Commit 4456b18b authored Nov 05, 2015 by Steven Cahail

soc: qcom: glink: Increase SSR timeout, notify RPM in single-threaded mode

In some cases, delays in context switching can cause SSR notifications
to take a long time to send. This can cause responses from remote
processors to be received after the half-second timeout has expired. In the
case of RPM, the delay is further increased because the notification is
sent in the "request intent" mode, which queues a work item to send the
notification rather than sending it in the same context. If the RPM fails
to respond in time, a kernel panic is initiated.

Increase the SSR timeout to one second and send notifications to RPM in
single-threaded mode to ensure that responses can be received in time.

Change-Id: I987e3970c7bdf4e677f046cc7a3d39c8586871fa
Signed-off-by: Steven Cahail <scahail@codeaurora.org>

parent 3b74bf1e

drivers/soc/qcom/glink_ssr.c

+12 −4

Original line number	Diff line number	Diff line
		@@ -25,7 +25,7 @@
		#include <soc/qcom/subsystem_restart.h>
		#include "glink_private.h"

		#define GLINK_SSR_REPLY_TIMEOUT (HZ / 2)
		#define GLINK_SSR_REPLY_TIMEOUT HZ
		#define GLINK_SSR_EVENT_INIT ~0
		#define NUM_LOG_PAGES 3

		@@ -535,9 +535,17 @@ int notify_for_subsystem(struct subsys_info *ss_info)
		continue;
		}

		ret = glink_tx(handle, (void *)ss_leaf_entry->cb_data,
		(void *)do_cleanup_data,
		sizeof(struct do_cleanup_msg), true);
		if (strcmp(ss_leaf_entry->ssr_name, "rpm"))
		ret = glink_tx(handle, ss_leaf_entry->cb_data,
		do_cleanup_data,
		sizeof(*do_cleanup_data),
		GLINK_TX_REQ_INTENT);
		else
		ret = glink_tx(handle, ss_leaf_entry->cb_data,
		do_cleanup_data,
		sizeof(*do_cleanup_data),
		GLINK_TX_SINGLE_THREADED);

		if (ret) {
		GLINK_SSR_ERR("<SSR> %s: tx failed, ret[%d], %s[%d]\n",
		__func__, ret, "resp. remaining",