mce-inject: extend mce-inject for support threshold interrupt event injection on ADM platform

There are three ways that have been used to report machine check event. And they are MCE, CMCI/Threshold Interrupt, and POLL. On the Intel platform, CMCI/Threshold Interrupt and POLL share the same event handler
- machine_check_poll(). However, on the AMD platform, they have a separate event handler. amd_threshold_interrupt() is used for handling Threshold Interrupt event. And machine_check_poll() has been used for polling other events.

This patch introduces a new flag MCJ_INTERRUPT that will be used to separate CMCI/Threshold Interrupt and POLL handler in mce-inject.

Signed-off-by Chen Yucong <[email protected] <mailto:[email protected]>>

---
 arch/x86/include/asm/mce.h              |    5 +++--
 arch/x86/kernel/cpu/mcheck/mce-inject.c |   16 ++++++++++++++++
 arch/x86/kernel/cpu/mcheck/threshold.c  |    1 +
 3 files changed, 20 insertions(+), 2 deletions(-)

arch/x86/include/asm/mce.h +3 -2

index 3a430ad..cf25839 100644
@@ -70,8 +70,9 @@
 #define MCJ_CTX_PROCESS                0x1  /* inject context: process */
 #define MCJ_CTX_IRQ            0x2  /* inject context: IRQ */
 #define MCJ_NMI_BROADCAST      0x4  /* do NMI broadcasting */
-#define MCJ_EXCEPTION          0x8  /* raise as exception */
-#define MCJ_IRQ_BROADCAST      0x10 /* do IRQ broadcasting */
+#define MCJ_IRQ_BROADCAST      0x8  /* do IRQ broadcasting */
+#define MCJ_EXCEPTION          0x10  /* raise as exception */
+#define MCJ_INTERRUPT          0x20  /* raise as interruption */
 
 #define MCE_OVERFLOW 0         /* bit 0 in flags means overflow */
 

arch/x86/kernel/cpu/mcheck/mce-inject.c +12 -0

index 4cfba43..8428746 100644
@@ -59,6 +59,16 @@ static void raise_poll(struct mce *m)
        m->finished = 0;
 }
 
+static void raise_interrupt(struct mce *m)
+{
+       unsigned long flags;
+
+       local_irq_save(flags);
+       mce_threshold_vector();
+       local_irq_restore(flags);
+       m->finished = 0;
+}
+
 static void raise_exception(struct mce *m, struct pt_regs *pregs)
 {
        struct pt_regs regs;

@@ -89,6 +99,8 @@ static int mce_raise_notify(unsigned int cmd, struct 
 pt_regs *regs)
        cpumask_clear_cpu(cpu, mce_inject_cpumask);
        if (m->inject_flags & MCJ_EXCEPTION)
                raise_exception(m, regs);
+       else if (m->inject_flags & MCJ_INTERRUPT)
+               raise_interrupt(m);
        else if (m->status)
                raise_poll(m);

        return NMI_HANDLED;

arch/x86/kernel/cpu/mcheck/mce-inject.c +4 -0

@@ -132,6 +144,10 @@ static int raise_local(void)
                        ret = -EINVAL;
                }
                printk(KERN_INFO "MCE exception done on CPU %d\n", cpu);
+       } else if (m->inject_flags & MCJ_INTERRUPT) {
+               printk(KERN_INFO "Raising threshold interrupt on CPU 
 %d\n", cpu);
+               raise_interrupt(m);
+               printk(KERN_INFO "Threshold interrupt done on CPU 
 %d\n", cpu);
        } else if (m->status) {

printk(KERN_INFO "Starting machine check poll CPU

%d\n", cpu);
                raise_poll(m);

arch/x86/kernel/cpu/mcheck/threshold.c +1 -1

index 7245980..e324bf9 100644
@@ -17,6 +17,7 @@ static void default_threshold_interrupt(void)
 }
 
 void (*mce_threshold_vector)(void) = default_threshold_interrupt;
+EXPORT_SYMBOL_GPL(mce_threshold_vector);
 
 static inline void __smp_threshold_interrupt(void)
 {
--

1.7.10.4

Quoted 12 lines · Aravind Gopalakrishnan

There are three ways that have been used to report machine check event. And they are MCE, CMCI/Threshold Interrupt, and POLL. On the Intel platform, CMCI/Threshold Interrupt and POLL share the same event handler
- machine_check_poll(). However, on the AMD platform, they have a separate event handler. amd_threshold_interrupt() is used for handling Threshold Interrupt event. And machine_check_poll() has been used for polling other events.

This patch introduces a new flag MCJ_INTERRUPT that will be used to separate CMCI/Threshold Interrupt and POLL handler in mce-inject.

Quoted 7 lines

Signed-off-by Chen Yucong <[email protected] <mailto:[email protected]>>

---

arch/x86/include/asm/mce.h              |    5 +++--
arch/x86/kernel/cpu/mcheck/mce-inject.c |   16 ++++++++++++++++
arch/x86/kernel/cpu/mcheck/threshold.c  |    1 +
3 files changed, 20 insertions(+), 2 deletions(-)

We currently test decoding logic on AMD by performing mce injections using edac/mce_amd_inj.c,
So instead of modifying mce-inject just for testing amd_threshold_interrupt(),
Why not put it under mce_amd_inj? (It's AMD specific code anyway)

Quoted 22 lines · Aravind Gopalakrishnan

Quoted 19 lines · Aravind Gopalakrishnan

There are three ways that have been used to report machine check event. And they are MCE, CMCI/Threshold Interrupt, and POLL. On the Intel platform, CMCI/Threshold Interrupt and POLL share the same event handler
- machine_check_poll(). However, on the AMD platform, they have a separate event handler. amd_threshold_interrupt() is used for handling Threshold Interrupt event. And machine_check_poll() has been used for polling other events.

This patch introduces a new flag MCJ_INTERRUPT that will be used to separate CMCI/Threshold Interrupt and POLL handler in mce-inject.

Signed-off-by Chen Yucong <[email protected] <mailto:[email protected]>>

---

arch/x86/include/asm/mce.h              |    5 +++--
arch/x86/kernel/cpu/mcheck/mce-inject.c |   16 ++++++++++++++++
arch/x86/kernel/cpu/mcheck/threshold.c  |    1 +
3 files changed, 20 insertions(+), 2 deletions(-)

Quoted 5 lines

We currently test decoding logic on AMD by performing mce injections using edac/mce_amd_inj.c,
So instead of modifying mce-inject just for testing amd_threshold_interrupt(),
Why not put it under mce_amd_inj? (It's AMD specific code anyway)

Right, I think this is supposed to be vendor-agnostic as it is calling mce_threshold_vector() directly.

Btw, I wouldn't mind if someone would sit down and unify those injection methods and come up with a saner interface which can actually be used by humans, not those yucky files you feed mce-inject with...

Quoted 12 lines · Aravind Gopalakrishnan

There are three ways that have been used to report machine check event. And they are MCE, CMCI/Threshold Interrupt, and POLL. On the Intel platform, CMCI/Threshold Interrupt and POLL share the same event handler
- machine_check_poll(). However, on the AMD platform, they have a separate event handler. amd_threshold_interrupt() is used for handling Threshold Interrupt event. And machine_check_poll() has been used for polling other events.

This patch introduces a new flag MCJ_INTERRUPT that will be used to separate CMCI/Threshold Interrupt and POLL handler in mce-inject.

Quoted 7 lines

Signed-off-by Chen Yucong <[email protected] <mailto:[email protected]>>

---
 arch/x86/include/asm/mce.h              |    5 +++--
 arch/x86/kernel/cpu/mcheck/mce-inject.c |   16 ++++++++++++++++
 arch/x86/kernel/cpu/mcheck/threshold.c  |    1 +
 3 files changed, 20 insertions(+), 2 deletions(-)

We currently test decoding logic on AMD by performing mce injections using edac/mce_amd_inj.c,
So instead of modifying mce-inject just for testing amd_threshold_interrupt(),
Why not put it under mce_amd_inj? (It's AMD specific code anyway)