From: René Rebe <rene@exactco.de>
Subject: [PATCH] MIPS: IP32: count CRIME error interrupts, rate limit their messages

Count CRIME memory and CPU error interrupts with the last address and
status in debugfs crime-errors, and rate limit the non-fatal messages,
so error bursts (e.g. from speculative accesses) can be measured without
flooding the log.  Multiple-bit and hard memory errors stay fatal.

Signed-off-by: René Rebe <rene@exactco.de>
---
--- a/arch/mips/sgi-ip32/crime.c
+++ b/arch/mips/sgi-ip32/crime.c
@@ -6,6 +6,8 @@
  * Copyright (C) 2001, 2003 Keith M Wesolowski
  * Copyright (C) 2005 Ilya A. Volynets <ilya@total-knowledge.com>
  */
+#include <linux/debugfs.h>
+#include <linux/seq_file.h>
 #include <linux/types.h>
 #include <linux/init.h>
 #include <linux/kernel.h>
@@ -25,6 +27,12 @@ struct sgi_mace __iomem *mace;
 
 EXPORT_SYMBOL_GPL(mace);
 
+/* error interrupt counts for debugfs crime-errors, e.g. speculative accesses */
+static struct {
+	unsigned long mem, mem_fatal, cpu;
+	unsigned long mem_addr, mem_stat, cpu_addr, cpu_stat;
+} crime_errors;
+
 void __init crime_init(void)
 {
 	unsigned int id, rev;
@@ -48,6 +56,15 @@ irqreturn_t crime_memerr_intr(int irq, v
 
 	stat = crime->mem_error_stat & CRIME_MEM_ERROR_STAT_MASK;
 	addr = crime->mem_error_addr & CRIME_MEM_ERROR_ADDR_MASK;
+	crime_errors.mem++;
+	crime_errors.mem_addr = addr;
+	crime_errors.mem_stat = stat;
+	/* uncorrectable errors stay fatal, the rest is counted and rate limited */
+	if (!(stat & (CRIME_MEM_ERROR_MULTIPLE | CRIME_MEM_ERROR_HARD_ERR)) &&
+	    !printk_ratelimit()) {
+		crime->mem_error_stat = 0;
+		return IRQ_HANDLED;
+	}
 
 	printk("CRIME memory error at 0x%08lx status 0x%08lx<", addr, stat);
 
@@ -84,6 +101,7 @@ irqreturn_t crime_memerr_intr(int irq, v
 	crime->mem_error_stat = 0;
 
 	if (fatal) {
+		crime_errors.mem_fatal++;
 		printk(KERN_CONT "FATAL>\n");
 		panic("Fatal memory error.");
 	} else
@@ -98,8 +116,29 @@ irqreturn_t crime_cpuerr_intr(int irq, v
 	unsigned long addr = crime->cpu_error_addr & CRIME_CPU_ERROR_ADDR_MASK;
 
 	addr <<= 2;
-	printk("CRIME CPU error at 0x%09lx status 0x%08lx\n", addr, stat);
+	crime_errors.cpu++;
+	crime_errors.cpu_addr = addr;
+	crime_errors.cpu_stat = stat;
+	pr_err_ratelimited("CRIME CPU error at 0x%09lx status 0x%08lx\n", addr, stat);
 	crime->cpu_error_stat = 0;
 
 	return IRQ_HANDLED;
 }
+
+static int crime_errors_show(struct seq_file *m, void *v)
+{
+	seq_printf(m, "memory errors %lu (fatal %lu), last at 0x%08lx status 0x%08lx\n",
+		   crime_errors.mem, crime_errors.mem_fatal, crime_errors.mem_addr,
+		   crime_errors.mem_stat);
+	seq_printf(m, "cpu errors %lu, last at 0x%09lx status 0x%08lx\n", crime_errors.cpu,
+		   crime_errors.cpu_addr, crime_errors.cpu_stat);
+	return 0;
+}
+DEFINE_SHOW_ATTRIBUTE(crime_errors);
+
+static int __init crime_errors_debugfs_init(void)
+{
+	debugfs_create_file("crime-errors", 0444, NULL, NULL, &crime_errors_fops);
+	return 0;
+}
+late_initcall(crime_errors_debugfs_init);