x86/mce: Stop mce_reign() from re-computing severity for every CPU
authorTony Luck <tony.luck@intel.com>
Tue, 8 Sep 2020 17:55:12 +0000 (10:55 -0700)
committerBorislav Petkov <bp@suse.de>
Mon, 14 Sep 2020 17:25:23 +0000 (19:25 +0200)
Back in commit:

  20d51a426fe9 ("x86/mce: Reuse one of the u16 padding fields in 'struct mce'")

a field was added to "struct mce" to save the computed error severity.

Make use of this in mce_reign() to avoid re-computing the severity
for every CPU.

In the case where the machine panics, one call to mce_severity() is
still needed in order to provide the correct message giving the reason
for the panic.

Signed-off-by: Tony Luck <tony.luck@intel.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lkml.kernel.org/r/20200908175519.14223-2-tony.luck@intel.com
arch/x86/kernel/cpu/mce/core.c

index a697baee9aa0b2ffaa9e849da09e3bdaff16d84d..5b1d5f33d60bb7d27da4b26a85052b5c3ce1ec74 100644 (file)
@@ -920,7 +920,6 @@ static void mce_reign(void)
        struct mce *m = NULL;
        int global_worst = 0;
        char *msg = NULL;
-       char *nmsg = NULL;
 
        /*
         * This CPU is the Monarch and the other CPUs have run
@@ -928,12 +927,10 @@ static void mce_reign(void)
         * Grade the severity of the errors of all the CPUs.
         */
        for_each_possible_cpu(cpu) {
-               int severity = mce_severity(&per_cpu(mces_seen, cpu),
-                                           mca_cfg.tolerant,
-                                           &nmsg, true);
-               if (severity > global_worst) {
-                       msg = nmsg;
-                       global_worst = severity;
+               struct mce *mtmp = &per_cpu(mces_seen, cpu);
+
+               if (mtmp->severity > global_worst) {
+                       global_worst = mtmp->severity;
                        m = &per_cpu(mces_seen, cpu);
                }
        }
@@ -943,8 +940,11 @@ static void mce_reign(void)
         * This dumps all the mces in the log buffer and stops the
         * other CPUs.
         */
-       if (m && global_worst >= MCE_PANIC_SEVERITY && mca_cfg.tolerant < 3)
+       if (m && global_worst >= MCE_PANIC_SEVERITY && mca_cfg.tolerant < 3) {
+               /* call mce_severity() to get "msg" for panic */
+               mce_severity(m, mca_cfg.tolerant, &msg, true);
                mce_panic("Fatal machine check", m, msg);
+       }
 
        /*
         * For UC somewhere we let the CPU who detects it handle it.