1..f6bc43b 100644
@@ -1790,10 +1790,7 @@ static void pq_init(int node, int pnode)
 	size_t plsize;
 	char *cp;
 	void *vp;
-	unsigned long pn;
-	unsigned long first;
-	unsigned long pn_first;
-	unsigned long last;
+	unsigned long gnode, first, last, tail;
 	struct bau_pq_entry *pqp;
 	struct bau_control *bcp;
 
@@ -1814,16 +1811,16 @@ static void pq_init(int node, int pnode)
 		bcp->bau_msg_head	= pqp;
 		bcp->queue_last		= pqp + (DEST_Q_SIZE - 1);
 	}
-	/*
-	 * need the gnode of where the memory was really allocated
-	 */
-	pn = uv_gpa_to_gnode(uv_gpa(pqp));
+
 	first = uv_physnodeaddr(pqp);
-	pn_first = ((unsigned long)pn << UV_PAYLOADQ_PNODE_SHIFT) | first;
 	last = uv_physnodeaddr(pqp + (DEST_Q_SIZE - 1));
-	write_mmr_payload_first(pnode, pn_first);
-	write_mmr_payload_tail(pnode, first);
+	tail = first;
+	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
+	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
+
+	write_mmr_payload_first(pnode, first);
 	write_mmr_payload_last(pnode, last);
+	write_mmr_payload_tail(pnode, tail);
 	write_gmmr_sw_ack(pnode, 0xffffUL);
 
 	/* in effect, all msg_type's are set to MSG_NOOP */
Signature · 1 line
1.8.5.6
#19 lore raw

[tip:x86/platform] x86/platform/uv/BAU: Clean up pq_init()

x86/platform/uv/BAU: Clean up pq_init()
Commit-ID:  d2a57afa53f3fdf9f68d1f4240ace85a7d20ca20
Gitweb:     http://git.kernel.org/tip/d2a57afa53f3fdf9f68d1f4240ace85a7d20ca20
Author:     Andrew Banman <[email protected]>
AuthorDate: Wed, 21 Sep 2016 11:09:14 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 22 Sep 2016 11:16:13 +0200

x86/platform/uv/BAU: Clean up pq_init()

The payload queue first MMR requires the physical memory address and hub GNODE of where the payload queue resides in memory, but the associated variables are named as if the PNODE were used. Rename gnode-related variables and clarify the definitions of the payload queue head, last, and tail pointers.

---
 arch/x86/include/asm/uv/uv_bau.h |  2 +-
 arch/x86/platform/uv/tlb_uv.c    | 19 ++++++++-----------
 2 files changed, 9 insertions(+), 12 deletions(-)
arch/x86/include/asm/uv/uv_bau.h +1 -1
index cc44d92..cc058c6 100644
@@ -49,7 +49,7 @@
 #define UV_NET_ENDPOINT_INTD		(is_uv1_hub() ?			\
 			UV1_NET_ENDPOINT_INTD : UV2_NET_ENDPOINT_INTD)
 #define UV_DESC_PSHIFT			49
-#define UV_PAYLOADQ_PNODE_SHIFT		49
+#define UV_PAYLOADQ_GNODE_SHIFT		49
 #define UV_PTC_BASENAME			"sgi_uv/ptc_statistics"
 #define UV_BAU_BASENAME			"sgi_uv/bau_tunables"
 #define UV_BAU_TUNABLES_DIR		"sgi_uv"
arch/x86/platform/uv/tlb_uv.c +8 -11
index 8462fd1..f6bc43b 100644
@@ -1790,10 +1790,7 @@ static void pq_init(int node, int pnode)
 	size_t plsize;
 	char *cp;
 	void *vp;
-	unsigned long pn;
-	unsigned long first;
-	unsigned long pn_first;
-	unsigned long last;
+	unsigned long gnode, first, last, tail;
 	struct bau_pq_entry *pqp;
 	struct bau_control *bcp;
 
@@ -1814,16 +1811,16 @@ static void pq_init(int node, int pnode)
 		bcp->bau_msg_head	= pqp;
 		bcp->queue_last		= pqp + (DEST_Q_SIZE - 1);
 	}
-	/*
-	 * need the gnode of where the memory was really allocated
-	 */
-	pn = uv_gpa_to_gnode(uv_gpa(pqp));
+
 	first = uv_physnodeaddr(pqp);
-	pn_first = ((unsigned long)pn << UV_PAYLOADQ_PNODE_SHIFT) | first;
 	last = uv_physnodeaddr(pqp + (DEST_Q_SIZE - 1));
-	write_mmr_payload_first(pnode, pn_first);
-	write_mmr_payload_tail(pnode, first);
+	tail = first;
+	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
+	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
+
+	write_mmr_payload_first(pnode, first);
 	write_mmr_payload_last(pnode, last);
+	write_mmr_payload_tail(pnode, tail);
 	write_gmmr_sw_ack(pnode, 0xffffUL);
 
 	/* in effect, all msg_type's are set to MSG_NOOP */
#20 lore raw

PATCH v2 4/10 arch/x86/platform/uv: BAU replace uv_physnodeaddr

In reply to Andrew Banman

The BAU driver should use the functions provided by uv_hub.h rather than its own implementations. uv_physnodeaddr converts vaddrs to paddrs for BAU MMR fields, but this is done better by uv_gpa_to_offset.

---
 arch/x86/include/asm/uv/uv_bau.h | 2 --
 arch/x86/platform/uv/tlb_uv.c    | 4 ++--
 2 files changed, 2 insertions(+), 4 deletions(-)
arch/x86/include/asm/uv/uv_bau.h +0 -2
index cc058c6..a46f270 100644
@@ -55,8 +55,6 @@
 #define UV_BAU_TUNABLES_DIR		"sgi_uv"
 #define UV_BAU_TUNABLES_FILE		"bau_tunables"
 #define WHITESPACE			" \t\n"
-#define uv_mmask			((1UL << uv_hub_info->m_val) - 1)
-#define uv_physnodeaddr(x)		((__pa((unsigned long)(x)) & uv_mmask))
 #define cpubit_isset(cpu, bau_local_cpumask) \
 	test_bit((cpu), (bau_local_cpumask).bits)
 
arch/x86/platform/uv/tlb_uv.c +2 -2
index f6bc43b..34b2a48 100644
@@ -1812,8 +1812,8 @@ static void pq_init(int node, int pnode)
 		bcp->queue_last		= pqp + (DEST_Q_SIZE - 1);
 	}
 
-	first = uv_physnodeaddr(pqp);
-	last = uv_physnodeaddr(pqp + (DEST_Q_SIZE - 1));
+	first = uv_gpa_to_offset(uv_gpa(pqp));
+	last = uv_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
 	tail = first;
 	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
 	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
Signature · 1 line
1.8.5.6
#21 lore raw

[tip:x86/platform] x86/platform/uv/BAU: Convert uv_physnodeaddr() use to uv_gpa_to_offset()

x86/platform/uv/BAU: Convert uv_physnodeaddr() use to uv_gpa_to_offset()
Commit-ID:  60e1c842c7ea3dd6a65660864554565cc737dd86
Gitweb:     http://git.kernel.org/tip/60e1c842c7ea3dd6a65660864554565cc737dd86
Author:     Andrew Banman <[email protected]>
AuthorDate: Wed, 21 Sep 2016 11:09:15 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 22 Sep 2016 11:16:13 +0200

x86/platform/uv/BAU: Convert uv_physnodeaddr() use to uv_gpa_to_offset()

The BAU driver should use the functions provided by uv_hub.h rather than its own implementations. uv_physnodeaddr converts vaddrs to paddrs for BAU MMR fields, but this is done better by uv_gpa_to_offset.

---
 arch/x86/include/asm/uv/uv_bau.h | 2 --
 arch/x86/platform/uv/tlb_uv.c    | 4 ++--
 2 files changed, 2 insertions(+), 4 deletions(-)
arch/x86/include/asm/uv/uv_bau.h +0 -2
index cc058c6..a46f270 100644
@@ -55,8 +55,6 @@
 #define UV_BAU_TUNABLES_DIR		"sgi_uv"
 #define UV_BAU_TUNABLES_FILE		"bau_tunables"
 #define WHITESPACE			" \t\n"
-#define uv_mmask			((1UL << uv_hub_info->m_val) - 1)
-#define uv_physnodeaddr(x)		((__pa((unsigned long)(x)) & uv_mmask))
 #define cpubit_isset(cpu, bau_local_cpumask) \
 	test_bit((cpu), (bau_local_cpumask).bits)
 
arch/x86/platform/uv/tlb_uv.c +2 -2
index f6bc43b..34b2a48 100644
@@ -1812,8 +1812,8 @@ static void pq_init(int node, int pnode)
 		bcp->queue_last		= pqp + (DEST_Q_SIZE - 1);
 	}
 
-	first = uv_physnodeaddr(pqp);
-	last = uv_physnodeaddr(pqp + (DEST_Q_SIZE - 1));
+	first = uv_gpa_to_offset(uv_gpa(pqp));
+	last = uv_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
 	tail = first;
 	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
 	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
#22 lore raw

PATCH v2 5/10 arch/x86/platform/uv: BAU add generic function pointers

In reply to Andrew Banman

Many BAU functions have different implementations depending on the UV version. Rather than switching on the uvhub_version throughout the driver, we can define a set of operations for each version. This is especially beneficial for UV4, which will require many new MMR read/write functions.

Currently, the set of abstracted functions are the same for UV1, UV2, and UV3. The functions were chosen because each one will have a different implementation for UV4. Other functions will be added as needed to handle new implementations or to cleanup the existing differences between UV1, UV2, and UV3, i.e. read_status and wait_completion.

---
 arch/x86/include/asm/uv/uv_bau.h | 11 +++++++++++
 arch/x86/platform/uv/tlb_uv.c    | 19 +++++++++++++++++++
 2 files changed, 30 insertions(+)
arch/x86/include/asm/uv/uv_bau.h +11 -0
index a46f270..a7a93a5 100644
@@ -385,6 +385,17 @@ struct uv2_3_bau_msg_header {
 	/* bits 127:120 */
 };
 
+/* Abstracted BAU functions */
+struct bau_operations {
+	unsigned long (*read_l_sw_ack)(void);
+	unsigned long (*read_g_sw_ack)(int pnode);
+	unsigned long (*bau_gpa_to_offset)(unsigned long vaddr);
+	void (*write_l_sw_ack)(unsigned long mmr);
+	void (*write_g_sw_ack)(int pnode, unsigned long mmr);
+	void (*write_payload_first)(int pnode, unsigned long mmr);
+	void (*write_payload_last)(int pnode, unsigned long mmr);
+};
+
 /*
  * The activation descriptor:
  * The format of the message to send, plus all accompanying control
arch/x86/platform/uv/tlb_uv.c +19 -0
index 34b2a48..a33a433 100644
@@ -24,6 +24,18 @@
 #include <asm/irq_vectors.h>
 #include <asm/timer.h>
 
+static struct bau_operations ops;
+
+static struct bau_operations uv123_bau_ops = {
+	.bau_gpa_to_offset       = uv_gpa_to_offset,
+	.read_l_sw_ack           = read_mmr_sw_ack,
+	.read_g_sw_ack           = read_gmmr_sw_ack,
+	.write_l_sw_ack          = write_mmr_sw_ack,
+	.write_g_sw_ack          = write_gmmr_sw_ack,
+	.write_payload_first     = write_mmr_payload_first,
+	.write_payload_last      = write_mmr_payload_last,
+};
+
 /* timeouts in nanoseconds (indexed by UVH_AGING_PRESCALE_SEL urgency7 30:28) */
 static int timeout_base_ns[] = {
 		20,
@@ -2135,6 +2147,13 @@ static int __init uv_bau_init(void)
 	if (!is_uv_system())
 		return 0;
 
+	if (is_uv3_hub())
+		ops = uv123_bau_ops;
+	else if (is_uv2_hub())
+		ops = uv123_bau_ops;
+	else if (is_uv1_hub())
+		ops = uv123_bau_ops;
+
 	for_each_possible_cpu(cur_cpu) {
 		mask = &per_cpu(uv_flush_tlb_mask, cur_cpu);
 		zalloc_cpumask_var_node(mask, GFP_KERNEL, cpu_to_node(cur_cpu));
Signature · 1 line
1.8.5.6
#23 lore raw

[tip:x86/platform] x86/platform/uv/BAU: Add generic function pointers

x86/platform/uv/BAU: Add generic function pointers
Commit-ID:  5e4f96fe2a61c759d5d47f8112813618805c85a0
Gitweb:     http://git.kernel.org/tip/5e4f96fe2a61c759d5d47f8112813618805c85a0
Author:     Andrew Banman <[email protected]>
AuthorDate: Wed, 21 Sep 2016 11:09:16 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 22 Sep 2016 11:16:13 +0200

x86/platform/uv/BAU: Add generic function pointers

Many BAU functions have different implementations depending on the UV version. Rather than switching on the uvhub_version throughout the driver, we can define a set of operations for each version. This is especially beneficial for UV4, which will require many new MMR read/write functions.

Currently, the set of abstracted functions are the same for UV1, UV2, and UV3. The functions were chosen because each one will have a different implementation for UV4. Other functions will be added as needed to handle new implementations or to cleanup the existing differences between UV1, UV2, and UV3, i.e. read_status and wait_completion.

---
 arch/x86/include/asm/uv/uv_bau.h | 11 +++++++++++
 arch/x86/platform/uv/tlb_uv.c    | 19 +++++++++++++++++++
 2 files changed, 30 insertions(+)
arch/x86/include/asm/uv/uv_bau.h +11 -0
index a46f270..a7a93a5 100644
@@ -385,6 +385,17 @@ struct uv2_3_bau_msg_header {
 	/* bits 127:120 */
 };
 
+/* Abstracted BAU functions */
+struct bau_operations {
+	unsigned long (*read_l_sw_ack)(void);
+	unsigned long (*read_g_sw_ack)(int pnode);
+	unsigned long (*bau_gpa_to_offset)(unsigned long vaddr);
+	void (*write_l_sw_ack)(unsigned long mmr);
+	void (*write_g_sw_ack)(int pnode, unsigned long mmr);
+	void (*write_payload_first)(int pnode, unsigned long mmr);
+	void (*write_payload_last)(int pnode, unsigned long mmr);
+};
+
 /*
  * The activation descriptor:
  * The format of the message to send, plus all accompanying control
arch/x86/platform/uv/tlb_uv.c +19 -0
index 34b2a48..a33a433 100644
@@ -24,6 +24,18 @@
 #include <asm/irq_vectors.h>
 #include <asm/timer.h>
 
+static struct bau_operations ops;
+
+static struct bau_operations uv123_bau_ops = {
+	.bau_gpa_to_offset       = uv_gpa_to_offset,
+	.read_l_sw_ack           = read_mmr_sw_ack,
+	.read_g_sw_ack           = read_gmmr_sw_ack,
+	.write_l_sw_ack          = write_mmr_sw_ack,
+	.write_g_sw_ack          = write_gmmr_sw_ack,
+	.write_payload_first     = write_mmr_payload_first,
+	.write_payload_last      = write_mmr_payload_last,
+};
+
 /* timeouts in nanoseconds (indexed by UVH_AGING_PRESCALE_SEL urgency7 30:28) */
 static int timeout_base_ns[] = {
 		20,
@@ -2135,6 +2147,13 @@ static int __init uv_bau_init(void)
 	if (!is_uv_system())
 		return 0;
 
+	if (is_uv3_hub())
+		ops = uv123_bau_ops;
+	else if (is_uv2_hub())
+		ops = uv123_bau_ops;
+	else if (is_uv1_hub())
+		ops = uv123_bau_ops;
+
 	for_each_possible_cpu(cur_cpu) {
 		mask = &per_cpu(uv_flush_tlb_mask, cur_cpu);
 		zalloc_cpumask_var_node(mask, GFP_KERNEL, cpu_to_node(cur_cpu));
#24 lore raw

PATCH v2 6/10 arch/x86/platform/uv: BAU use generic function pointers

In reply to Andrew Banman

Convert the use of UV version-specific functions to their abstracted counterparts.

---
 arch/x86/platform/uv/tlb_uv.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)
arch/x86/platform/uv/tlb_uv.c +13 -13
index a33a433..030d452 100644
@@ -228,7 +228,7 @@ static void reply_to_message(struct msg_desc *mdp, struct bau_control *bcp,
 	msg = mdp->msg;
 	if (!msg->canceled && do_acknowledge) {
 		dw = (msg->swack_vec << UV_SW_ACK_NPENDING) | msg->swack_vec;
-		write_mmr_sw_ack(dw);
+		ops.write_l_sw_ack(dw);
 	}
 	msg->replied_to = 1;
 	msg->swack_vec = 0;
@@ -264,7 +264,7 @@ static void bau_process_retry_msg(struct msg_desc *mdp,
 			msg->swack_vec) == 0) &&
 		    (msg2->sending_cpu == msg->sending_cpu) &&
 		    (msg2->msg_type != MSG_NOOP)) {
-			mmr = read_mmr_sw_ack();
+			mmr = ops.read_l_sw_ack();
 			msg_res = msg2->swack_vec;
 			/*
 			 * This is a message retry; clear the resources held
@@ -282,7 +282,7 @@ static void bau_process_retry_msg(struct msg_desc *mdp,
 				stat->d_canceled++;
 				cancel_count++;
 				mr = (msg_res << UV_SW_ACK_NPENDING) | msg_res;
-				write_mmr_sw_ack(mr);
+				ops.write_l_sw_ack(mr);
 			}
 		}
 	}
@@ -415,12 +415,12 @@ static void do_reset(void *ptr)
 			/*
 			 * only reset the resource if it is still pending
 			 */
-			mmr = read_mmr_sw_ack();
+			mmr = ops.read_l_sw_ack();
 			msg_res = msg->swack_vec;
 			mr = (msg_res << UV_SW_ACK_NPENDING) | msg_res;
 			if (mmr & msg_res) {
 				stat->d_rcanceled++;
-				write_mmr_sw_ack(mr);
+				ops.write_l_sw_ack(mr);
 			}
 		}
 	}
@@ -1214,7 +1214,7 @@ void process_uv2_message(struct msg_desc *mdp, struct bau_control *bcp)
 	struct bau_pq_entry *msg = mdp->msg;
 	struct bau_pq_entry *other_msg;
 
-	mmr_image = read_mmr_sw_ack();
+	mmr_image = ops.read_l_sw_ack();
 	swack_vec = msg->swack_vec;
 
 	if ((swack_vec & mmr_image) == 0) {
@@ -1443,7 +1443,7 @@ static int ptc_seq_show(struct seq_file *file, void *data)
 		/* destination side statistics */
 		seq_printf(file,
 			"%lx %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld\n",
-			   read_gmmr_sw_ack(uv_cpu_to_pnode(cpu)),
+			   ops.read_g_sw_ack(uv_cpu_to_pnode(cpu)),
 			   stat->d_requestee, cycles_2_us(stat->d_time),
 			   stat->d_alltlb, stat->d_onetlb, stat->d_multmsg,
 			   stat->d_nomsg, stat->d_retries, stat->d_canceled,
@@ -1737,7 +1737,7 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode)
 
 	gpa = uv_gpa(bau_desc);
 	n = uv_gpa_to_gnode(gpa);
-	m = uv_gpa_to_offset(gpa);
+	m = ops.bau_gpa_to_offset(gpa);
 	if (is_uv1_hub())
 		uv1 = 1;
 
@@ -1824,16 +1824,16 @@ static void pq_init(int node, int pnode)
 		bcp->queue_last		= pqp + (DEST_Q_SIZE - 1);
 	}
 
-	first = uv_gpa_to_offset(uv_gpa(pqp));
-	last = uv_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
+	first = ops.bau_gpa_to_offset(uv_gpa(pqp));
+	last = ops.bau_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
 	tail = first;
 	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
 	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
 
-	write_mmr_payload_first(pnode, first);
-	write_mmr_payload_last(pnode, last);
 	write_mmr_payload_tail(pnode, tail);
-	write_gmmr_sw_ack(pnode, 0xffffUL);
+	ops.write_payload_first(pnode, first);
+	ops.write_payload_last(pnode, last);
+	ops.write_g_sw_ack(pnode, 0xffffUL);
 
 	/* in effect, all msg_type's are set to MSG_NOOP */
 	memset(pqp, 0, sizeof(struct bau_pq_entry) * DEST_Q_SIZE);
Signature · 1 line
1.8.5.6
#25 lore raw

[tip:x86/platform] x86/platform/uv/BAU: Use generic function pointers

x86/platform/uv/BAU: Use generic function pointers
Commit-ID:  21e3f12fc0e12181102ad0400bcb50bc7a027106
Gitweb:     http://git.kernel.org/tip/21e3f12fc0e12181102ad0400bcb50bc7a027106
Author:     Andrew Banman <[email protected]>
AuthorDate: Wed, 21 Sep 2016 11:09:17 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 22 Sep 2016 11:16:14 +0200

x86/platform/uv/BAU: Use generic function pointers

Convert the use of UV version-specific functions to their abstracted counterparts.

---
 arch/x86/platform/uv/tlb_uv.c | 26 +++++++++++++-------------
 1 file changed, 13 insertions(+), 13 deletions(-)
arch/x86/platform/uv/tlb_uv.c +13 -13
index a33a433..030d452 100644
@@ -228,7 +228,7 @@ static void reply_to_message(struct msg_desc *mdp, struct bau_control *bcp,
 	msg = mdp->msg;
 	if (!msg->canceled && do_acknowledge) {
 		dw = (msg->swack_vec << UV_SW_ACK_NPENDING) | msg->swack_vec;
-		write_mmr_sw_ack(dw);
+		ops.write_l_sw_ack(dw);
 	}
 	msg->replied_to = 1;
 	msg->swack_vec = 0;
@@ -264,7 +264,7 @@ static void bau_process_retry_msg(struct msg_desc *mdp,
 			msg->swack_vec) == 0) &&
 		    (msg2->sending_cpu == msg->sending_cpu) &&
 		    (msg2->msg_type != MSG_NOOP)) {
-			mmr = read_mmr_sw_ack();
+			mmr = ops.read_l_sw_ack();
 			msg_res = msg2->swack_vec;
 			/*
 			 * This is a message retry; clear the resources held
@@ -282,7 +282,7 @@ static void bau_process_retry_msg(struct msg_desc *mdp,
 				stat->d_canceled++;
 				cancel_count++;
 				mr = (msg_res << UV_SW_ACK_NPENDING) | msg_res;
-				write_mmr_sw_ack(mr);
+				ops.write_l_sw_ack(mr);
 			}
 		}
 	}
@@ -415,12 +415,12 @@ static void do_reset(void *ptr)
 			/*
 			 * only reset the resource if it is still pending
 			 */
-			mmr = read_mmr_sw_ack();
+			mmr = ops.read_l_sw_ack();
 			msg_res = msg->swack_vec;
 			mr = (msg_res << UV_SW_ACK_NPENDING) | msg_res;
 			if (mmr & msg_res) {
 				stat->d_rcanceled++;
-				write_mmr_sw_ack(mr);
+				ops.write_l_sw_ack(mr);
 			}
 		}
 	}
@@ -1214,7 +1214,7 @@ void process_uv2_message(struct msg_desc *mdp, struct bau_control *bcp)
 	struct bau_pq_entry *msg = mdp->msg;
 	struct bau_pq_entry *other_msg;
 
-	mmr_image = read_mmr_sw_ack();
+	mmr_image = ops.read_l_sw_ack();
 	swack_vec = msg->swack_vec;
 
 	if ((swack_vec & mmr_image) == 0) {
@@ -1443,7 +1443,7 @@ static int ptc_seq_show(struct seq_file *file, void *data)
 		/* destination side statistics */
 		seq_printf(file,
 			"%lx %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld %ld\n",
-			   read_gmmr_sw_ack(uv_cpu_to_pnode(cpu)),
+			   ops.read_g_sw_ack(uv_cpu_to_pnode(cpu)),
 			   stat->d_requestee, cycles_2_us(stat->d_time),
 			   stat->d_alltlb, stat->d_onetlb, stat->d_multmsg,
 			   stat->d_nomsg, stat->d_retries, stat->d_canceled,
@@ -1737,7 +1737,7 @@ static void activation_descriptor_init(int node, int pnode, int base_pnode)
 
 	gpa = uv_gpa(bau_desc);
 	n = uv_gpa_to_gnode(gpa);
-	m = uv_gpa_to_offset(gpa);
+	m = ops.bau_gpa_to_offset(gpa);
 	if (is_uv1_hub())
 		uv1 = 1;
 
@@ -1824,16 +1824,16 @@ static void pq_init(int node, int pnode)
 		bcp->queue_last		= pqp + (DEST_Q_SIZE - 1);
 	}
 
-	first = uv_gpa_to_offset(uv_gpa(pqp));
-	last = uv_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
+	first = ops.bau_gpa_to_offset(uv_gpa(pqp));
+	last = ops.bau_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
 	tail = first;
 	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
 	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
 
-	write_mmr_payload_first(pnode, first);
-	write_mmr_payload_last(pnode, last);
 	write_mmr_payload_tail(pnode, tail);
-	write_gmmr_sw_ack(pnode, 0xffffUL);
+	ops.write_payload_first(pnode, first);
+	ops.write_payload_last(pnode, last);
+	ops.write_g_sw_ack(pnode, 0xffffUL);
 
 	/* in effect, all msg_type's are set to MSG_NOOP */
 	memset(pqp, 0, sizeof(struct bau_pq_entry) * DEST_Q_SIZE);
#26 lore raw

PATCH v2 7/10 arch/x86/platform/uv: BAU UV4 populate uvhub_version

In reply to Andrew Banman
---
 arch/x86/platform/uv/tlb_uv.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/platform/uv/tlb_uv.c +3 -1
index 030d452..ddbeb16 100644
@@ -2044,8 +2044,10 @@ static int scan_sock(struct socket_desc *sdp, struct uvhub_desc *bdp,
 			bcp->uvhub_version = 2;
 		else if (is_uv3_hub())
 			bcp->uvhub_version = 3;
+		else if (is_uv4_hub())
+			bcp->uvhub_version = 4;
 		else {
-			pr_emerg("uvhub version not 1, 2, or 3\n");
+			pr_emerg("uvhub version not 1, 2, 3, or 4\n");
 			return 1;
 		}
 		bcp->uvhub_master = *hmasterp;
Signature · 1 line
1.8.5.6
#27 lore raw

[tip:x86/platform] x86/platform/uv/BAU: Populate ->uvhub_version with UV4 version information

x86/platform/uv/BAU: Populate ->uvhub_version with UV4 version information
Commit-ID:  58d4ab46f21e7e800a7597f271a23ec602796247
Gitweb:     http://git.kernel.org/tip/58d4ab46f21e7e800a7597f271a23ec602796247
Author:     Andrew Banman <[email protected]>
AuthorDate: Wed, 21 Sep 2016 11:09:18 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 22 Sep 2016 11:16:14 +0200

x86/platform/uv/BAU: Populate ->uvhub_version with UV4 version information

---
 arch/x86/platform/uv/tlb_uv.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
arch/x86/platform/uv/tlb_uv.c +3 -1
index 030d452..ddbeb16 100644
@@ -2044,8 +2044,10 @@ static int scan_sock(struct socket_desc *sdp, struct uvhub_desc *bdp,
 			bcp->uvhub_version = 2;
 		else if (is_uv3_hub())
 			bcp->uvhub_version = 3;
+		else if (is_uv4_hub())
+			bcp->uvhub_version = 4;
 		else {
-			pr_emerg("uvhub version not 1, 2, or 3\n");
+			pr_emerg("uvhub version not 1, 2, 3, or 4\n");
 			return 1;
 		}
 		bcp->uvhub_master = *hmasterp;
#28 lore raw

PATCH v2 8/10 arch/x86/platform/uv: BAU UV4 disable software timeout

In reply to Andrew Banman

Software timeouts are not currently supported on BAU for UV4. Instead, the BAU will rely on hardware-level fairness protocols to determine broadcast timeouts.

Do not call enable_timeouts or calculate_destination_timeout on UV4. These functions write to pre-UV4 MMRs so they generate error messages on UV4.

---
 arch/x86/platform/uv/tlb_uv.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
arch/x86/platform/uv/tlb_uv.c +5 -2
index ddbeb16..72a5de7 100644
@@ -2109,7 +2109,8 @@ static int __init init_per_cpu(int nuvhubs, int base_part_pnode)
 	void *vp;
 	struct uvhub_desc *uvhub_descs;
 
-	timeout_us = calculate_destination_timeout();
+	if (is_uv3_hub() || is_uv2_hub() || is_uv1_hub())
+		timeout_us = calculate_destination_timeout();
 
 	vp = kmalloc(nuvhubs * sizeof(struct uvhub_desc), GFP_KERNEL);
 	uvhub_descs = (struct uvhub_desc *)vp;
@@ -2171,7 +2172,9 @@ static int __init uv_bau_init(void)
 			uv_base_pnode = uv_blade_to_pnode(uvhub);
 	}
 
-	enable_timeouts();
+	/* software timeouts are not supported on UV4 */
+	if (is_uv3_hub() || is_uv2_hub() || is_uv1_hub())
+		enable_timeouts();
 
 	if (init_per_cpu(nuvhubs, uv_base_pnode)) {
 		set_bau_off();
Signature · 1 line
1.8.5.6
#29 lore raw

[tip:x86/platform] x86/platform/uv/BAU: Disable software timeout on UV4 hardware

x86/platform/uv/BAU: Disable software timeout on UV4 hardware
Commit-ID:  e879c1124a6c5c3367f20a254909605e7ee938c1
Gitweb:     http://git.kernel.org/tip/e879c1124a6c5c3367f20a254909605e7ee938c1
Author:     Andrew Banman <[email protected]>
AuthorDate: Wed, 21 Sep 2016 11:09:19 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 22 Sep 2016 11:16:14 +0200

x86/platform/uv/BAU: Disable software timeout on UV4 hardware

Software timeouts are not currently supported on BAU for UV4. Instead, the BAU will rely on hardware-level fairness protocols to determine broadcast timeouts.

Do not call enable_timeouts or calculate_destination_timeout on UV4. These functions write to pre-UV4 MMRs so they generate error messages on UV4.

---
 arch/x86/platform/uv/tlb_uv.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
arch/x86/platform/uv/tlb_uv.c +5 -2
index ddbeb16..72a5de7 100644
@@ -2109,7 +2109,8 @@ static int __init init_per_cpu(int nuvhubs, int base_part_pnode)
 	void *vp;
 	struct uvhub_desc *uvhub_descs;
 
-	timeout_us = calculate_destination_timeout();
+	if (is_uv3_hub() || is_uv2_hub() || is_uv1_hub())
+		timeout_us = calculate_destination_timeout();
 
 	vp = kmalloc(nuvhubs * sizeof(struct uvhub_desc), GFP_KERNEL);
 	uvhub_descs = (struct uvhub_desc *)vp;
@@ -2171,7 +2172,9 @@ static int __init uv_bau_init(void)
 			uv_base_pnode = uv_blade_to_pnode(uvhub);
 	}
 
-	enable_timeouts();
+	/* software timeouts are not supported on UV4 */
+	if (is_uv3_hub() || is_uv2_hub() || is_uv1_hub())
+		enable_timeouts();
 
 	if (init_per_cpu(nuvhubs, uv_base_pnode)) {
 		set_bau_off();
#30 lore raw

PATCH v2 9/10 arch/x86/platform/uv: BAU UV4 fix payload queue setup

In reply to Andrew Banman

The BAU on UV4 does not need to maintain the payload queue tail pointer. Do not initialize the tail pointer MMR on UV4.

Note that write_payload_tail is not an abstracted BAU function since it is an operation specific to pre-UV4 versions. Then we must switch on the UV version to control its usage, for which we use uvhub_version rather than is_uv*_hub because it is quicker/more concise.

---
 arch/x86/platform/uv/tlb_uv.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
arch/x86/platform/uv/tlb_uv.c +12 -4
index 72a5de7..7ca0e5c 100644
@@ -1826,11 +1826,19 @@ static void pq_init(int node, int pnode)
 
 	first = ops.bau_gpa_to_offset(uv_gpa(pqp));
 	last = ops.bau_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
-	tail = first;
-	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
-	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
 
-	write_mmr_payload_tail(pnode, tail);
+	/*
+	 * Pre UV4, the gnode is required to locate the payload queue
+	 * and the payload queue tail must be maintained by the kernel.
+	 */
+	bcp = &per_cpu(bau_control, smp_processor_id());
+	if (bcp->uvhub_version <= 3) {
+		tail = first;
+		gnode = uv_gpa_to_gnode(uv_gpa(pqp));
+		first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
+		write_mmr_payload_tail(pnode, tail);
+	}
+
 	ops.write_payload_first(pnode, first);
 	ops.write_payload_last(pnode, last);
 	ops.write_g_sw_ack(pnode, 0xffffUL);
Signature · 1 line
1.8.5.6
#31 lore raw

[tip:x86/platform] x86/platform/uv/BAU: Fix payload queue setup on UV4 hardware

x86/platform/uv/BAU: Fix payload queue setup on UV4 hardware
Commit-ID:  6d78059bbc0ace5461938aaea8cda95eb6719898
Gitweb:     http://git.kernel.org/tip/6d78059bbc0ace5461938aaea8cda95eb6719898
Author:     Andrew Banman <[email protected]>
AuthorDate: Wed, 21 Sep 2016 11:09:20 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 22 Sep 2016 11:16:15 +0200

x86/platform/uv/BAU: Fix payload queue setup on UV4 hardware

The BAU on UV4 does not need to maintain the payload queue tail pointer. Do not initialize the tail pointer MMR on UV4.

Note that write_payload_tail is not an abstracted BAU function since it is an operation specific to pre-UV4 versions. Then we must switch on the UV version to control its usage, for which we use uvhub_version rather than is_uv*_hub because it is quicker/more concise.

---
 arch/x86/platform/uv/tlb_uv.c | 16 ++++++++++++----
 1 file changed, 12 insertions(+), 4 deletions(-)
arch/x86/platform/uv/tlb_uv.c +12 -4
index 72a5de7..7ca0e5c 100644
@@ -1826,11 +1826,19 @@ static void pq_init(int node, int pnode)
 
 	first = ops.bau_gpa_to_offset(uv_gpa(pqp));
 	last = ops.bau_gpa_to_offset(uv_gpa(pqp + (DEST_Q_SIZE - 1)));
-	tail = first;
-	gnode = uv_gpa_to_gnode(uv_gpa(pqp));
-	first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
 
-	write_mmr_payload_tail(pnode, tail);
+	/*
+	 * Pre UV4, the gnode is required to locate the payload queue
+	 * and the payload queue tail must be maintained by the kernel.
+	 */
+	bcp = &per_cpu(bau_control, smp_processor_id());
+	if (bcp->uvhub_version <= 3) {
+		tail = first;
+		gnode = uv_gpa_to_gnode(uv_gpa(pqp));
+		first = (gnode << UV_PAYLOADQ_GNODE_SHIFT) | tail;
+		write_mmr_payload_tail(pnode, tail);
+	}
+
 	ops.write_payload_first(pnode, first);
 	ops.write_payload_last(pnode, last);
 	ops.write_g_sw_ack(pnode, 0xffffUL);
#32 lore raw

PATCH v2 10/10 arch/x86/platform/uv: BAU UV4 add version-specific functions

In reply to Andrew Banman

Add the UV4-specific function definitions and define an operations struct to implement them in the BAU driver.

Many BAU MMRs, although functionally the same, have new addresses on UV4 due to hardware changes. Each MMR requires new read/write functions, but their implementation in the driver does not change. Thus, it is enough to enumerate them in the operations struct for the changes to take effect.

---
 arch/x86/include/asm/uv/uv_bau.h | 30 ++++++++++++++++++++++++++++++
 arch/x86/platform/uv/tlb_uv.c    | 15 ++++++++++++++-
 2 files changed, 44 insertions(+), 1 deletion(-)
arch/x86/include/asm/uv/uv_bau.h +30 -0
index a7a93a5..57ab86d 100644
@@ -664,6 +664,16 @@ static inline void write_gmmr_activation(int pnode, unsigned long mmr_image)
 	write_gmmr(pnode, UVH_LB_BAU_SB_ACTIVATION_CONTROL, mmr_image);
 }
 
+static inline void write_mmr_proc_payload_first(int pnode, unsigned long mmr_image)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_FIRST, mmr_image);
+}
+
+static inline void write_mmr_proc_payload_last(int pnode, unsigned long mmr_image)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_LAST, mmr_image);
+}
+
 static inline void write_mmr_payload_first(int pnode, unsigned long mmr_image)
 {
 	write_gmmr(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST, mmr_image);
@@ -709,6 +719,26 @@ static inline unsigned long read_gmmr_sw_ack(int pnode)
 	return read_gmmr(pnode, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
 }
 
+static inline void write_mmr_proc_sw_ack(unsigned long mr)
+{
+	uv_write_local_mmr(UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
+}
+
+static inline void write_gmmr_proc_sw_ack(int pnode, unsigned long mr)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
+}
+
+static inline unsigned long read_mmr_proc_sw_ack(void)
+{
+	return read_lmmr(UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
+}
+
+static inline unsigned long read_gmmr_proc_sw_ack(int pnode)
+{
+	return read_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
+}
+
 static inline void write_mmr_data_config(int pnode, unsigned long mr)
 {
 	uv_write_global_mmr64(pnode, UVH_BAU_DATA_CONFIG, mr);
arch/x86/platform/uv/tlb_uv.c +14 -1
index 7ca0e5c..56c5a3a 100644
@@ -36,6 +36,17 @@ static struct bau_operations uv123_bau_ops = {
 	.write_payload_last      = write_mmr_payload_last,
 };
 
+static struct bau_operations uv4_bau_ops = {
+	.bau_gpa_to_offset       = uv_gpa_to_soc_phys_ram,
+	.read_l_sw_ack           = read_mmr_proc_sw_ack,
+	.read_g_sw_ack           = read_gmmr_proc_sw_ack,
+	.write_l_sw_ack          = write_mmr_proc_sw_ack,
+	.write_g_sw_ack          = write_gmmr_proc_sw_ack,
+	.write_payload_first     = write_mmr_proc_payload_first,
+	.write_payload_last      = write_mmr_proc_payload_last,
+};
+
+
 /* timeouts in nanoseconds (indexed by UVH_AGING_PRESCALE_SEL urgency7 30:28) */
 static int timeout_base_ns[] = {
 		20,
@@ -2158,7 +2169,9 @@ static int __init uv_bau_init(void)
 	if (!is_uv_system())
 		return 0;
 
-	if (is_uv3_hub())
+	if (is_uv4_hub())
+		ops = uv4_bau_ops;
+	else if (is_uv3_hub())
 		ops = uv123_bau_ops;
 	else if (is_uv2_hub())
 		ops = uv123_bau_ops;
Signature · 1 line
1.8.5.6
#33 lore raw

[tip:x86/platform] x86/platform/uv/BAU: Add UV4-specific functions

x86/platform/uv/BAU: Add UV4-specific functions
Commit-ID:  4f059d514f7119a4fdd9934189ff31f2c26b2647
Gitweb:     http://git.kernel.org/tip/4f059d514f7119a4fdd9934189ff31f2c26b2647
Author:     Andrew Banman <[email protected]>
AuthorDate: Wed, 21 Sep 2016 11:09:21 -0500
Committer:  Ingo Molnar <[email protected]>
CommitDate: Thu, 22 Sep 2016 11:16:15 +0200

x86/platform/uv/BAU: Add UV4-specific functions

Add the UV4-specific function definitions and define an operations struct to implement them in the BAU driver.

Many BAU MMRs, although functionally the same, have new addresses on UV4 due to hardware changes. Each MMR requires new read/write functions, but their implementation in the driver does not change. Thus, it is enough to enumerate them in the operations struct for the changes to take effect.

---
 arch/x86/include/asm/uv/uv_bau.h | 30 ++++++++++++++++++++++++++++++
 arch/x86/platform/uv/tlb_uv.c    | 15 ++++++++++++++-
 2 files changed, 44 insertions(+), 1 deletion(-)
arch/x86/include/asm/uv/uv_bau.h +30 -0
index a7a93a5..57ab86d 100644
@@ -664,6 +664,16 @@ static inline void write_gmmr_activation(int pnode, unsigned long mmr_image)
 	write_gmmr(pnode, UVH_LB_BAU_SB_ACTIVATION_CONTROL, mmr_image);
 }
 
+static inline void write_mmr_proc_payload_first(int pnode, unsigned long mmr_image)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_FIRST, mmr_image);
+}
+
+static inline void write_mmr_proc_payload_last(int pnode, unsigned long mmr_image)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_QUEUE_LAST, mmr_image);
+}
+
 static inline void write_mmr_payload_first(int pnode, unsigned long mmr_image)
 {
 	write_gmmr(pnode, UVH_LB_BAU_INTD_PAYLOAD_QUEUE_FIRST, mmr_image);
@@ -709,6 +719,26 @@ static inline unsigned long read_gmmr_sw_ack(int pnode)
 	return read_gmmr(pnode, UVH_LB_BAU_INTD_SOFTWARE_ACKNOWLEDGE);
 }
 
+static inline void write_mmr_proc_sw_ack(unsigned long mr)
+{
+	uv_write_local_mmr(UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
+}
+
+static inline void write_gmmr_proc_sw_ack(int pnode, unsigned long mr)
+{
+	write_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_CLEAR, mr);
+}
+
+static inline unsigned long read_mmr_proc_sw_ack(void)
+{
+	return read_lmmr(UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
+}
+
+static inline unsigned long read_gmmr_proc_sw_ack(int pnode)
+{
+	return read_gmmr(pnode, UV4H_LB_PROC_INTD_SOFT_ACK_PENDING);
+}
+
 static inline void write_mmr_data_config(int pnode, unsigned long mr)
 {
 	uv_write_global_mmr64(pnode, UVH_BAU_DATA_CONFIG, mr);
arch/x86/platform/uv/tlb_uv.c +14 -1
index 7ca0e5c..56c5a3a 100644
@@ -36,6 +36,17 @@ static struct bau_operations uv123_bau_ops = {
 	.write_payload_last      = write_mmr_payload_last,
 };
 
+static struct bau_operations uv4_bau_ops = {
+	.bau_gpa_to_offset       = uv_gpa_to_soc_phys_ram,
+	.read_l_sw_ack           = read_mmr_proc_sw_ack,
+	.read_g_sw_ack           = read_gmmr_proc_sw_ack,
+	.write_l_sw_ack          = write_mmr_proc_sw_ack,
+	.write_g_sw_ack          = write_gmmr_proc_sw_ack,
+	.write_payload_first     = write_mmr_proc_payload_first,
+	.write_payload_last      = write_mmr_proc_payload_last,
+};
+
+
 /* timeouts in nanoseconds (indexed by UVH_AGING_PRESCALE_SEL urgency7 30:28) */
 static int timeout_base_ns[] = {
 		20,
@@ -2158,7 +2169,9 @@ static int __init uv_bau_init(void)
 	if (!is_uv_system())
 		return 0;
 
-	if (is_uv3_hub())
+	if (is_uv4_hub())
+		ops = uv4_bau_ops;
+	else if (is_uv3_hub())
 		ops = uv123_bau_ops;
 	else if (is_uv2_hub())
 		ops = uv123_bau_ops;