Hi Rui,
> With your patch, the Intel DG35EC board will not allow my distribution
> reboot or halt the machine. In order to circumvent that problem, I've
> made a few addition to your previous patch witch allows the restore of
> the changed Bit 0 to it's previous value if the module is unloaded.
> My only doubt is if it should be done every time the gbl_smi_en is zero,
> or in conjunction with nowayout when the value also equals zero. This
> patch has what I described and a commented gbl_smi_en only.
Forget the previous patch. I don't like having this hack in the main iTCO_wdt code.
So I added it to the iTCO_vendor_support code with the necessary warnings.
Can you test this (please note that the iTCO_vendor_support module needs to be
loaded with the vendorsupport=911 module parameter).
Thanks in advance,
Wim.
---
commit 8a590d97277819e6693a47ca776ceee9ac74fda3
Author: Wim Van Sebroeck <wim@xxxxxxxxx>
Date: Mon Jun 8 17:41:51 2009 +0000
[WATCHDOG] iTCO_wdt: Fix ICH7+ reboot issue.
Bugzilla: 9868 & 10195.
There seems to be a bug into the SMM code that handles TCO Timeout SMI.
Andriy Gapon found that the code on his DG33TL system does the following:
> The handler is quite simple - it tests value in TCO1_CNT against 0x800, i.e.
> checks TCO_TMR_HLT. If the bit is set the handler goes into an infinite loop,
> apparently to allow the second timeout and reboot. Otherwise it simply clears
> TIMEOUT bit in TCO1_STS and that's it.
> So the logic seems to be reversed, because it is hard to see how TIMEOUT can
> get set to 1 and SMI generated when TCO_TMR_HLT is set (other than a
> transitional effect).
The only trick we have is to bypass the SMM code by turning of the generation
of the SMI#. The trick can only be enabled by setting the vendorsupport module
parameter to 911. This trick doesn't work well on laptop's.
Note: this is a dirty hack. Please handle with care. The only real fix is that
the bug in the SMM bios code get's fixed.
Signed-off-by: Wim Van Sebroeck <wim@xxxxxxxxx>
diff --git a/drivers/watchdog/iTCO_vendor_support.c b/drivers/watchdog/iTCO_vendor_support.c
index 843ef62..5133bca 100644
--- a/drivers/watchdog/iTCO_vendor_support.c
+++ b/drivers/watchdog/iTCO_vendor_support.c
@@ -19,7 +19,7 @@
/* Module and version information */
#define DRV_NAME "iTCO_vendor_support"
-#define DRV_VERSION "1.03"
+#define DRV_VERSION "1.04"
#define PFX DRV_NAME ": "
/* Includes */
@@ -44,11 +44,14 @@
#define SUPERMICRO_OLD_BOARD 1
/* SuperMicro Pentium 4 / Xeon 4 / EMT64T Era Systems */
#define SUPERMICRO_NEW_BOARD 2
+/* Broken BIOS */
+#define BROKEN_BIOS 911
static int vendorsupport;
module_param(vendorsupport, int, 0);
MODULE_PARM_DESC(vendorsupport, "iTCO vendor specific support mode, default="
- "0 (none), 1=SuperMicro Pent3, 2=SuperMicro Pent4+");
+ "0 (none), 1=SuperMicro Pent3, 2=SuperMicro Pent4+, "
+ "911=Broken SMI BIOS");
/*
* Vendor Specific Support
@@ -243,25 +246,92 @@ static void supermicro_new_pre_set_heartbeat(unsigned int heartbeat)
}
/*
+ * Vendor Support: 911
+ * Board: Some Intel ICHx based motherboards
+ * iTCO chipset: ICH7+
+ *
+ * Some Intel motherboards have a broken BIOS implementation: i.e.
+ * the SMI handler clear's the TIMEOUT bit in the TC01_STS register
+ * and does not reload the time. Thus the TCO watchdog does not reboot
+ * the system.
+ *
+ * These are the conclusions of Andriy Gapon <avg@xxxxxxxxxxx> after
+ * debugging: the SMI handler is quite simple - it tests value in
+ * TCO1_CNT against 0x800, i.e. checks TCO_TMR_HLT. If the bit is set
+ * the handler goes into an infinite loop, apparently to allow the
+ * second timeout and reboot. Otherwise it simply clears TIMEOUT bit
+ * in TCO1_STS and that's it.
+ * So the logic seems to be reversed, because it is hard to see how
+ * TIMEOUT can get set to 1 and SMI generated when TCO_TMR_HLT is set
+ * (other than a transitional effect).
+ *
+ * The only fix found to get the motherboard(s) to reboot is to put
+ * the glb_smi_en bit to 0. This is a dirty hack that bypasses the
+ * broken code by disabling Global SMI.
+ *
+ * WARNING: globally disabling SMI could possibly lead to dramatic
+ * problems, especially on laptops! I.e. various ACPI things where
+ * SMI is used for communication between OS and firmware.
+ *
+ * Don't use this fix if you don't need to!!!
+ */
+
+static void broken_bios_start(unsigned long acpibase)
+{
+ unsigned long val32;
+
+ val32 = inl(SMI_EN);
+ /* Bit 13: TCO_EN -> 0 = Disables TCO logic generating an SMI#
+ Bit 0: GBL_SMI_EN -> 0 = No SMI# will be generated by ICH. */
+ val32 &= 0xffffdffe;
+ outl(val32, SMI_EN);
+}
+
+static void broken_bios_stop(unsigned long acpibase)
+{
+ unsigned long val32;
+
+ val32 = inl(SMI_EN);
+ /* Bit 13: TCO_EN -> 1 = Enables TCO logic generating an SMI#
+ Bit 0: GBL_SMI_EN -> 1 = Turn global SMI on again. */
+ val32 |= 0x00002001;
+ outl(val32, SMI_EN);
+}
+
+/*
* Generic Support Functions
*/
void iTCO_vendor_pre_start(unsigned long acpibase,
unsigned int heartbeat)
{
- if (vendorsupport == SUPERMICRO_OLD_BOARD)
+ switch (vendorsupport) {
+ case SUPERMICRO_OLD_BOARD:
supermicro_old_pre_start(acpibase);
- else if (vendorsupport == SUPERMICRO_NEW_BOARD)
+ break;
+ case SUPERMICRO_NEW_BOARD:
supermicro_new_pre_start(heartbeat);
+ break;
+ case BROKEN_BIOS:
+ broken_bios_start(acpibase);
+ break;
+ }
}
EXPORT_SYMBOL(iTCO_vendor_pre_start);
void iTCO_vendor_pre_stop(unsigned long acpibase)
{
- if (vendorsupport == SUPERMICRO_OLD_BOARD)
+ switch (vendorsupport) {
+ case SUPERMICRO_OLD_BOARD:
supermicro_old_pre_stop(acpibase);
- else if (vendorsupport == SUPERMICRO_NEW_BOARD)
+ break;
+ case SUPERMICRO_NEW_BOARD:
supermicro_new_pre_stop();
+ break;
+ case BROKEN_BIOS:
+ broken_bios_stop(acpibase);
+ break;
+ }
}
EXPORT_SYMBOL(iTCO_vendor_pre_stop);
--
To unsubscribe from this list: send the line "unsubscribe kernel-testers" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
