On Mon, Feb 15, 2016 at 05:46:56PM +0000, Will Deacon wrote:

On Thu, Feb 11, 2016 at 06:40:02PM +0000, Marc Zyngier wrote:

quoted

With VHE, we place kernel {watch,break}-points at EL2 to get things
like kgdb and "perf -e mem:..." working.

This requires a bit of repainting in the low-level encore/decode,
but is otherwise pretty simple.

Signed-off-by: Marc Zyngier <redacted>
---
 arch/arm64/include/asm/hw_breakpoint.h | 49 +++++++++++++++++++++-------------
 1 file changed, 31 insertions(+), 18 deletions(-)

diff --git a/arch/arm64/include/asm/hw_breakpoint.h b/arch/arm64/include/asm/hw_breakpoint.h
index 9732908..4d8d5a8 100644
--- a/arch/arm64/include/asm/hw_breakpoint.h
+++ b/arch/arm64/include/asm/hw_breakpoint.h

@@ -18,6 +18,7 @@
 
 #include <asm/cputype.h>
 #include <asm/cpufeature.h>
+#include <asm/virt.h>
 
 #ifdef __KERNEL__
 

@@ -35,24 +36,6 @@ struct arch_hw_breakpoint {
 	struct arch_hw_breakpoint_ctrl ctrl;
 };
 
-static inline u32 encode_ctrl_reg(struct arch_hw_breakpoint_ctrl ctrl)
-{
-	return (ctrl.len << 5) | (ctrl.type << 3) | (ctrl.privilege << 1) |
-		ctrl.enabled;
-}
-
-static inline void decode_ctrl_reg(u32 reg,
-				   struct arch_hw_breakpoint_ctrl *ctrl)
-{
-	ctrl->enabled	= reg & 0x1;
-	reg >>= 1;
-	ctrl->privilege	= reg & 0x3;
-	reg >>= 2;
-	ctrl->type	= reg & 0x3;
-	reg >>= 2;
-	ctrl->len	= reg & 0xff;
-}
-
 /* Breakpoint */
 #define ARM_BREAKPOINT_EXECUTE	0
 

@@ -62,6 +45,7 @@ static inline void decode_ctrl_reg(u32 reg,
 #define AARCH64_ESR_ACCESS_MASK	(1 << 6)
 
 /* Privilege Levels */
+#define AARCH64_BREAKPOINT_EL2	0
 #define AARCH64_BREAKPOINT_EL1	1
 #define AARCH64_BREAKPOINT_EL0	2
 

@@ -76,6 +60,35 @@ static inline void decode_ctrl_reg(u32 reg,
 #define ARM_KERNEL_STEP_ACTIVE	1
 #define ARM_KERNEL_STEP_SUSPEND	2
 
+#define DBG_HMC_HYP		(1 << 13)
+#define DBG_SSC_HYP		(3 << 14)

Why do we need to touch the SSC field at all?

quoted

+
+static inline u32 encode_ctrl_reg(struct arch_hw_breakpoint_ctrl ctrl)
+{
+	u32 val = (ctrl.len << 5) | (ctrl.type << 3) | ctrl.enabled;
+
+	if (is_kernel_in_hyp_mode() && ctrl.privilege == AARCH64_BREAKPOINT_EL1)
+		val |= DBG_HMC_HYP | DBG_SSC_HYP | (AARCH64_BREAKPOINT_EL2 << 1);

I don't think this is correct. We want to allow, for example, a userspace
watchpoint to fire thanks to something like put_user, so the encoding
really needs to build up the PMC field (like we do already), then orr in
the HMC field.

Hmm, I got my arm and my arm64 mixed up here. For the latter, we don't
actually support EL0+EL1 watchpoints, but I still think that the
{HMC,SSC,PMC} encoding of {1,00,xx} is cleaner.

Will