On Tue, 2014-09-30 at 12:09 +0200, Borislav Petkov wrote:
>
> Now let me repeat my question: how are you testing your patches?
>
There are no any hardware facilities that can help me to inject some
MCE errors. So I have to modify the kernel source code for testing my
patches.
My method is based on the `mce-injection' that is better suited to
Intel processors. So I have replaced rdmsrl/wrmsrl/rdmsr_safe with
mce_rdmsrl/mce_wrmsrl/mce_rdmsr_safe in mce_amd.c. But I use a new
kernel module for error injection instead of writing /dev/mcelog.
For more detailed information about testing, you can refer the
attachments.
thx!
cyc
diff -uNr amd_inject/linux-3.16.3/arch/x86/include/asm/mce.h linux-3.16.3/arch/x86/include/asm/mce.h
--- amd_inject/linux-3.16.3/arch/x86/include/asm/mce.h 2014-09-18 01:22:16.000000000 +0800
+++ linux-3.16.3/arch/x86/include/asm/mce.h 2014-10-01 09:36:06.302670241 +0800
@@ -166,6 +166,7 @@
#endif
#ifdef CONFIG_X86_MCE_AMD
+void raise_amd_threshold_event(void);
void mce_amd_feature_init(struct cpuinfo_x86 *c);
#else
static inline void mce_amd_feature_init(struct cpuinfo_x86 *c) { }
@@ -185,10 +186,14 @@
MCP_DONTLOG = (1 << 2), /* only clear, don't log */
};
void machine_check_poll(enum mcp_flags flags, mce_banks_t *b);
+u64 mce_rdmsrl(u32 msr);
+void mce_wrmsrl(u32 msr, u64 v);
+int mce_rdmsr_safe(u32 msr, u32 *low, u32 *high);
int mce_notify_irq(void);
void mce_notify_process(void);
+extern int amd_inject;
DECLARE_PER_CPU(struct mce, injectm);
extern void register_mce_write_callback(ssize_t (*)(struct file *filp,
diff -uNr amd_inject/linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce_amd.c linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce_amd.c
--- amd_inject/linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce_amd.c 2014-09-18 01:22:16.000000000 +0800
+++ linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce_amd.c 2014-10-01 11:09:07.817585622 +0800
@@ -274,6 +274,7 @@
struct mce m;
mce_setup(&m);
+ m.mcgstatus = mce_rdmsrl(MSR_IA32_MCG_STATUS);
/* assume first bank caused it */
for (bank = 0; bank < mca_cfg.banks; ++bank) {
@@ -291,7 +292,7 @@
++address;
}
- if (rdmsr_safe(address, &low, &high))
+ if (mce_rdmsr_safe(address, &low, &high))
break;
if (!(high & MASK_VALID_HI)) {
@@ -305,26 +306,35 @@
(high & MASK_LOCKED_HI))
continue;
- /*
- * Log the machine check that caused the threshold
- * event.
- */
- machine_check_poll(MCP_TIMESTAMP,
- &__get_cpu_var(mce_poll_banks));
-
if (high & MASK_OVERFLOW_HI) {
- rdmsrl(address, m.misc);
- rdmsrl(MSR_IA32_MC0_STATUS + bank * 4,
- m.status);
+ m.misc = mce_rdmsrl(address);
+ m.status = mce_rdmsrl(MSR_IA32_MC0_STATUS + bank * 4);
+ if (m.status & MCI_STATUS_ADDRV)
+ m.addr = mce_rdmsrl(MSR_IA32_MC0_ADDR + bank * 4);
m.bank = K8_MCE_THRESHOLD_BASE
+ bank * NR_BLOCKS
+ block;
mce_log(&m);
+ mce_wrmsrl(MSR_IA32_MC0_STATUS + bank * 4, 0);
return;
}
}
}
+
+ /*
+ * Log the machine check that caused the threshold
+ * event.
+ */
+ machine_check_poll(MCP_TIMESTAMP,
+ &__get_cpu_var(mce_poll_banks));
+
+}
+
+void raise_amd_threshold_event(void)
+{
+ amd_threshold_interrupt();
}
+EXPORT_SYMBOL_GPL(raise_amd_threshold_event);
/*
* Sysfs Interface
diff -uNr amd_inject/linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce.c linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce.c
--- amd_inject/linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce.c 2014-09-18 01:22:16.000000000 +0800
+++ linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce.c 2014-10-01 09:40:13.269228358 +0800
@@ -48,6 +48,9 @@
#include "mce-internal.h"
+int amd_inject = 0;
+EXPORT_PER_CPU_SYMBOL_GPL(amd_inject);
+
static DEFINE_MUTEX(mce_chrdev_read_mutex);
#define rcu_dereference_check_mce(p) \
@@ -131,6 +134,7 @@
m->apicid = cpu_data(m->extcpu).initial_apicid;
rdmsrl(MSR_IA32_MCG_CAP, m->mcgcap);
}
+EXPORT_SYMBOL_GPL(mce_setup);
DEFINE_PER_CPU(struct mce, injectm);
EXPORT_PER_CPU_SYMBOL_GPL(injectm);
@@ -391,7 +395,7 @@
}
/* MSR access wrappers used for error injection */
-static u64 mce_rdmsrl(u32 msr)
+u64 mce_rdmsrl(u32 msr)
{
u64 v;
@@ -415,8 +419,9 @@
return v;
}
-static void mce_wrmsrl(u32 msr, u64 v)
+void mce_wrmsrl(u32 msr, u64 v)
{
if (__this_cpu_read(injectm.finished)) {
int offset = msr_to_offset(msr);
@@ -427,6 +432,18 @@
}
wrmsrl(msr, v);
}
+
+int mce_rdmsr_safe(u32 msr, u32 *low, u32 *high)
+{
+ u64 __val = mce_rdmsrl(msr);
+
+ (*low) = (u32)__val;
+ (*high) = (u32)(__val >> 32);
+
+ return 0;
+}
/*
* Collect all global (w.r.t. this processor) status about this machine
@@ -1637,6 +1654,7 @@
mce_adjust_timer = mce_intel_adjust_timer;
break;
case X86_VENDOR_AMD:
+ amd_inject = 1;
mce_amd_feature_init(c);
break;
default:
diff -uNr amd_inject/linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce-inject.c linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce-inject.c
--- amd_inject/linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce-inject.c 2014-09-18 01:22:16.000000000 +0800
+++ linux-3.16.3/arch/x86/kernel/cpu/mcheck/mce-inject.c 2014-09-30 22:38:30.138557839 +0800
@@ -54,7 +54,10 @@
memset(&b, 0xff, sizeof(mce_banks_t));
local_irq_save(flags);
- machine_check_poll(0, &b);
+ if (!amd_inject)
+ machine_check_poll(0, &b);
+ else
+ mce_threshold_vector();
local_irq_restore(flags);
m->finished = 0;
}
diff -uNr amd_inject/linux-3.16.3/arch/x86/kernel/cpu/mcheck/threshold.c linux-3.16.3/arch/x86/kernel/cpu/mcheck/threshold.c
--- amd_inject/linux-3.16.3/arch/x86/kernel/cpu/mcheck/threshold.c 2014-09-18 01:22:16.000000000 +0800
+++ linux-3.16.3/arch/x86/kernel/cpu/mcheck/threshold.c 2014-10-01 08:49:06.140738192 +0800
@@ -17,6 +17,7 @@
}
void (*mce_threshold_vector)(void) = default_threshold_interrupt;
+EXPORT_SYMBOL_GPL(mce_threshold_vector);
static inline void __smp_threshold_interrupt(void)
{
/*
* Copyright Chen Yucong<sla...@gmail.com> 2014
*
* This program is free software; you can redistribute it and/or
* modify it under the terms of the GNU General Public License
* as published by the Free Software Foundation; version 2
* of the License.
*/
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/smp.h>
#include <linux/cpu.h>
#include <asm/mce.h>
#include <asm/msr.h>
#include <asm/amd_nb.h>
#define MASK_OVERFLOW 0x0001000000000000
/* Update fake mce registers on current CPU. */
static void inject_mce(struct mce *m)
{
struct mce *i = &per_cpu(injectm, m->extcpu);
/* Make sure no one reads partially written injectm */
i->finished = 0;
mb();
m->finished = 0;
/* First set the fields after finished */
i->extcpu = m->extcpu;
mb();
/* Now write record in order, finished last (except above) */
memcpy(i, m, sizeof(struct mce));
/* Finally activate it */
mb();
i->finished = 1;
}
static void raise_mce(void)
{
struct mce m;
mce_setup(&m);
m.status = 0X8C00000000000000;
m.misc = 0XC008000000000000 | MASK_OVERFLOW;
//m.misc = 0XC008000000000000;
m.bank = 4;
m.addr = 0xabcdef;
inject_mce(&m);
raise_amd_threshold_event();
}
static int __init amd_inject_init(void)
{
raise_mce();
pr_info("amd_inject module loaded ...\n");
return 0;
}
static void __exit amd_inject_exit(void)
{
pr_info("amd_inject module unloaded ...\n");
}
module_init(amd_inject_init);
module_exit(amd_inject_exit);
/*
* Cannot tolerate unloading currently because we cannot
* guarantee all openers of mce_chrdev will get a reference to us.
*/
MODULE_LICENSE("GPL");
