Re: [Patch V1] x86, mce: Don't clear global error reporting banks during cpu_offline

From: Borislav Petkov
Date: Fri Sep 11 2015 - 15:31:14 EST


On Fri, Sep 04, 2015 at 02:29:50PM -0400, Ashok Raj wrote:
> During CPU offline, or during suspend/resume operations, its not safe to
> clear MCi_CTL. These MSR's are either thread scoped (meaning private to
> thread), or core scoped (private to threads in that core only), or socket
> scope i.e visible and controllable from all threads in the socket.
>
> When we turn off during CPU_OFFLINE, just offlining a single CPU will
> stop signaling for all the socket wide resources, such as LLC, iMC for e.g.
>
> It is true for Intel CPU's. But there seems some history that other processors
> may require to turn these off during every CPU offline.
>
> Intel Secure Guard eXtentions (SGX) is worried that it might be possible to
> compromise integrity in a SGX system if the attacker has control of host system
> to inject errors which would be otherwise ignored when MCi_CTL bits are
> cleared. Hence on SGX enabled systems, if MCi_CTL is cleared SGX becomes not
> available anymore.
>
> - Consolidated some code to use sharing
> - Minor changes to some prototypes to fit usage.
> - Left handling same for non-Intel CPU models to avoid any unknown regressions.
> - Fixed review comments from Boris

I massaged the text a bit more and ended up committing this:

---
From: Ashok Raj <ashok.raj@xxxxxxxxx>
Date: Fri, 4 Sep 2015 14:29:50 -0400
Subject: [PATCH] x86/mce: Don't clear shared banks on Intel when offlining
CPUs

It is not safe to clear global MCi_CTL banks during CPU offline or
suspend/resume operations. These MSRs are either thread-scoped (meaning
private to a thread), or core-scoped (private to threads in that core
only), or with a socket scope: visible and controllable from all threads
in the socket.

When we offline a single CPU, clearing those MCi_CTL bits will stop
signaling for all the shared, i.e., socket-wide resources, such as LLC,
iMC, etc.

In addition, it might be possible to compromise the integrity of an
Intel Secure Guard eXtentions (SGX) system if the attacker has control
of the host system and is able to inject errors which would be otherwise
ignored when MCi_CTL bits are cleared.

Hence on SGX enabled systems, if MCi_CTL is cleared, SGX gets disabled.

Reviewed-by: Tony Luck <tony.luck@xxxxxxxxx>
Tested-by: Serge Ayoun <serge.ayoun@xxxxxxxxx>
Signed-off-by: Ashok Raj <ashok.raj@xxxxxxxxx>
Cc: "H. Peter Anvin" <hpa@xxxxxxxxx>
Cc: Ingo Molnar <mingo@xxxxxxxxxx>
Cc: Serge Ayoun <serge.ayoun@xxxxxxxxx>
Cc: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
Cc: Tony Luck <tony.luck@xxxxxxxxx>
Cc: linux-edac <linux-edac@xxxxxxxxxxxxxxx>
Cc: x86-ml <x86@xxxxxxxxxx>
Link: http://lkml.kernel.org/r/1441391390-16985-1-git-send-email-ashok.raj@xxxxxxxxx
[ Cleanup text. ]
Signed-off-by: Borislav Petkov <bp@xxxxxxx>
---
arch/x86/kernel/cpu/mcheck/mce.c | 30 ++++++++++++++++++++----------
1 file changed, 20 insertions(+), 10 deletions(-)

diff --git a/arch/x86/kernel/cpu/mcheck/mce.c b/arch/x86/kernel/cpu/mcheck/mce.c
index 9d014b82a124..17b5ec6edcb6 100644
--- a/arch/x86/kernel/cpu/mcheck/mce.c
+++ b/arch/x86/kernel/cpu/mcheck/mce.c
@@ -2042,7 +2042,7 @@ int __init mcheck_init(void)
* Disable machine checks on suspend and shutdown. We can't really handle
* them later.
*/
-static int mce_disable_error_reporting(void)
+static void mce_disable_error_reporting(void)
{
int i;

@@ -2052,17 +2052,32 @@ static int mce_disable_error_reporting(void)
if (b->init)
wrmsrl(MSR_IA32_MCx_CTL(i), 0);
}
- return 0;
+ return;
+}
+
+static void vendor_disable_error_reporting(void)
+{
+ /*
+ * Don't clear on Intel CPUs. Some of these MSRs are socket-wide.
+ * Disabling them for just a single offlined CPU is bad, since it will
+ * inhibit reporting for all shared resources on the socket like the
+ * last level cache (LLC), the integrated memory controller (iMC), etc.
+ */
+ if (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL)
+ return;
+
+ mce_disable_error_reporting();
}

static int mce_syscore_suspend(void)
{
- return mce_disable_error_reporting();
+ vendor_disable_error_reporting();
+ return 0;
}

static void mce_syscore_shutdown(void)
{
- mce_disable_error_reporting();
+ vendor_disable_error_reporting();
}

/*
@@ -2342,19 +2357,14 @@ static void mce_device_remove(unsigned int cpu)
static void mce_disable_cpu(void *h)
{
unsigned long action = *(unsigned long *)h;
- int i;

if (!mce_available(raw_cpu_ptr(&cpu_info)))
return;

if (!(action & CPU_TASKS_FROZEN))
cmci_clear();
- for (i = 0; i < mca_cfg.banks; i++) {
- struct mce_bank *b = &mce_banks[i];

- if (b->init)
- wrmsrl(MSR_IA32_MCx_CTL(i), 0);
- }
+ vendor_disable_error_reporting();
}

static void mce_reenable_cpu(void *h)
--
2.3.5


--
Regards/Gruss,
Boris.

ECO tip #101: Trim your mails when you reply.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/