[PATCH 1/6] x86, UV2: use native UV2 broadcast mode
From: Cliff Wickman
Date: Mon Jan 16 2012 - 16:17:28 EST
Update the use of the Broadcast Assist Unit on SGI Altix UV2 to the
use of native UV2 mode (not the legacy mode).
UV2 native mode has a different format for a broadcast message. We also
need quick differentiaton between UV1 and UV2.
Signed-off-by: Cliff Wickman <cpw@xxxxxxx>
---
arch/x86/include/asm/uv/uv_bau.h | 93 ++++++++++++++++++++++++++++++++++++---
arch/x86/platform/uv/tlb_uv.c | 88 +++++++++++++++++++++++++++---------
2 files changed, 151 insertions(+), 30 deletions(-)
Index: 120113.linux-3.2.1/arch/x86/include/asm/uv/uv_bau.h
===================================================================
--- 120113.linux-3.2.1.orig/arch/x86/include/asm/uv/uv_bau.h
+++ 120113.linux-3.2.1/arch/x86/include/asm/uv/uv_bau.h
@@ -65,7 +65,7 @@
* UV2: Bit 19 selects between
* (0): 10 microsecond timebase and
* (1): 80 microseconds
- * we're using 655us, similar to UV1: 65 units of 10us
+ * we're using 560us, similar to UV1: 65 units of 10us
*/
#define UV1_INTD_SOFT_ACK_TIMEOUT_PERIOD (9UL)
#define UV2_INTD_SOFT_ACK_TIMEOUT_PERIOD (15UL)
@@ -235,10 +235,10 @@ struct bau_msg_payload {
/*
- * Message header: 16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
+ * UV1 Message header: 16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
* see table 4.2.3.0.1 in broacast_assist spec.
*/
-struct bau_msg_header {
+struct uv1_bau_msg_header {
unsigned int dest_subnodeid:6; /* must be 0x10, for the LB */
/* bits 5:0 */
unsigned int base_dest_nasid:15; /* nasid of the first bit */
@@ -318,19 +318,87 @@ struct bau_msg_header {
};
/*
+ * UV2 Message header: 16 bytes (128 bits) (bytes 0x30-0x3f of descriptor)
+ * see figure 9-2 of harp_sys.pdf
+ */
+struct uv2_bau_msg_header {
+ unsigned int base_dest_nasid:15; /* nasid of the first bit */
+ /* bits 14:0 */ /* in uvhub map */
+ unsigned int dest_subnodeid:5; /* must be 0x10, for the LB */
+ /* bits 19:15 */
+ unsigned int rsvd_1:1; /* must be zero */
+ /* bit 20 */
+ /* Address bits 59:21 */
+ /* bits 25:2 of address (44:21) are payload */
+ /* these next 24 bits become bytes 12-14 of msg */
+ /* bits 28:21 land in byte 12 */
+ unsigned int replied_to:1; /* sent as 0 by the source to
+ byte 12 */
+ /* bit 21 */
+ unsigned int msg_type:3; /* software type of the
+ message */
+ /* bits 24:22 */
+ unsigned int canceled:1; /* message canceled, resource
+ is to be freed*/
+ /* bit 25 */
+ unsigned int payload_1:3; /* not currently used */
+ /* bits 28:26 */
+
+ /* bits 36:29 land in byte 13 */
+ unsigned int payload_2a:3; /* not currently used */
+ unsigned int payload_2b:5; /* not currently used */
+ /* bits 36:29 */
+
+ /* bits 44:37 land in byte 14 */
+ unsigned int payload_3:8; /* not currently used */
+ /* bits 44:37 */
+
+ unsigned int rsvd_2:7; /* reserved */
+ /* bits 51:45 */
+ unsigned int swack_flag:1; /* software acknowledge flag */
+ /* bit 52 */
+ unsigned int rsvd_3a:3; /* must be zero */
+ unsigned int rsvd_3b:8; /* must be zero */
+ unsigned int rsvd_3c:8; /* must be zero */
+ unsigned int rsvd_3d:3; /* must be zero */
+ /* bits 74:53 */
+ unsigned int fairness:3; /* usually zero */
+ /* bits 77:75 */
+
+ unsigned int sequence:16; /* message sequence number */
+ /* bits 93:78 Suppl_A */
+ unsigned int chaining:1; /* next descriptor is part of
+ this activation*/
+ /* bit 94 */
+ unsigned int multilevel:1; /* multi-level multicast
+ format */
+ /* bit 95 */
+ unsigned int rsvd_4:24; /* ordered / source node /
+ source subnode / aging
+ must be zero */
+ /* bits 119:96 */
+ unsigned int command:8; /* message type */
+ /* bits 127:120 */
+};
+
+/*
* The activation descriptor:
* The format of the message to send, plus all accompanying control
* Should be 64 bytes
*/
struct bau_desc {
- struct pnmask distribution;
+ struct pnmask distribution;
/*
* message template, consisting of header and payload:
*/
- struct bau_msg_header header;
- struct bau_msg_payload payload;
+ union bau_msg_header {
+ struct uv1_bau_msg_header uv1_hdr;
+ struct uv2_bau_msg_header uv2_hdr;
+ } header;
+
+ struct bau_msg_payload payload;
};
-/*
+/* UV1:
* -payload-- ---------header------
* bytes 0-11 bits 41-56 bits 58-81
* A B (2) C (3)
@@ -340,6 +408,16 @@ struct bau_desc {
* bytes 0-11 bytes 12-14 bytes 16-17 (byte 15 filled in by hw as vector)
* ------------payload queue-----------
*/
+/* UV2:
+ * -payload-- ---------header------
+ * bytes 0-11 bits 70-78 bits 21-44
+ * A B (2) C (3)
+ *
+ * A/B/C are moved to:
+ * A C B
+ * bytes 0-11 bytes 12-14 bytes 16-17 (byte 15 filled in by hw as vector)
+ * ------------payload queue-----------
+ */
/*
* The payload queue on the destination side is an array of these.
@@ -511,6 +589,7 @@ struct bau_control {
short osnode;
short uvhub_cpu;
short uvhub;
+ short uvhub_version;
short cpus_in_socket;
short cpus_in_uvhub;
short partition_base_pnode;
Index: 120113.linux-3.2.1/arch/x86/platform/uv/tlb_uv.c
===================================================================
--- 120113.linux-3.2.1.orig/arch/x86/platform/uv/tlb_uv.c
+++ 120113.linux-3.2.1/arch/x86/platform/uv/tlb_uv.c
@@ -573,7 +573,7 @@ static int wait_completion(struct bau_de
right_shift = ((cpu - UV_CPUS_PER_AS) * UV_ACT_STATUS_SIZE);
}
- if (is_uv1_hub())
+ if (bcp->uvhub_version == 1)
return uv1_wait_completion(bau_desc, mmr_offset, right_shift,
bcp, try);
else
@@ -757,15 +757,22 @@ int uv_flush_send_and_wait(struct bau_de
{
int seq_number = 0;
int completion_stat = 0;
+ int uv1 = 0;
long try = 0;
unsigned long index;
cycles_t time1;
cycles_t time2;
struct ptc_stats *stat = bcp->statp;
struct bau_control *hmaster = bcp->uvhub_master;
+ struct uv1_bau_msg_header *uv1_hdr = NULL;
+ struct uv2_bau_msg_header *uv2_hdr = NULL;
- if (is_uv1_hub())
+ if (bcp->uvhub_version == 1) {
+ uv1 = 1;
uv1_throttle(hmaster, stat);
+ uv1_hdr = &bau_desc->header.uv1_hdr;
+ } else
+ uv2_hdr = &bau_desc->header.uv2_hdr;
while (hmaster->uvhub_quiesce)
cpu_relax();
@@ -773,14 +780,23 @@ int uv_flush_send_and_wait(struct bau_de
time1 = get_cycles();
do {
if (try == 0) {
- bau_desc->header.msg_type = MSG_REGULAR;
+ if (uv1)
+ uv1_hdr->msg_type = MSG_REGULAR;
+ else
+ uv2_hdr->msg_type = MSG_REGULAR;
seq_number = bcp->message_number++;
} else {
- bau_desc->header.msg_type = MSG_RETRY;
+ if (uv1)
+ uv1_hdr->msg_type = MSG_RETRY;
+ else
+ uv2_hdr->msg_type = MSG_RETRY;
stat->s_retry_messages++;
}
- bau_desc->header.sequence = seq_number;
+ if (uv1)
+ uv1_hdr->sequence = seq_number;
+ else
+ uv2_hdr->sequence = seq_number;
index = (1UL << AS_PUSH_SHIFT) | bcp->uvhub_cpu;
bcp->send_message = get_cycles();
@@ -967,7 +983,7 @@ const struct cpumask *uv_flush_tlb_other
stat->s_ntargself++;
bau_desc = bcp->descriptor_base;
- bau_desc += ITEMS_PER_DESC * bcp->uvhub_cpu;
+ bau_desc += (ITEMS_PER_DESC * bcp->uvhub_cpu);
bau_uvhubs_clear(&bau_desc->distribution, UV_DISTRIBUTION_SIZE);
if (set_distrib_bits(flush_mask, bcp, bau_desc, &locals, &remotes))
return NULL;
@@ -1083,7 +1099,7 @@ static void __init enable_timeouts(void)
*/
mmr_image |= (1L << SOFTACK_MSHIFT);
if (is_uv2_hub()) {
- mmr_image |= (1L << UV2_LEG_SHFT);
+ mmr_image &= ~(1L << UV2_LEG_SHFT);
mmr_image |= (1L << UV2_EXT_SHFT);
}
write_mmr_misc_control(pnode, mmr_image);
@@ -1432,12 +1448,15 @@ static void activation_descriptor_init(i
{
int i;
int cpu;
+ int uv1 = 0;
unsigned long gpa;
unsigned long m;
unsigned long n;
size_t dsize;
struct bau_desc *bau_desc;
struct bau_desc *bd2;
+ struct uv1_bau_msg_header *uv1_hdr;
+ struct uv2_bau_msg_header *uv2_hdr;
struct bau_control *bcp;
/*
@@ -1451,6 +1470,8 @@ static void activation_descriptor_init(i
gpa = uv_gpa(bau_desc);
n = uv_gpa_to_gnode(gpa);
m = uv_gpa_to_offset(gpa);
+ if (is_uv1_hub())
+ uv1 = 1;
/* the 14-bit pnode */
write_mmr_descriptor_base(pnode, (n << UV_DESC_PSHIFT | m));
@@ -1461,21 +1482,33 @@ static void activation_descriptor_init(i
*/
for (i = 0, bd2 = bau_desc; i < (ADP_SZ * ITEMS_PER_DESC); i++, bd2++) {
memset(bd2, 0, sizeof(struct bau_desc));
- bd2->header.swack_flag = 1;
- /*
- * The base_dest_nasid set in the message header is the nasid
- * of the first uvhub in the partition. The bit map will
- * indicate destination pnode numbers relative to that base.
- * They may not be consecutive if nasid striding is being used.
- */
- bd2->header.base_dest_nasid = UV_PNODE_TO_NASID(base_pnode);
- bd2->header.dest_subnodeid = UV_LB_SUBNODEID;
- bd2->header.command = UV_NET_ENDPOINT_INTD;
- bd2->header.int_both = 1;
- /*
- * all others need to be set to zero:
- * fairness chaining multilevel count replied_to
- */
+ if (uv1) {
+ uv1_hdr = &bd2->header.uv1_hdr;
+ uv1_hdr->swack_flag = 1;
+ /*
+ * The base_dest_nasid set in the message header
+ * is the nasid of the first uvhub in the partition.
+ * The bit map will indicate destination pnode numbers
+ * relative to that base. They may not be consecutive
+ * 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;
+ /*
+ * all others need to be set to zero:
+ * fairness chaining multilevel count replied_to
+ */
+ } else {
+ uv2_hdr = &bd2->header.uv2_hdr;
+ uv2_hdr->swack_flag = 1;
+ uv2_hdr->base_dest_nasid =
+ UV_PNODE_TO_NASID(base_pnode);
+ uv2_hdr->dest_subnodeid = UV_LB_SUBNODEID;
+ uv2_hdr->command = UV_NET_ENDPOINT_INTD;
+ }
}
for_each_present_cpu(cpu) {
if (pnode != uv_blade_to_pnode(uv_cpu_to_blade_id(cpu)))
@@ -1728,6 +1761,14 @@ static int scan_sock(struct socket_desc
bcp->cpus_in_socket = sdp->num_cpus;
bcp->socket_master = *smasterp;
bcp->uvhub = bdp->uvhub;
+ if (is_uv1_hub())
+ bcp->uvhub_version = 1;
+ else if (is_uv2_hub())
+ bcp->uvhub_version = 2;
+ else {
+ printk(KERN_EMERG "uvhub version not 1 or 2\n");
+ return 1;
+ }
bcp->uvhub_master = *hmasterp;
bcp->uvhub_cpu = uv_cpu_hub_info(cpu)->blade_processor_id;
if (bcp->uvhub_cpu >= MAX_CPUS_PER_UVHUB) {
@@ -1867,7 +1908,8 @@ static int __init uv_bau_init(void)
val = 1L << 63;
write_gmmr_activation(pnode, val);
mmr = 1; /* should be 1 to broadcast to both sockets */
- write_mmr_data_broadcast(pnode, mmr);
+ if (!is_uv1_hub())
+ write_mmr_data_broadcast(pnode, mmr);
}
}
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/