soc: qcom: Add support PFE WA

Sometimes the PFTLB entries get stuck in the invalid state
and new prefetches get dropped. As a workaround,
when a large number of L1 prefetches have been dropped,
above a threshold, the PMU can generate an interrupt.
And then reset PFE. The default threshold is 257.

Change-Id: Id6142037fa411575c3cbaa30dd08140ab6e5cb3f
Signed-off-by: Se Wang (Patrick) Oh <sewango@codeaurora.org>

drivers/soc/qcom/Kconfig

@@ -9,6 +9,16 @@ config MSM_INRUSH_CURRENT_MITIGATION
 	  huge load on a rail. This driver introduces an intermediate
 	  load to mitigate the in-rush current.
 
+config MSM_PFE_WA
+	depends on HW_PERF_EVENTS
+	bool "Enable a H/W PFE WA"
+	help
+	  Sometimes the PFTLB entries get stuck in the invalid state and new
+	  prefetches get dropped. For a workaround, count L1 prefeches dropped
+	  due to PFTLB miss and reset H/W PFE when a overflow happens.
+
+	  If unsure, say N.
+
 config MSM_SMEM
 	depends on ARCH_QCOM
 	depends on REMOTE_SPINLOCK_MSM

drivers/soc/qcom/Makefile

@@ -33,6 +33,7 @@ obj-$(CONFIG_MSM_PIL_SSR_GENERIC) += subsys-pil-tz.o
 obj-$(CONFIG_MSM_PIL_MSS_QDSP6V5) += pil-q6v5.o pil-msa.o pil-q6v5-mss.o
 obj-$(CONFIG_MSM_PIL)   +=      peripheral-loader.o
 obj-$(CONFIG_MSM_CORE_CTL_HELPER) += core_ctl_helper.o
+obj-$(CONFIG_MSM_PFE_WA) += pfe-wa.o
 
 ifdef CONFIG_MSM_SUBSYSTEM_RESTART
 	obj-y += subsystem_notif.o

drivers/soc/qcom/pfe-wa.c

@@ -0,0 +1,144 @@
+/* Copyright (c) 2015, The Linux Foundation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 and
+ * only version 2 as published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ */
+
+#include <linux/cpu.h>
+#include <linux/cpu_pm.h>
+#include <linux/irq.h>
+#include <linux/module.h>
+#include <linux/of.h>
+#include <linux/perf_event.h>
+#include <linux/slab.h>
+
+/*
+ * BIT 19:16 -> prefix
+ * BIT 15:12 -> Reg number
+ * BIT 11:4  -> Code number
+ * BIT  3:0  -> Group number
+ */
+#define L1_DROP_EVENT 0x12513
+
+#define EHWPF_BITS 0x3
+#define EHWPF_SHIFT 0x0
+#define EHWPF_MASK ~(EHWPF_BITS << EHWPF_SHIFT)
+
+/*
+ * EHWPF
+ * 0x0: turn off PFE
+ * 0x1: turn on l1 cache PFE
+ * 0x2: turn on l1 and l2 cache PFE
+ * 0x3: turn on l1,l2, l3 cache PFE
+ */
+#define EHWPF_OFF  (0x0 << EHWPF_SHIFT)
+#define EHWPF_L1L2 (0x2 << EHWPF_SHIFT)
+
+struct l1_drop_data {
+	struct notifier_block nb_cpu;
+	struct perf_event *l1_drop_events[NR_CPUS];
+};
+
+static unsigned long max_allowed_cntr = 257;
+module_param(max_allowed_cntr, ulong, 0);
+
+static void l1_drop_handler(struct perf_event *event,
+		struct perf_sample_data *data,
+		struct pt_regs *regs)
+{
+	u32 val;
+
+	asm volatile ("mrs %0, s3_1_c11_c4_7" : "=r" (val));
+	val = (val & EHWPF_MASK) | EHWPF_OFF;
+	asm volatile ("msr s3_1_c11_c4_7, %0" : : "r" (val));
+	val = (val & EHWPF_MASK) | EHWPF_L1L2;
+	isb();
+	asm volatile ("msr s3_1_c11_c4_7, %0" : : "r" (val));
+	isb();
+}
+
+static void l1_drop_event_create(int cpu, void *info)
+{
+	struct l1_drop_data *drv = info;
+	struct perf_event *event = drv->l1_drop_events[cpu];
+	struct perf_event_attr attr = {
+		.pinned = 1,
+		.disabled = 0,
+		.sample_period = max_allowed_cntr,
+		.type = PERF_TYPE_RAW,
+		.config = L1_DROP_EVENT,
+		.size = sizeof(struct perf_event_attr),
+	};
+
+	if (event)
+		return;
+	event = perf_event_create_kernel_counter(&attr, cpu, NULL,
+			l1_drop_handler,
+			drv);
+	if (IS_ERR(event)) {
+		pr_err("PERF Event creation failed on cpu %d ptr_err %ld\n",
+				cpu, PTR_ERR(event));
+		return;
+	}
+
+	drv->l1_drop_events[cpu] = event;
+}
+
+static int l1_drop_cpu_notify(struct notifier_block *self,
+		unsigned long action, void *hcpu)
+{
+	struct l1_drop_data *data = container_of(self, struct l1_drop_data,
+					nb_cpu);
+	unsigned long cpu = (unsigned long)hcpu;
+
+	switch (action & ~CPU_TASKS_FROZEN) {
+	case CPU_ONLINE:
+		l1_drop_event_create(cpu, data);
+		break;
+	};
+
+	return NOTIFY_OK;
+}
+
+static struct of_device_id pfe_wa_of_match[] = {
+	{.compatible = "qcom,kryo-pmuv3", },
+	{}
+};
+
+static int __init msm_pfe_wa_init(void)
+{
+	int cpu;
+	struct device_node *np;
+	struct l1_drop_data *l1_drop;
+
+	np = of_find_matching_node(NULL, pfe_wa_of_match);
+	if (!np) {
+		pr_debug("Disable PFE WA - PMU is not enabled\n");
+		return -ENODEV;
+	}
+	if (!of_find_property(np, "qcom,enable-pfe-wa", NULL)) {
+		pr_debug("Disable PFE WA\n");
+		return -ENODEV;
+	}
+
+	l1_drop = kzalloc(sizeof(struct l1_drop_data), GFP_KERNEL);
+	if (!l1_drop)
+		return -ENOMEM;
+
+	l1_drop->nb_cpu.notifier_call = l1_drop_cpu_notify;
+	register_cpu_notifier(&l1_drop->nb_cpu);
+	get_online_cpus();
+	for_each_online_cpu(cpu)
+		l1_drop_event_create(cpu, l1_drop);
+	put_online_cpus();
+
+	return 0;
+}
+
+late_initcall(msm_pfe_wa_init);