powerpc: Fix corruption when grabbing FWNMI data

The FWNMI code uses a global buffer without any locks to read the RTAS error
information. If two CPUs take a machine check at once then we will corrupt
this buffer.

Since most FWNMI rtas messages are not of the extended type, we can create a
64bit percpu buffer and use it where possible. If we do receive an extended
RTAS log then we fall back to the old behaviour of using the global buffer.

Signed-off-by: Anton Blanchard <anton@samba.org>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>

diff --git a/arch/powerpc/platforms/pseries/ras.c b/arch/powerpc/platforms/pseries/ras.c
index d194150cf342e0b2e331740c588fbba10a66c8ff..c55d7ad9c6483f480b611a09444249f8db0b876b 100644 (file)
--- a/arch/powerpc/platforms/pseries/ras.c
+++ b/arch/powerpc/platforms/pseries/ras.c
@@ -54,7 +54,8 @@
 static unsigned char ras_log_buf[RTAS_ERROR_LOG_MAX];
 static DEFINE_SPINLOCK(ras_log_buf_lock);
 
-static char mce_data_buf[RTAS_ERROR_LOG_MAX];
+static char global_mce_data_buf[RTAS_ERROR_LOG_MAX];
+static DEFINE_PER_CPU(__u64, mce_data_buf);
 
 static int ras_get_sensor_state_token;
 static int ras_check_exception_token;
@@ -196,12 +197,24 @@ static irqreturn_t ras_error_interrupt(int irq, void *dev_id)
        return IRQ_HANDLED;
 }
 
-/* Get the error information for errors coming through the
+/*
+ * Some versions of FWNMI place the buffer inside the 4kB page starting at
+ * 0x7000. Other versions place it inside the rtas buffer. We check both.
+ */
+#define VALID_FWNMI_BUFFER(A) \
+       ((((A) >= 0x7000) && ((A) < 0x7ff0)) || \
+       (((A) >= rtas.base) && ((A) < (rtas.base + rtas.size - 16))))
+
+/*
+ * Get the error information for errors coming through the
  * FWNMI vectors.  The pt_regs' r3 will be updated to reflect
  * the actual r3 if possible, and a ptr to the error log entry
  * will be returned if found.
  *
- * The mce_data_buf does not have any locks or protection around it,
+ * If the RTAS error is not of the extended type, then we put it in a per
+ * cpu 64bit buffer. If it is the extended type we use global_mce_data_buf.
+ *
+ * The global_mce_data_buf does not have any locks or protection around it,
  * if a second machine check comes in, or a system reset is done
  * before we have logged the error, then we will get corruption in the
  * error log.  This is preferable over holding off on calling
@@ -210,20 +223,31 @@ static irqreturn_t ras_error_interrupt(int irq, void *dev_id)
  */
 static struct rtas_error_log *fwnmi_get_errinfo(struct pt_regs *regs)
 {
-       unsigned long errdata = regs->gpr[3];
-       struct rtas_error_log *errhdr = NULL;
        unsigned long *savep;
+       struct rtas_error_log *h, *errhdr = NULL;
 
-       if ((errdata >= 0x7000 && errdata < 0x7fff0) ||
-           (errdata >= rtas.base && errdata < rtas.base + rtas.size - 16)) {
-               savep = __va(errdata);
-               regs->gpr[3] = savep[0];        /* restore original r3 */
-               memset(mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
-               memcpy(mce_data_buf, (char *)(savep + 1), RTAS_ERROR_LOG_MAX);
-               errhdr = (struct rtas_error_log *)mce_data_buf;
+       if (!VALID_FWNMI_BUFFER(regs->gpr[3])) {
+               printk(KERN_ERR "FWNMI: corrupt r3\n");
+               return NULL;
+       }
+
+       savep = __va(regs->gpr[3]);
+       regs->gpr[3] = savep[0];        /* restore original r3 */
+
+       /* If it isn't an extended log we can use the per cpu 64bit buffer */
+       h = (struct rtas_error_log *)&savep[1];
+       if (!h->extended) {
+               memcpy(&__get_cpu_var(mce_data_buf), h, sizeof(__u64));
+               errhdr = (struct rtas_error_log *)&__get_cpu_var(mce_data_buf);
        } else {
-               printk("FWNMI: corrupt r3\n");
+               int len;
+
+               len = max_t(int, 8+h->extended_log_length, RTAS_ERROR_LOG_MAX);
+               memset(global_mce_data_buf, 0, RTAS_ERROR_LOG_MAX);
+               memcpy(global_mce_data_buf, h, len);
+               errhdr = (struct rtas_error_log *)global_mce_data_buf;
        }
+
        return errhdr;
 }
 
@@ -235,7 +259,7 @@ static void fwnmi_release_errinfo(void)
 {
        int ret = rtas_call(rtas_token("ibm,nmi-interlock"), 0, 1, NULL);
        if (ret != 0)
-               printk("FWNMI: nmi-interlock failed: %d\n", ret);
+               printk(KERN_ERR "FWNMI: nmi-interlock failed: %d\n", ret);
 }
 
 int pSeries_system_reset_exception(struct pt_regs *regs)