ARM/hw_breakpoint: add ARMv8.1/ARMv8.2 debug architecutre versions support in enable_monitor_mode()
Subsystems (2)
PATCH RESEND 1/1 ARM/hw_breakpoint: add ARMv8.1/ARMv8.2 debug architecutre versions support in enable_monitor_mode()
From: Candle Sun <[email protected]>
When ARMv8.1/ARMv8.2 cores are used in AArch32 mode, arch_hw_breakpoint_init() in arch/arm/kernel/hw_breakpoint.c will be used.
From ARMv8 specification, different debug architecture versions defined:
* 0110 ARMv8, v8 Debug architecture.
* 0111 ARMv8.1, v8 Debug architecture, with Virtualization Host Extensions.
* 1000 ARMv8.2, v8.2 Debug architecture.
So missing ARMv8.1/ARMv8.2 cases will cause enable_monitor_mode() function returns -ENODEV, and arch_hw_breakpoint_init() will fail.
--- arch/arm/include/asm/hw_breakpoint.h | 2 ++ arch/arm/kernel/hw_breakpoint.c | 2 ++ 2 files changed, 4 insertions(+)
arch/arm/include/asm/hw_breakpoint.h +2 -0
@@ -53,6 +53,8 @@ static inline void decode_ctrl_reg(u32 reg,#define ARM_DEBUG_ARCH_V7_MM 4#define ARM_DEBUG_ARCH_V7_1 5#define ARM_DEBUG_ARCH_V8 6 +#define ARM_DEBUG_ARCH_V8_1 7 +#define ARM_DEBUG_ARCH_V8_2 8/* Breakpoint */ #define ARM_BREAKPOINT_EXECUTE 0
arch/arm/kernel/hw_breakpoint.c +2 -0
@@ -246,6 +246,8 @@ static int enable_monitor_mode(void)case ARM_DEBUG_ARCH_V7_ECP14 :case ARM_DEBUG_ARCH_V7_1 :case ARM_DEBUG_ARCH_V8 : +case ARM_DEBUG_ARCH_V8_1 : +case ARM_DEBUG_ARCH_V8_2 :ARM_DBG_WRITE (c0, c2,2 , (dscr |ARM_DSCR_MDBGEN ));isb ();break ;
Signature · 1 line
2.7.4
Will Deacon
maintainerQuoted 13 lines · Candle Sun
From: Candle Sun <[email protected]>
When ARMv8.1/ARMv8.2 cores are used in AArch32 mode,
arch_hw_breakpoint_init()inarch/arm/kernel/hw_breakpoint.cwill be used.From ARMv8 specification, different debug architecture versions defined:
* 0110 ARMv8, v8 Debug architecture.
* 0111 ARMv8.1, v8 Debug architecture, with Virtualization Host Extensions.
* 1000 ARMv8.2, v8.2 Debug architecture.So missing ARMv8.1/ARMv8.2 cases will cause
enable_monitor_mode()function returns-ENODEV, andarch_hw_breakpoint_init()will fail.
Quoted 6 lines
Signed-off-by Candle Sun <[email protected]>Signed-off-by Nianfu Bai <[email protected]>--- arch/arm/include/asm/hw_breakpoint.h | 2 ++ arch/arm/kernel/hw_breakpoint.c | 2 ++ 2 files changed, 4 insertions(+)
How did you test this?
Will
Hi Will,
Sorry for not instant respond.
Quoted 22 lines · Will Deacon
Quoted 13 lines · Candle Sun
From: Candle Sun <[email protected]>
When ARMv8.1/ARMv8.2 cores are used in AArch32 mode,
arch_hw_breakpoint_init()inarch/arm/kernel/hw_breakpoint.cwill be used.From ARMv8 specification, different debug architecture versions defined:
* 0110 ARMv8, v8 Debug architecture.
* 0111 ARMv8.1, v8 Debug architecture, with Virtualization Host Extensions.
* 1000 ARMv8.2, v8.2 Debug architecture.So missing ARMv8.1/ARMv8.2 cases will cause
enable_monitor_mode()function returns-ENODEV, andarch_hw_breakpoint_init()will fail.Quoted 6 lines
Signed-off-by Candle Sun <[email protected]>Signed-off-by Nianfu Bai <[email protected]>--- arch/arm/include/asm/hw_breakpoint.h | 2 ++ arch/arm/kernel/hw_breakpoint.c | 2 ++ 2 files changed, 4 insertions(+)
Quoted 3 lines
How did you test this?
Will
We have the SoC with A55 cores. On one Android project, for saving memory usage, we let A55 run in aarch32 mode.
While the following failures occue on Android CtsBionicTestCases:
--sys_ptrace#watchpoint_imprecisede --sys_ptrace#hardware_breakpoint
--sys_ptrace#watchpoint_stress
The code snippet for testing is:
static void check_hw_feature_supported(pid_t child, HwFeature feature) { #if defined(__arm__)
long capabilities;
long result = ptrace(PTRACE_GETHBPREGS, child, 0, &capabilities); if (result == -1) {
EXPECT_EQ(EIO, errno);
GTEST_SKIP() << "Hardware debug support disabled at kernel
configuration time";
}
uint8_t hb_count = capabilities & 0xff;
capabilities >>= 8;
uint8_t wp_count = capabilities & 0xff;
capabilities >>= 8;
uint8_t max_wp_size = capabilities & 0xff;
if (max_wp_size == 0) {
GTEST_SKIP() << "Kernel reports zero maximum watchpoint size";
} else if (feature == HwFeature::Watchpoint && wp_count == 0) {
GTEST_SKIP() << "Kernel reports zero hardware watchpoints";
} else if (feature == HwFeature::Breakpoint && hb_count == 0) {
GTEST_SKIP() << "Kernel reports zero hardware breakpoints"; }
#elif defined(__aarch64__)
user_hwdebug_state dreg_state;
iovec iov;
iov.iov_base = &dreg_state;
iov.iov_len = sizeof(dreg_state);
long result = ptrace(PTRACE_GETREGSET, child,
feature == HwFeature::Watchpoint ?
NT_ARM_HW_WATCH : NT_ARM_HW_BREAK, &iov);
if (result == -1) {
ASSERT_EQ(EINVAL, errno); }
if ((dreg_state.dbg_info & 0xff) == 0) GTEST_SKIP() << "hardware
support missing";
#else
// We assume watchpoints and breakpoints are always supported on x86. UNUSED(child);
UNUSED(feature);
#endif
}
The max_wp_size field returned by __ptrace() from kernel is zero, which causes the test failures.
After futher analysis, we found max_watchpoint_len variable is not right initialized in kernelarch_hw_breakpoint_init() function. Missing the case of ARM_DEBUG_ARCH_V8_2 in enable_monitor_mode() directly aborts the arch_hw_breakpoint_int().
Candle
Best regards
Will,
Is the patch useful for you? Would you please give me some suggestions? Thank you.
Regards,
Candle
Quoted 93 lines · Candle Sun
Hi Will,
Sorry for not instant respond.Quoted 22 lines · Will Deacon
Quoted 13 lines · Candle Sun
From: Candle Sun <[email protected]>
When ARMv8.1/ARMv8.2 cores are used in AArch32 mode,
arch_hw_breakpoint_init()inarch/arm/kernel/hw_breakpoint.cwill be used.From ARMv8 specification, different debug architecture versions defined:
* 0110 ARMv8, v8 Debug architecture.
* 0111 ARMv8.1, v8 Debug architecture, with Virtualization Host Extensions.
* 1000 ARMv8.2, v8.2 Debug architecture.So missing ARMv8.1/ARMv8.2 cases will cause
enable_monitor_mode()function returns-ENODEV, andarch_hw_breakpoint_init()will fail.Quoted 6 lines
Signed-off-by Candle Sun <[email protected]>Signed-off-by Nianfu Bai <[email protected]>--- arch/arm/include/asm/hw_breakpoint.h | 2 ++ arch/arm/kernel/hw_breakpoint.c | 2 ++ 2 files changed, 4 insertions(+)Quoted 3 lines
How did you test this?
Will
We have the SoC with A55 cores. On one Android project, for saving memory usage, we let A55 run in aarch32 mode.
While the following failures occue on Android CtsBionicTestCases:
--sys_ptrace#watchpoint_imprecisede --sys_ptrace#hardware_breakpoint
--sys_ptrace#watchpoint_stressThe code snippet for testing is:
static void
check_hw_feature_supported(pid_t child, HwFeature feature){ #ifdefined(__arm__)
long capabilities;
long result =ptrace(PTRACE_GETHBPREGS, child, 0, &capabilities); if (result == -1) {
EXPECT_EQ(EIO, errno);
GTEST_SKIP()<< "Hardware debug support disabled at kernel
configuration time";
}
uint8_thb_count= capabilities &0xff;
capabilities >>= 8;
uint8_twp_count= capabilities &0xff;
capabilities >>= 8;
uint8_tmax_wp_size= capabilities &0xff;if (max_wp_size == 0) { GTEST_SKIP() << "Kernel reports zero maximum watchpoint size"; } else if (feature == HwFeature::Watchpoint && wp_count == 0) { GTEST_SKIP() << "Kernel reports zero hardware watchpoints"; } else if (feature == HwFeature::Breakpoint && hb_count == 0) { GTEST_SKIP() << "Kernel reports zero hardware breakpoints";}
#elifdefined(__aarch64__)
user_hwdebug_statedreg_state;
iovec iov;
iov.iov_base =&dreg_state;
iov.iov_len =sizeof(dreg_state);long result = ptrace(PTRACE_GETREGSET, child,
feature == HwFeature::Watchpoint ? NT_ARM_HW_WATCH : NT_ARM_HW_BREAK, &iov); if (result == -1) { ASSERT_EQ(EINVAL, errno);}
if ((dreg_state.dbg_info &0xff) == 0)GTEST_SKIP()<< "hardware
support missing";
#else
// We assume watchpoints and breakpoints are always supported on x86. UNUSED(child);
UNUSED(feature);
#endif
}The
max_wp_sizefield returned by__ptrace()from kernel is zero, which causes the test failures.After futher analysis, we found
max_watchpoint_lenvariable is not right initialized in kernelarch_hw_breakpoint_init()function. Missing the case ofARM_DEBUG_ARCH_V8_2inenable_monitor_mode()directly aborts thearch_hw_breakpoint_int().Candle
Best regards
Hello,
just noticed a typo in the subject line while going through my lakml mailbox:
s/architecutre/architecture/
Best regards
Uwe
Signature · 2 lines
Pengutronix e.K. | Uwe Kleine-König | Industrial Linux Solutions | http://www.pengutronix.de/ |
Thanks Uwe for pointing out my typing error.
Will,
Is the patch ok? Do I need to send another version?
Candle
Candle
Quoted 14 lines · Uwe Kleine-König
Hello,
just noticed a typo in the subject line while going through my lakml mailbox:
s/architecutre/architecture/
Best regards
Uwe--
Pengutronix e.K. | Uwe Kleine-König |
Industrial Linux Solutions | http://www.pengutronix.de/ |
Quoted 19 lines · Candle Sun
From: Candle Sun <[email protected]>
When ARMv8.1/ARMv8.2 cores are used in AArch32 mode,
arch_hw_breakpoint_init()inarch/arm/kernel/hw_breakpoint.cwill be used.From ARMv8 specification, different debug architecture versions defined:
* 0110 ARMv8, v8 Debug architecture.
* 0111 ARMv8.1, v8 Debug architecture, with Virtualization Host Extensions.
* 1000 ARMv8.2, v8.2 Debug architecture.So missing ARMv8.1/ARMv8.2 cases will cause
enable_monitor_mode()function returns-ENODEV, andarch_hw_breakpoint_init()will fail.Signed-off-by Candle Sun <[email protected]>Signed-off-by Nianfu Bai <[email protected]>--- arch/arm/include/asm/hw_breakpoint.h | 2 ++ arch/arm/kernel/hw_breakpoint.c | 2 ++ 2 files changed, 4 insertions(+)
[...]
arch/arm/include/asm/hw_breakpoint.h static inline void decode_ctrl_reg(u32 reg, · line 57
Quoted 10 lines
arch/arm/include/asm/hw_breakpoint.h +2
-0@@ -53,6 +53,8 @@ static inline void decode_ctrl_reg(u32 reg,#define ARM_DEBUG_ARCH_V7_MM 4#define ARM_DEBUG_ARCH_V7_1 5#define ARM_DEBUG_ARCH_V8 6 +#define ARM_DEBUG_ARCH_V8_1 7 +#define ARM_DEBUG_ARCH_V8_2 8
Looks like you can also add:
#define ARM_DEBUG_ARCH_V8_4 9
and treat that the same way. With that, and a fix to $SUBJECT:
Please put this into the patch system [1].
Will
Quoted 41 lines · Will Deacon
Quoted 19 lines · Candle Sun
From: Candle Sun <[email protected]>
When ARMv8.1/ARMv8.2 cores are used in AArch32 mode,
arch_hw_breakpoint_init()inarch/arm/kernel/hw_breakpoint.cwill be used.From ARMv8 specification, different debug architecture versions defined:
* 0110 ARMv8, v8 Debug architecture.
* 0111 ARMv8.1, v8 Debug architecture, with Virtualization Host Extensions.
* 1000 ARMv8.2, v8.2 Debug architecture.So missing ARMv8.1/ARMv8.2 cases will cause
enable_monitor_mode()function returns-ENODEV, andarch_hw_breakpoint_init()will fail.Signed-off-by Candle Sun <[email protected]>Signed-off-by Nianfu Bai <[email protected]>--- arch/arm/include/asm/hw_breakpoint.h | 2 ++ arch/arm/kernel/hw_breakpoint.c | 2 ++ 2 files changed, 4 insertions(+)[...]
arch/arm/include/asm/hw_breakpoint.h static inline void decode_ctrl_reg(u32 reg, · line 57
Quoted 10 lines
arch/arm/include/asm/hw_breakpoint.h +2
-0@@ -53,6 +53,8 @@ static inline void decode_ctrl_reg(u32 reg,#define ARM_DEBUG_ARCH_V7_MM 4#define ARM_DEBUG_ARCH_V7_1 5#define ARM_DEBUG_ARCH_V8 6 +#define ARM_DEBUG_ARCH_V8_1 7 +#define ARM_DEBUG_ARCH_V8_2 8Looks like you can also add:
#define
ARM_DEBUG_ARCH_V8_49and treat that the same way. With that, and a fix to $SUBJECT:
Quoted 7 lines
Acked-by Will Deacon <[email protected]>Please put this into the patch system [1].
Will
Thanks, Will.
I will do it ASAP.
Best regards,
Candle