[Skiboot] [PATCH 10/13] hw/prd: Add firmware PRD handling subsystem

Jeremy Kerr jk at ozlabs.org
Fri Feb 27 20:11:06 AEDT 2015


This change adds Processor Recovery Diagnostics (PRD) code to skiboot
firmware. This allows certain hardware RAS events to be handled by a
userspace application.

The core of the PRD code is a messaging interface to the kernel (and
onwards to userspace). PRD events are logged with the prd.c code, and
sent to the kernel as opal_msg messages. For responses to these
messages, the kernel will reply using a new OPAL call, opal_prd_msg.

Only one message is outstanding at a time; we collect events from
hardware interrupts (hooked up by subsequent patches), and set
per-processor event bits. Once an event has been consumed by the
kernel, we clear that event from out pending set, and send any further
pending events.

Certain events (hardware attentions from the psi layer) need to be
masked at interrupt time. For these, we have an acknowledgement facility
to clear the mask once the userspace PRD application has cleared the
source of the error.

Includes multiple contributions from:

  Neelesh Gupta <neelegup at linux.vnet.ibm.com>
  Vaidyanathan Srinivasan <svaidy at linux.vnet.ibm.com>

Signed-off-by: Jeremy Kerr <jk at ozlabs.org>
Signed-off-by: Neelesh Gupta <neelegup at linux.vnet.ibm.com>
Signed-off-by: Vaidyanathan Srinivasan <svaidy at linux.vnet.ibm.com>

---
 doc/opal-api/opal-messages.txt    |   45 +++
 doc/opal-api/opal-prd-msg-113.txt |   11 
 hw/Makefile.inc                   |    2 
 hw/prd.c                          |  360 ++++++++++++++++++++++++++++++
 hw/psi.c                          |    2 
 include/skiboot.h                 |    6 
 platforms/astbmc/habanero.c       |    2 
 platforms/astbmc/palmetto.c       |    2 
 8 files changed, 428 insertions(+), 2 deletions(-)

diff --git a/doc/opal-api/opal-messages.txt b/doc/opal-api/opal-messages.txt
index 448e9c7..1091d8b 100644
--- a/doc/opal-api/opal-messages.txt
+++ b/doc/opal-api/opal-messages.txt
@@ -77,3 +77,48 @@ perform a shutdown in the future.
 
 The host OS can use the separate API OPAL_GET_DPO_STATUS to query OPAL for the
 number of seconds before a forced shutdown will occur.
+
+OPAL_MSG_PRD
+------------
+
+This message is a OPAL-to-HBRT notification, and contains a
+struct opal_prd_msg:
+
+	enum opal_prd_msg_type {
+		OPAL_PRD_MSG_TYPE_INIT = 0,	/* HBRT --> OPAL */
+		OPAL_PRD_MSG_TYPE_FINI,		/* HBRT --> OPAL */
+		OPAL_PRD_MSG_TYPE_ATTN,		/* HBRT <-- OPAL */
+		OPAL_PRD_MSG_TYPE_ATTN_ACK,	/* HBRT --> OPAL */
+		OPAL_PRD_MSG_TYPE_OCC_ERROR,	/* HBRT <-- OPAL */
+		OPAL_PRD_MSG_TYPE_OCC_RESET,	/* HBRT <-- OPAL */
+	};
+
+	struct opal_prd_msg {
+		uint8_t		type;
+		uint8_t		pad[3];
+		__be32		token;
+		union {
+			struct {
+				__be64	version;
+				__be64	ipoll;
+			} init;
+			struct {
+				__be64	proc;
+				__be64	ipoll_status;
+				__be64	ipoll_mask;
+			} attn;
+			struct {
+				__be64	proc;
+				__be64	ipoll_ack;
+			} attn_ack;
+			struct {
+				__be64	chip;
+			} occ_error;
+			struct {
+				__be64	chip;
+			} occ_reset;
+		};
+	};
+
+Responses from the kernel use the same message format, but are passed
+through the opal_prd_msg call.
diff --git a/doc/opal-api/opal-prd-msg-113.txt b/doc/opal-api/opal-prd-msg-113.txt
new file mode 100644
index 0000000..b41f17e
--- /dev/null
+++ b/doc/opal-api/opal-prd-msg-113.txt
@@ -0,0 +1,11 @@
+
+OPAL_PRD_MSG call
+-------------
+
+The OPAL_PRD_MSG call is used to pass a struct opal_prd_msg from the HBRT
+code into opal, and is paired with the OPAL_PRD_MSG message type.
+
+Parameters:
+	struct opal_msg *msg
+
+Passes an opal_msg, of type OPAL_PRD_MSG, from the OS to OPAL.
diff --git a/hw/Makefile.inc b/hw/Makefile.inc
index 416d23c..395a65f 100644
--- a/hw/Makefile.inc
+++ b/hw/Makefile.inc
@@ -5,7 +5,7 @@ HW_OBJS  = xscom.o chiptod.o gx.o cec.o lpc.o lpc-uart.o psi.o
 HW_OBJS += homer.o slw.o occ.o fsi-master.o centaur.o
 HW_OBJS += nx.o nx-rng.o nx-crypto.o nx-842.o
 HW_OBJS += p7ioc.o p7ioc-inits.o p7ioc-phb.o p5ioc2.o p5ioc2-phb.o
-HW_OBJS += phb3.o sfc-ctrl.o fake-rtc.o bt.o p8-i2c.o
+HW_OBJS += phb3.o sfc-ctrl.o fake-rtc.o bt.o p8-i2c.o prd.o
 HW=hw/built-in.o
 
 include $(SRC)/hw/fsp/Makefile.inc
diff --git a/hw/prd.c b/hw/prd.c
new file mode 100644
index 0000000..1d9c617
--- /dev/null
+++ b/hw/prd.c
@@ -0,0 +1,360 @@
+/* Copyright 2014-2015 IBM Corp.
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *	http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+ * implied.
+ * See the License for the specific language governing permissions and
+ * imitations under the License.
+ */
+
+#include <skiboot.h>
+#include <opal.h>
+#include <lock.h>
+#include <xscom.h>
+#include <chip.h>
+#include <opal-api.h>
+#include <opal-msg.h>
+#include <fsp.h>
+
+enum events {
+	EVENT_ATTN	= 1 << 0,
+	EVENT_OCC_ERROR	= 1 << 1,
+	EVENT_OCC_RESET	= 1 << 2,
+};
+
+static uint8_t events[MAX_CHIPS];
+static uint64_t ipoll_status[MAX_CHIPS];
+static struct opal_prd_msg prd_msg;
+static bool prd_msg_inuse, prd_active;
+
+/* Locking:
+ *
+ * The events lock serialises access to the events, ipoll_status,
+ * prd_msg_inuse, and prd_active variables.
+ *
+ * The ipoll_lock protects against concurrent updates to the ipoll registers.
+ *
+ * The ipoll_lock may be acquired with events_lock held. This order must
+ * be preserved.
+ */
+static struct lock events_lock = LOCK_UNLOCKED;
+static struct lock ipoll_lock = LOCK_UNLOCKED;
+
+/* PRD registers */
+#define PRD_IPOLL_REG_MASK	0x01020013
+#define PRD_IPOLL_REG_STATUS	0x01020014
+#define PRD_IPOLL_XSTOP		PPC_BIT(0) /* Xstop for host/core/millicode */
+#define PRD_IPOLL_RECOV		PPC_BIT(1) /* Recoverable */
+#define PRD_IPOLL_SPEC_ATTN	PPC_BIT(2) /* Special attention */
+#define PRD_IPOLL_HOST_ATTN	PPC_BIT(3) /* Host attention */
+#define PRD_IPOLL_MASK		PPC_BITMASK(0, 3)
+
+static int queue_prd_msg_hbrt(struct opal_prd_msg *msg,
+		void (*consumed)(void *data))
+{
+	uint64_t *buf;
+
+	BUILD_ASSERT(sizeof(*msg) / sizeof(uint64_t) == 4);
+
+	buf = (uint64_t *)msg;
+
+	return _opal_queue_msg(OPAL_MSG_PRD, msg, consumed, 4, buf);
+}
+
+static int queue_prd_msg_nop(struct opal_prd_msg *msg,
+		void (*consumed)(void *data))
+{
+	(void)msg;
+	(void)consumed;
+	return OPAL_UNSUPPORTED;
+}
+
+static int (*queue_prd_msg)(struct opal_prd_msg *msg,
+		void (*consumed)(void *data)) = queue_prd_msg_nop;
+
+static void send_pending_events(void);
+
+static void prd_msg_consumed(void *data)
+{
+	struct opal_prd_msg *msg = data;
+	uint32_t proc;
+	uint8_t event = 0;
+
+	lock(&events_lock);
+	switch (msg->type) {
+	case OPAL_PRD_MSG_TYPE_ATTN:
+		proc = msg->attn.proc;
+
+		/* If other ipoll events have been received in the time
+		 * between prd_msg creation and consumption, we'll need to
+		 * raise a separate ATTN message for those. So, we only
+		 * clear the event if we don't have any further ipoll_status
+		 * bits.
+		 */
+		ipoll_status[proc] &= ~msg->attn.ipoll_status;
+		if (!ipoll_status[proc])
+			event = EVENT_ATTN;
+
+		break;
+	case OPAL_PRD_MSG_TYPE_OCC_ERROR:
+		proc = msg->occ_error.chip;
+		event = EVENT_OCC_ERROR;
+		break;
+	case OPAL_PRD_MSG_TYPE_OCC_RESET:
+		proc = msg->occ_reset.chip;
+		event = EVENT_OCC_RESET;
+		break;
+	default:
+		return;
+	}
+
+	events[proc] &= ~event;
+	prd_msg_inuse = false;
+	send_pending_events();
+	unlock(&events_lock);
+}
+
+static int populate_ipoll_msg(struct opal_prd_msg *msg, uint32_t proc)
+{
+	uint64_t ipoll_mask;
+	int rc;
+
+	lock(&ipoll_lock);
+	rc = xscom_read(proc, PRD_IPOLL_REG_MASK, &ipoll_mask);
+	unlock(&ipoll_lock);
+
+	if (rc) {
+		prlog(PR_ERR, "PRD: Unable to read ipoll status (chip %d)!\n",
+				proc);
+		return -1;
+	}
+
+	msg->attn.proc = proc;
+	msg->attn.ipoll_status = ipoll_status[proc];
+	msg->attn.ipoll_mask = ipoll_mask;
+	return 0;
+}
+
+static void send_pending_events(void)
+{
+	struct proc_chip *chip;
+	uint32_t proc;
+	uint8_t event;
+
+	assert(!prd_msg_inuse);
+
+	if (!prd_active)
+		return;
+
+	event = 0;
+
+	for_each_chip(chip) {
+		proc = chip->id;
+		if (events[proc]) {
+			event = events[proc];
+			break;
+		}
+	}
+
+	if (!event)
+		return;
+
+	prd_msg_inuse = true;
+	prd_msg.token = 0;
+
+	if (event & EVENT_ATTN) {
+		prd_msg.type = OPAL_PRD_MSG_TYPE_ATTN;
+		populate_ipoll_msg(&prd_msg, proc);
+		event = EVENT_ATTN;
+
+	} else if (event & EVENT_OCC_ERROR) {
+		prd_msg.type = OPAL_PRD_MSG_TYPE_OCC_ERROR;
+		prd_msg.occ_error.chip = proc;
+		event = EVENT_OCC_ERROR;
+
+	} else if (event & EVENT_OCC_RESET) {
+		prd_msg.type = OPAL_PRD_MSG_TYPE_OCC_RESET;
+		prd_msg.occ_reset.chip = proc;
+		event = EVENT_OCC_RESET;
+
+	}
+
+	queue_prd_msg(&prd_msg, prd_msg_consumed);
+}
+
+static void __prd_event(uint32_t proc, uint8_t event)
+{
+	events[proc] |= event;
+	if (!prd_msg_inuse)
+		send_pending_events();
+}
+
+static void prd_event(uint32_t proc, uint8_t event)
+{
+	lock(&events_lock);
+	__prd_event(proc, event);
+	unlock(&events_lock);
+}
+
+static int __ipoll_update_mask(uint32_t proc, bool set, uint64_t bits)
+{
+	uint64_t mask;
+	int rc;
+
+	rc = xscom_read(proc, PRD_IPOLL_REG_MASK, &mask);
+	if (rc)
+		return rc;
+
+	if (set)
+		mask |= bits;
+	else
+		mask &= ~bits;
+
+	return xscom_write(proc, PRD_IPOLL_REG_MASK, mask);
+}
+
+static int ipoll_record_and_mask_pending(uint32_t proc)
+{
+	uint64_t status;
+	int rc;
+
+	lock(&ipoll_lock);
+	rc = xscom_read(proc, PRD_IPOLL_REG_STATUS, &status);
+	if (!rc)
+		__ipoll_update_mask(proc, true, status);
+	unlock(&ipoll_lock);
+
+	if (!rc)
+		ipoll_status[proc] |= (status & PRD_IPOLL_MASK);
+
+	return rc;
+}
+
+/* Entry point for interrupts */
+void prd_psi_interrupt(uint32_t proc)
+{
+	int rc;
+
+	lock(&events_lock);
+
+	rc = ipoll_record_and_mask_pending(proc);
+	if (rc)
+		prlog(PR_ERR, "PRD: Failed to update IPOLL mask\n");
+
+	__prd_event(proc, EVENT_ATTN);
+
+	unlock(&events_lock);
+}
+
+void prd_tmgt_interrupt(uint32_t proc)
+{
+	prd_event(proc, EVENT_OCC_ERROR);
+}
+
+void prd_occ_reset(uint32_t proc)
+{
+	prd_event(proc, EVENT_OCC_RESET);
+}
+
+/* incoming message handlers */
+static int prd_msg_handle_attn_ack(struct opal_prd_msg *msg)
+{
+	int rc;
+
+	lock(&ipoll_lock);
+	rc = __ipoll_update_mask(msg->attn_ack.proc, false,
+			msg->attn_ack.ipoll_ack & PRD_IPOLL_MASK);
+	unlock(&ipoll_lock);
+
+	if (rc)
+		prlog(PR_ERR, "PRD: Unable to unmask ipoll!\n");
+
+	return rc;
+}
+
+static int prd_msg_handle_init(struct opal_prd_msg *msg)
+{
+	struct proc_chip *chip;
+
+	lock(&ipoll_lock);
+	for_each_chip(chip) {
+		__ipoll_update_mask(chip->id, false, msg->init.ipoll);
+	}
+	unlock(&ipoll_lock);
+
+	/* we're transitioning from inactive to active; send any pending tmgt
+	 * interrupts */
+	lock(&events_lock);
+	prd_active = true;
+	if (!prd_msg_inuse)
+		send_pending_events();
+	unlock(&events_lock);
+
+	return OPAL_SUCCESS;
+}
+
+static int prd_msg_handle_fini(void)
+{
+	struct proc_chip *chip;
+
+	lock(&events_lock);
+	prd_active = false;
+	unlock(&events_lock);
+
+	lock(&ipoll_lock);
+	for_each_chip(chip) {
+		__ipoll_update_mask(chip->id, true, -1ul);
+	}
+	unlock(&ipoll_lock);
+
+	return OPAL_SUCCESS;
+}
+
+/* Entry from the host above */
+static int64_t opal_prd_msg(struct opal_prd_msg *msg)
+{
+	int rc;
+
+	switch (msg->type) {
+	case OPAL_PRD_MSG_TYPE_INIT:
+		rc = prd_msg_handle_init(msg);
+		break;
+	case OPAL_PRD_MSG_TYPE_FINI:
+		rc = prd_msg_handle_fini();
+		break;
+	case OPAL_PRD_MSG_TYPE_ATTN_ACK:
+		rc = prd_msg_handle_attn_ack(msg);
+		break;
+	default:
+		rc = OPAL_UNSUPPORTED;
+	}
+
+	return rc;
+}
+
+void prd_init(void)
+{
+	struct proc_chip *chip;
+
+	/* mask everything */
+	lock(&ipoll_lock);
+	for_each_chip(chip) {
+		__ipoll_update_mask(chip->id, true, -1ul);
+	}
+	unlock(&ipoll_lock);
+
+	if (fsp_present()) {
+		/* todo: FSP implementation */
+		queue_prd_msg = queue_prd_msg_nop;
+	} else {
+		queue_prd_msg = queue_prd_msg_hbrt;
+		opal_register(OPAL_PRD_MSG, opal_prd_msg, 1);
+	}
+}
diff --git a/hw/psi.c b/hw/psi.c
index b96cb0f..baba154 100644
--- a/hw/psi.c
+++ b/hw/psi.c
@@ -298,7 +298,7 @@ static void handle_extra_interrupt(struct psi *psi)
 		p8_i2c_interrupt(psi->chip_id);
 	}
 	if (val & PSIHB_IRQ_STAT_LOCAL_ERR)
-		printf("PSI: ATTN irq received\n");
+		prd_psi_interrupt(psi->chip_id);
 	if (val & PSIHB_IRQ_STAT_HOST_ERR) {
 		if (platform.external_irq)
 			platform.external_irq(psi->chip_id);
diff --git a/include/skiboot.h b/include/skiboot.h
index 146ff80..9dfe0e7 100644
--- a/include/skiboot.h
+++ b/include/skiboot.h
@@ -218,6 +218,12 @@ extern void occ_send_dummy_interrupt(void);
 /* OCC load support */
 extern void occ_poke_load_queue(void);
 
+/* PRD */
+extern void prd_psi_interrupt(uint32_t proc);
+extern void prd_tmgt_interrupt(uint32_t proc);
+extern void prd_occ_reset(uint32_t proc);
+extern void prd_init(void);
+
 /* Flatten device-tree */
 extern void *create_dtb(const struct dt_node *root);
 
diff --git a/platforms/astbmc/habanero.c b/platforms/astbmc/habanero.c
index 0ec867f..8575be5 100644
--- a/platforms/astbmc/habanero.c
+++ b/platforms/astbmc/habanero.c
@@ -38,6 +38,8 @@ static bool habanero_probe(void)
 	/* Lot of common early inits here */
 	astbmc_early_init();
 
+	prd_init();
+
 	return true;
 }
 
diff --git a/platforms/astbmc/palmetto.c b/platforms/astbmc/palmetto.c
index 3435d0b..dee2b06 100644
--- a/platforms/astbmc/palmetto.c
+++ b/platforms/astbmc/palmetto.c
@@ -39,6 +39,8 @@ static bool palmetto_probe(void)
 	/* Lot of common early inits here */
 	astbmc_early_init();
 
+	prd_init();
+
 	return true;
 }
 


More information about the Skiboot mailing list