[PATCH] powerpc/xmon: really enable xmon when a breakpoint is set

When single-stepping kernel code from xmon without a debug hook enabled
the kernel crashes. This can happen when kernel starts with xmon on
crash disabled but xmon is entered using sysrq.

Commit e1368d0c9edb ("powerpc/xmon: Setup debugger hooks when first
break-point is set") adds force_enable_xmon function that prints
"xmon: Enabling debugger hooks" but does not enable them.

Add the call to xmon_init to install the debugger hooks in
force_enable_xmon and also call force_enable_xmon when single-stepping
in xmon.

Fixes: e1368d0c9edb ("powerpc/xmon: Setup debugger hooks when first
break-point is set")

Signed-off-by: Michal Suchanek <redacted>
---
 arch/powerpc/xmon/xmon.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/arch/powerpc/xmon/xmon.c b/arch/powerpc/xmon/xmon.c
index a0842f1ff72c..504bd1c3d8b0 100644
--- a/arch/powerpc/xmon/xmon.c
+++ b/arch/powerpc/xmon/xmon.c

@@ -179,6 +179,9 @@ static const char *getvecname(unsigned long vec);
 
 static int do_spu_cmd(void);
 
+static void xmon_init(int enable);
+static inline void force_enable_xmon(void);
+
 #ifdef CONFIG_44x
 static void dump_tlb_44x(void);
 #endif

@@ -1094,6 +1097,7 @@ static int do_step(struct pt_regs *regs)
 	unsigned int instr;
 	int stepped;
 
+	force_enable_xmon();
 	/* check we are in 64-bit kernel mode, translation enabled */
 	if ((regs->msr & (MSR_64BIT|MSR_PR|MSR_IR)) == (MSR_64BIT|MSR_IR)) {
 		if (mread(regs->nip, &instr, 4) == 4) {

@@ -1275,6 +1279,7 @@ static inline void force_enable_xmon(void)
 	if (!xmon_on) {
 		printf("xmon: Enabling debugger hooks\n");
 		xmon_on = 1;
+		xmon_init(1);
 	}
 }
 

-- 
2.13.6

Thanks for the patch Michal,

Michal Suchanek [off-list ref] writes:

When single-stepping kernel code from xmon without a debug hook enabled
the kernel crashes. This can happen when kernel starts with xmon on
crash disabled but xmon is entered using sysrq.

Commit e1368d0c9edb ("powerpc/xmon: Setup debugger hooks when first
break-point is set") adds force_enable_xmon function that prints
"xmon: Enabling debugger hooks" but does not enable them.

Debugger hooks are enabled just befores debugger() is entered from
sysrq_handle_xmon(). Thats why force_enable_xmon() simply sets sets
'xmon_on=1' and exits.

The problem you are seeing is probably due to sysrq_handle_xmon()
clearing the debugger hooks on return from debugger() as xmon_on was
never set for the 's' xmon command.

Add the call to xmon_init to install the debugger hooks in
force_enable_xmon and also call force_enable_xmon when single-stepping
in xmon.

Only calling force_enable_xmon() from do_step() should be suffice as on exit
from the debugger() the value of xmon_on is checked and if required the
debugger hooks are kept instead of getting cleared.

As mentioned above call to force_enable_xmon() is usually done in context of
sysrq_handle_xmon() which sets the debugger hooks as soon as its
entered. So I think that this hunk is not really needed.


-- 
Vaibhav Jain [off-list ref]
Linux Technology Center, IBM India Pvt. Ltd.

On Tue, 22 May 2018 12:53:53 +0530
Vaibhav Jain [off-list ref] wrote:

Thanks for the patch Michal,

Michal Suchanek [off-list ref] writes:

quoted

When single-stepping kernel code from xmon without a debug hook
enabled the kernel crashes. This can happen when kernel starts with
xmon on crash disabled but xmon is entered using sysrq.

Commit e1368d0c9edb ("powerpc/xmon: Setup debugger hooks when first
break-point is set") adds force_enable_xmon function that prints
"xmon: Enabling debugger hooks" but does not enable them.  

Debugger hooks are enabled just befores debugger() is entered from
sysrq_handle_xmon(). Thats why force_enable_xmon() simply sets sets
'xmon_on=1' and exits.

The problem you are seeing is probably due to sysrq_handle_xmon()
clearing the debugger hooks on return from debugger() as xmon_on was
never set for the 's' xmon command.

Indeed, setting xmon_on is sufficient. Will resend the patch.

Thanks

Michal

quoted

Add the call to xmon_init to install the debugger hooks in
force_enable_xmon and also call force_enable_xmon when
single-stepping in xmon.  

Only calling force_enable_xmon() from do_step() should be suffice as
on exit from the debugger() the value of xmon_on is checked and if
required the debugger hooks are kept instead of getting cleared.

quoted

 arch/powerpc/xmon/xmon.c | 5 +++++
 1 file changed, 5 insertions(+)

diff --git a/arch/powerpc/xmon/xmon.c b/arch/powerpc/xmon/xmon.c
index a0842f1ff72c..504bd1c3d8b0 100644
--- a/arch/powerpc/xmon/xmon.c
+++ b/arch/powerpc/xmon/xmon.c  

quoted

@@ -1275,6 +1279,7 @@ static inline void force_enable_xmon(void)
 	if (!xmon_on) {
 		printf("xmon: Enabling debugger hooks\n");
 		xmon_on = 1;
+		xmon_init(1);
 	}
 }  

As mentioned above call to force_enable_xmon() is usually done in
context of sysrq_handle_xmon() which sets the debugger hooks as soon
as its entered. So I think that this hunk is not really needed.