* [PATCH linux-next 1/2] kgdb-x86: allow to temporarily disable trap activation
2013-11-19 14:53 Dan Aloni
@ 2013-11-19 14:53 ` Dan Aloni
2013-11-19 14:53 ` [PATCH linux-next 2/2] kvm-x86: emulator: disable kgdb-x86 on fastop and fpe flush Dan Aloni
2013-11-27 10:08 ` KVM/KGDB integration Gleb Natapov
2 siblings, 0 replies; 5+ messages in thread
From: Dan Aloni @ 2013-11-19 14:53 UTC (permalink / raw)
To: linux-kernel
Cc: kvm, kgdb-bugreport, x86, gleb, pbonzini, tglx, mingo, hpa,
Muli Ben-Yehuda
There are some users that would like the ability of a temporary
per-cpu deactivation of the debugger trap.
More specifically, we have seen that when kvm-unit-tests runs with
kgdb enabled, the kernel halts on the debugger during the KVM x86
instruction emulation that performs a test of a division by zero,
and also during fwait.
A second patch depending on this one addresses that issue.
Signed-off-by: Dan Aloni <alonid@stratoscale.com>
Signed-off-by: Muli Ben-Yehuda <muli@stratoscale.com>
---
arch/x86/include/asm/kgdb.h | 8 ++++++++
arch/x86/kernel/kgdb.c | 21 +++++++++++++++++++++
2 files changed, 29 insertions(+)
diff --git a/arch/x86/include/asm/kgdb.h b/arch/x86/include/asm/kgdb.h
index 332f98c..35bec81 100644
--- a/arch/x86/include/asm/kgdb.h
+++ b/arch/x86/include/asm/kgdb.h
@@ -86,4 +86,12 @@ static inline void arch_kgdb_breakpoint(void)
extern int kgdb_ll_trap(int cmd, const char *str,
struct pt_regs *regs, long err, int trap, int sig);
+#ifdef CONFIG_KGDB
+extern void kgdb_ll_local_disable(void);
+extern void kgdb_ll_local_enable(void);
+#else
+#define kgdb_ll_local_disable() do {} while (0)
+#define kgdb_ll_local_enable() do {} while (0)
+#endif
+
#endif /* _ASM_X86_KGDB_H */
diff --git a/arch/x86/kernel/kgdb.c b/arch/x86/kernel/kgdb.c
index 836f832..c269ad8 100644
--- a/arch/x86/kernel/kgdb.c
+++ b/arch/x86/kernel/kgdb.c
@@ -587,6 +587,22 @@ int kgdb_ll_trap(int cmd, const char *str,
return __kgdb_notify(&args, cmd);
}
+static DEFINE_PER_CPU(bool, kgdb_swallow_traps);
+
+void kgdb_ll_local_disable(void)
+{
+ __this_cpu_write(kgdb_swallow_traps, true);
+}
+
+EXPORT_SYMBOL(kgdb_ll_local_disable);
+
+void kgdb_ll_local_enable(void)
+{
+ __this_cpu_write(kgdb_swallow_traps, false);
+}
+
+EXPORT_SYMBOL(kgdb_ll_local_enable);
+
static int
kgdb_notify(struct notifier_block *self, unsigned long cmd, void *ptr)
{
@@ -594,6 +610,11 @@ kgdb_notify(struct notifier_block *self, unsigned long cmd, void *ptr)
int ret;
local_irq_save(flags);
+ if (__this_cpu_read(kgdb_swallow_traps) && cmd == DIE_TRAP) {
+ local_irq_restore(flags);
+ return NOTIFY_DONE;
+ }
+
ret = __kgdb_notify(ptr, cmd);
local_irq_restore(flags);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 5+ messages in thread* [PATCH linux-next 2/2] kvm-x86: emulator: disable kgdb-x86 on fastop and fpe flush
2013-11-19 14:53 Dan Aloni
2013-11-19 14:53 ` [PATCH linux-next 1/2] kgdb-x86: allow to temporarily disable trap activation Dan Aloni
@ 2013-11-19 14:53 ` Dan Aloni
2013-11-19 15:02 ` Paolo Bonzini
2013-11-27 10:08 ` KVM/KGDB integration Gleb Natapov
2 siblings, 1 reply; 5+ messages in thread
From: Dan Aloni @ 2013-11-19 14:53 UTC (permalink / raw)
To: linux-kernel
Cc: kvm, kgdb-bugreport, x86, gleb, pbonzini, tglx, mingo, hpa,
Muli Ben-Yehuda
We have seen that when kvm-unit-tests runs with kgdb enabled, the
kernel halts on the debugger during the KVM x86 instruction emulation
that performs a test of a division by zero, and also during fwait.
This patch adds calls that temporarily disable the debugger trap.
Signed-off-by: Dan Aloni <alonid@stratoscale.com>
Signed-off-by: Muli Ben-Yehuda <muli@stratoscale.com>
---
arch/x86/kvm/emulate.c | 7 +++++++
1 file changed, 7 insertions(+)
diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
index 07ffca0..05e8509 100644
--- a/arch/x86/kvm/emulate.c
+++ b/arch/x86/kvm/emulate.c
@@ -24,6 +24,7 @@
#include "kvm_cache_regs.h"
#include <linux/module.h>
#include <asm/kvm_emulate.h>
+#include <asm/kgdb.h>
#include <linux/stringify.h>
#include "x86.h"
@@ -4438,6 +4439,7 @@ static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
{
bool fault = false;
+ kgdb_ll_local_disable();
ctxt->ops->get_fpu(ctxt);
asm volatile("1: fwait \n\t"
"2: \n\t"
@@ -4449,6 +4451,7 @@ static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
_ASM_EXTABLE(1b, 3b)
: [fault]"+qm"(fault));
ctxt->ops->put_fpu(ctxt);
+ kgdb_ll_local_enable();
if (unlikely(fault))
return emulate_exception(ctxt, MF_VECTOR, 0, false);
@@ -4468,10 +4471,14 @@ static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *))
ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
if (!(ctxt->d & ByteOp))
fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
+
+ kgdb_ll_local_disable();
asm("push %[flags]; popf; call *%[fastop]; pushf; pop %[flags]\n"
: "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
[fastop]"+S"(fop)
: "c"(ctxt->src2.val));
+ kgdb_ll_local_enable();
+
ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
if (!fop) /* exception is returned in fop variable */
return emulate_de(ctxt);
--
1.8.3.1
^ permalink raw reply related [flat|nested] 5+ messages in thread* Re: [PATCH linux-next 2/2] kvm-x86: emulator: disable kgdb-x86 on fastop and fpe flush
2013-11-19 14:53 ` [PATCH linux-next 2/2] kvm-x86: emulator: disable kgdb-x86 on fastop and fpe flush Dan Aloni
@ 2013-11-19 15:02 ` Paolo Bonzini
0 siblings, 0 replies; 5+ messages in thread
From: Paolo Bonzini @ 2013-11-19 15:02 UTC (permalink / raw)
To: Dan Aloni
Cc: linux-kernel, kvm, kgdb-bugreport, x86, gleb, tglx, mingo, hpa,
Muli Ben-Yehuda
Il 19/11/2013 15:53, Dan Aloni ha scritto:
> We have seen that when kvm-unit-tests runs with kgdb enabled, the
> kernel halts on the debugger during the KVM x86 instruction emulation
> that performs a test of a division by zero, and also during fwait.
>
> This patch adds calls that temporarily disable the debugger trap.
>
> Signed-off-by: Dan Aloni <alonid@stratoscale.com>
> Signed-off-by: Muli Ben-Yehuda <muli@stratoscale.com>
> ---
> arch/x86/kvm/emulate.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/arch/x86/kvm/emulate.c b/arch/x86/kvm/emulate.c
> index 07ffca0..05e8509 100644
> --- a/arch/x86/kvm/emulate.c
> +++ b/arch/x86/kvm/emulate.c
> @@ -24,6 +24,7 @@
> #include "kvm_cache_regs.h"
> #include <linux/module.h>
> #include <asm/kvm_emulate.h>
> +#include <asm/kgdb.h>
> #include <linux/stringify.h>
>
> #include "x86.h"
> @@ -4438,6 +4439,7 @@ static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
> {
> bool fault = false;
>
> + kgdb_ll_local_disable();
> ctxt->ops->get_fpu(ctxt);
> asm volatile("1: fwait \n\t"
> "2: \n\t"
> @@ -4449,6 +4451,7 @@ static int flush_pending_x87_faults(struct x86_emulate_ctxt *ctxt)
> _ASM_EXTABLE(1b, 3b)
> : [fault]"+qm"(fault));
> ctxt->ops->put_fpu(ctxt);
> + kgdb_ll_local_enable();
>
> if (unlikely(fault))
> return emulate_exception(ctxt, MF_VECTOR, 0, false);
> @@ -4468,10 +4471,14 @@ static int fastop(struct x86_emulate_ctxt *ctxt, void (*fop)(struct fastop *))
> ulong flags = (ctxt->eflags & EFLAGS_MASK) | X86_EFLAGS_IF;
> if (!(ctxt->d & ByteOp))
> fop += __ffs(ctxt->dst.bytes) * FASTOP_SIZE;
> +
> + kgdb_ll_local_disable();
> asm("push %[flags]; popf; call *%[fastop]; pushf; pop %[flags]\n"
> : "+a"(ctxt->dst.val), "+d"(ctxt->src.val), [flags]"+D"(flags),
> [fastop]"+S"(fop)
> : "c"(ctxt->src2.val));
> + kgdb_ll_local_enable();
> +
> ctxt->eflags = (ctxt->eflags & ~EFLAGS_MASK) | (flags & EFLAGS_MASK);
> if (!fop) /* exception is returned in fop variable */
> return emulate_de(ctxt);
>
Acked-by: Paolo Bonzini <pbonzini@redhat.com>
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: KVM/KGDB integration
2013-11-19 14:53 Dan Aloni
2013-11-19 14:53 ` [PATCH linux-next 1/2] kgdb-x86: allow to temporarily disable trap activation Dan Aloni
2013-11-19 14:53 ` [PATCH linux-next 2/2] kvm-x86: emulator: disable kgdb-x86 on fastop and fpe flush Dan Aloni
@ 2013-11-27 10:08 ` Gleb Natapov
2 siblings, 0 replies; 5+ messages in thread
From: Gleb Natapov @ 2013-11-27 10:08 UTC (permalink / raw)
To: Dan Aloni
Cc: linux-kernel, kvm, kgdb-bugreport, x86, pbonzini, tglx, mingo,
hpa, Jason Wessel
Copying KGDB maintainer to get some feedback.
On Tue, Nov 19, 2013 at 04:53:28PM +0200, Dan Aloni wrote:
> Hello,
>
> The following two patches address an integration issue between KVM and
> KGDB. The issue described in the patches can be triggered with vanilla
> kernels that enable KGDB and KVM together on x86 (more specifically,
> we bump into this with Fedora's 3.11 kernel from FC19).
>
> On a kernel enabled with KGDB, running with kvm-unit-tests should
> reproduce the issue. On VM hosts servers where an admin accidently
> left an active KGDB, and unprivilged guest might be able to bring
> the host down.
>
> Patches apply to linux-next and earlier kernels.
--
Gleb.
^ permalink raw reply [flat|nested] 5+ messages in thread