On Tue, 2013-05-07 at 15:19 +0200, Borislav Petkov wrote:
> On Mon, May 06, 2013 at 08:32:46PM -0700, Max Asbock wrote:
> > But this is glue code between MCE and APEI, therefore I thought a
> > file named mce-apei.c would be a good place. If we put it into apei
> > then we might have to export some of the MCE interfaces, whereas
> > apei_hest_parse() is already exported.
>
> Right, I don't want to expose any MCA internals to other subsystems
> unless it is really necessary.
>
> AFAICT, you need an ON and OFF switch for CMCI which is callable from
> outside. So you can adapt/adjust the code in mce_intel.c to do so
> without adding any other code.
>
> Unless I'm missing something. But I don't think so, the high-level
> sequence looks like this:
>
> * MCA init
> -> CMCI init
> * APEI init:
> -> if FF
> -> CMCI off
>
> At a quick glance, simply doing:
>
> on_each_cpu(cmci_clear_func, NULL, 1);
>
> should work. You'd need to define a proper cmci_clear_func prototype but
> that should be trivial...
>
So something like the following patch might be closer:
Signed-off-by: Max Asbock <masbock@xxxxxxxxxxxxxxxxxx>
---
mce-apei.c | 52 ++++++++++++++++++++++++++++++++++++++++++++++++++++
1 file changed, 52 insertions(+)
diff -burpN linux-3.9/arch/x86/kernel/cpu/mcheck/mce-apei.c linux-3.9.ff2/arch/x86/kernel/cpu/mcheck/mce-apei.c
--- linux-3.9/arch/x86/kernel/cpu/mcheck/mce-apei.c 2013-05-02 09:38:16.000000000 -0700
+++ linux-3.9.ff2/arch/x86/kernel/cpu/mcheck/mce-apei.c 2013-05-07 08:12:16.000000000 -0700
@@ -32,6 +32,7 @@
#include <linux/kernel.h>
#include <linux/acpi.h>
#include <linux/cper.h>
+#include <linux/init.h>
#include <acpi/apei.h>
#include <asm/mce.h>
@@ -148,3 +149,54 @@ int apei_clear_mce(u64 record_id)
{
return erst_clear(record_id);
}
+
+
+/* Support for Firmware First (FF) mode for Corrected Machine Checks
+ * as defined by APEI in the Hardware Error Source Table (HEST).
+ * (see section 18.3.2.2 in the ACPI spec, version 5.0)
+ *
+ * When Firmware First mode is specified in HEST for Corrected Machine Checks,
+ * these errors are expected to be reported by the firmware through a
+ * Generic Harware Event Source (GHES). In this case error reporting
+ * through CMCI should not be enabled so it won't interfere with the firmware.
+ *
+ * In the boot sequence HEST parsing is initialized after
+ * CPUs, therefore initially we don't know if Firmware First is set.
+ * We enable CMCI at first and then disable it in a late init call if FF is set.
+ */
+
+static bool cmc_firmware_first;
+
+static int check_cmc_firmware_first(struct acpi_hest_header *hest_hdr, void *d)
+{
+ if (hest_hdr->type == ACPI_HEST_TYPE_IA32_CORRECTED_CHECK) {
+ struct acpi_hest_ia_corrected *cmc;
+
+ cmc = (struct acpi_hest_ia_corrected *)hest_hdr;
+ if (cmc->flags & ACPI_HEST_FIRMWARE_FIRST) {
+ cmc_firmware_first = true;
+ return 1;
+ }
+ }
+ return 0;
+}
+
+static void disable_cmci(void *data)
+{
+ if (!mce_available(__this_cpu_ptr(&cpu_info)))
+ return;
+ cmci_clear();
+}
+
+static __init int honor_cmc_firmware_first(void)
+{
+ apei_hest_parse(check_cmc_firmware_first, NULL);
+
+ if (cmc_firmware_first && !mca_cfg.cmci_disabled) {
+ on_each_cpu(disable_cmci, NULL, 1);
+ mca_cfg.cmci_disabled = true;
+ }
+ return 0;
+}
+
+late_initcall(honor_cmc_firmware_first);
--
To unsubscribe from this list: send the line "unsubscribe linux-acpi" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
