Skip to contentSkip to conversation

ARM: hw_breakpoint: fixes and improvements (v3)

PATCH New Will Deacon 2 participants 14 messages Dec 2, 2010

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover ARM: hw_breakpoint: fixes and improvements (v3)

Hello again,

This is version 3 of the patchset originally posted here:

[v1]: http://lists.infradead.org/pipermail/linux-arm-kernel/2010-November/032717.html
[v2]: http://lists.infradead.org/pipermail/linux-arm-kernel/2010-November/032962.html

It's grown a bit since v2, but I think I've caught all of the corner cases now. Thanks to the people who have highlighted any problems.

Changes since v2 include:

* Unified single-stepping mechanism for watchpoints and breakpoints

* Corrected misdiagnosis of the problem fixed in patch 6 (`do not allocate...')

* For cores where monitor mode cannot be enabled due to hardware restrictions, we no longer fail the initcall or WARN_ONCE.

* Per-cpu breakpoints are disallowed for the time being because they cannot be stepped reliably (since we may take an interrupt before the step has completed).

* Updated the comment describing why only a single watchpoint is available.

As per usual, all feedback is welcome.

Will Deacon (11):
ARM: hw_breakpoint: ensure OS lock is clear before writing to debug
registers
ARM: hw_breakpoint: reset control registers in hotplug path
ARM: hw_breakpoint: correct and simplify alignment fixup code
ARM: hw_breakpoint: disable preemption during debug exception
handling
ARM: hw_breakpoint: don't advertise reserved breakpoints
ARM: hw_breakpoint: do not allocate new breakpoints with preemption
disabled
ARM: hw_breakpoint: unify single-stepping code for watchpoints and
breakpoints
ARM: hw_breakpoint: disallow per-cpu breakpoints without overflow
handler
ARM: ptrace: fix style issue with hw_breakpoint interface
ARM: hw_breakpoint: fix warnings generated by sparse
ARM: hw_breakpoint: do not fail initcall if monitor mode is disabled

 arch/arm/include/asm/hw_breakpoint.h |    4 +-
 arch/arm/kernel/entry-armv.S         |    4 +
 arch/arm/kernel/entry-header.S       |   19 ++
 arch/arm/kernel/hw_breakpoint.c      |  545 ++++++++++++++++++++--------------
 arch/arm/kernel/ptrace.c             |    4 +-
 5 files changed, 349 insertions(+), 227 deletions(-)
#1 lore raw

PATCH 1/11 ARM: hw_breakpoint: ensure OS lock is clear before writing to debug registers

ARMv7 architects a system for saving and restoring the debug registers across low-power modes. At the heart of this system is a lock register which, when set, forbids writes to the debug registers. While locked, writes to debug registers via the co-processor interface will result in undefined instruction traps. Linux currently doesn't make use of this feature because we update the debug registers on context switch anyway, however the status of the lock is IMPLEMENTATION DEFINED on reset.

This patch ensures that the lock is cleared during boot so that we can write to the debug registers safely.

---
 arch/arm/kernel/hw_breakpoint.c |   27 ++++++++++++++++++++++-----
 1 files changed, 22 insertions(+), 5 deletions(-)
arch/arm/kernel/hw_breakpoint.c +22 -5
index 21e3a4a..793959e 100644
@@ -768,6 +768,23 @@ static void __init reset_ctrl_regs(void *unused)
 {
 	int i;
 
+	/*
+	 * v7 debug contains save and restore registers so that debug state
+	 * can be maintained across low-power modes without leaving
+	 * the debug logic powered up. It is IMPLEMENTATION DEFINED whether
+	 * we can write to the debug registers out of reset, so we must
+	 * unlock the OS Lock Access Register to avoid taking undefined
+	 * instruction exceptions later on.
+	 */
+	if (debug_arch >= ARM_DEBUG_ARCH_V7_ECP14) {
+		/*
+		 * Unconditionally clear the lock by writing a value
+		 * other than 0xC5ACCE55 to the access register.
+		 */
+		asm volatile("mcr p14, 0, %0, c1, c0, 4" : : "r" (0));
+		isb();
+	}
+
 	if (enable_monitor_mode())
 		return;
 
@@ -810,17 +827,17 @@ static int __init arch_hw_breakpoint_init(void)
 		pr_warning("halting debug mode enabled. Assuming maximum "
 				"watchpoint size of 4 bytes.");
 	} else {
-		/* Work out the maximum supported watchpoint length. */
-		max_watchpoint_len = get_max_wp_len();
-		pr_info("maximum watchpoint size is %u bytes.\n",
-				max_watchpoint_len);
-
 		/*
 		 * Reset the breakpoint resources. We assume that a halting
 		 * debugger will leave the world in a nice state for us.
 		 */
 		smp_call_function(reset_ctrl_regs, NULL, 1);
 		reset_ctrl_regs(NULL);
+
+		/* Work out the maximum supported watchpoint length. */
+		max_watchpoint_len = get_max_wp_len();
+		pr_info("maximum watchpoint size is %u bytes.\n",
+				max_watchpoint_len);
 	}
 
 	/* Register debug fault handler. */
Signature · 1 line
1.7.0.4
#2 lore raw

PATCH 2/11 ARM: hw_breakpoint: reset control registers in hotplug path

In reply to Will Deacon

The ARMv7 debug architecture doesn't make any guarantees about the contents of debug control registers following a debug logic reset.

This patch ensures that we reset the control registers when a cpu comes ONLINE (for example, with hotplug) so that when we enable monitor mode while inserting a breakpoint we won't exhibit random behaviour.

---
 arch/arm/kernel/hw_breakpoint.c |   16 +++++++++++++++-
 1 files changed, 15 insertions(+), 1 deletions(-)
arch/arm/kernel/hw_breakpoint.c +15 -1
index 793959e..515a3c4 100644
@@ -764,7 +764,7 @@ out:
 /*
  * One-time initialisation.
  */
-static void __init reset_ctrl_regs(void *unused)
+static void reset_ctrl_regs(void *unused)
 {
 	int i;
 
@@ -799,6 +799,18 @@ static void __init reset_ctrl_regs(void *unused)
 	}
 }
 
+static int __cpuinit dbg_reset_notify(struct notifier_block *self,
+				      unsigned long action, void *cpu)
+{
+	if (action == CPU_ONLINE)
+		smp_call_function_single((int)cpu, reset_ctrl_regs, NULL, 1);
+	return NOTIFY_OK;
+}
+
+static struct notifier_block __cpuinitdata dbg_reset_nb = {
+	.notifier_call = dbg_reset_notify,
+};
+
 static int __init arch_hw_breakpoint_init(void)
 {
 	int ret = 0;
@@ -846,6 +858,8 @@ static int __init arch_hw_breakpoint_init(void)
 	hook_ifault_code(2, hw_breakpoint_pending, SIGTRAP, TRAP_HWBKPT,
 			"breakpoint debug exception");
 
+	/* Register hotplug notifier. */
+	register_cpu_notifier(&dbg_reset_nb);
 out:
 	return ret;
 }
Signature · 1 line
1.7.0.4
#3 lore raw

PATCH 3/11 ARM: hw_breakpoint: correct and simplify alignment fixup code

In reply to Will Deacon

The current hw_breakpoint code tries to fix up the alignment of breakpoints so that we can make use of sparse byte-address-select bits in the control register and give the illusion that we can set breakpoints on unaligned addresses.

Although this works on v6 cores, v7 forbids this behaviour, instead requiring breakpoints to be set on aligned addresses and have contiguous byte-address-select ranges depending on the instruction set in use. For ARM the only supported size is 4 bytes, whilst Thumb-2 also permits 2 byte breakpoints (watchpoints can be of 1, 2, 4 or 8 bytes long).

This patch simplifies the alignment fixup code so that we require addresses to be aligned to the size of the corresponding breakpoint. This allows us to handle the common case of breaking on a half-word aligned Thumb-2 instruction and also allows us to set byte watchpoints on arbitrary addresses.

---
 arch/arm/kernel/hw_breakpoint.c |   57 +++++++++++++++++++++-----------------
 1 files changed, 31 insertions(+), 26 deletions(-)
arch/arm/kernel/hw_breakpoint.c +31 -26
index 515a3c4..d37ed35 100644
@@ -537,6 +537,17 @@ static int arch_build_bp_info(struct perf_event *bp)
 		return -EINVAL;
 	}
 
+	/*
+	 * Breakpoints must be of length 2 (thumb) or 4 (ARM) bytes.
+	 * Watchpoints can be of length 1, 2, 4 or 8 bytes if supported
+	 * by the hardware and must be aligned to the appropriate number of
+	 * bytes.
+	 */
+	if (info->ctrl.type == ARM_BREAKPOINT_EXECUTE &&
+	    info->ctrl.len != ARM_BREAKPOINT_LEN_2 &&
+	    info->ctrl.len != ARM_BREAKPOINT_LEN_4)
+		return -EINVAL;
+
 	/* Address */
 	info->address = bp->attr.bp_addr;
 
@@ -561,7 +572,7 @@ int arch_validate_hwbkpt_settings(struct perf_event *bp)
 {
 	struct arch_hw_breakpoint *info = counter_arch_bp(bp);
 	int ret = 0;
-	u32 bytelen, max_len, offset, alignment_mask = 0x3;
+	u32 offset, alignment_mask = 0x3;
 
 	/* Build the arch_hw_breakpoint. */
 	ret = arch_build_bp_info(bp);
@@ -571,32 +582,27 @@ int arch_validate_hwbkpt_settings(struct perf_event *bp)
 	/* Check address alignment. */
 	if (info->ctrl.len == ARM_BREAKPOINT_LEN_8)
 		alignment_mask = 0x7;
-	if (info->address & alignment_mask) {
-		/*
-		 * Try to fix the alignment. This may result in a length
-		 * that is too large, so we must check for that.
-		 */
-		bytelen = get_hbp_len(info->ctrl.len);
-		max_len = info->ctrl.type == ARM_BREAKPOINT_EXECUTE ? 4 :
-				max_watchpoint_len;
-
-		if (max_len >= 8)
-			offset = info->address & 0x7;
-		else
-			offset = info->address & 0x3;
-
-		if (bytelen > (1 << ((max_len - (offset + 1)) >> 1))) {
-			ret = -EFBIG;
-			goto out;
-		}
-
-		info->ctrl.len <<= offset;
-		info->address &= ~offset;
-
-		pr_debug("breakpoint alignment fixup: length = 0x%x, "
-			"address = 0x%x\n", info->ctrl.len, info->address);
+	offset = info->address & alignment_mask;
+	switch (offset) {
+	case 0:
+		/* Aligned */
+		break;
+	case 1:
+		/* Allow single byte watchpoint. */
+		if (info->ctrl.len == ARM_BREAKPOINT_LEN_1)
+			break;
+	case 2:
+		/* Allow halfword watchpoints and breakpoints. */
+		if (info->ctrl.len == ARM_BREAKPOINT_LEN_2)
+			break;
+	default:
+		ret = -EINVAL;
+		goto out;
 	}
 
+	info->address &= ~alignment_mask;
+	info->ctrl.len <<= offset;
+
 	/*
 	 * Currently we rely on an overflow handler to take
 	 * care of single-stepping the breakpoint when it fires.
@@ -607,7 +613,6 @@ int arch_validate_hwbkpt_settings(struct perf_event *bp)
 		(arch_check_bp_in_kernelspace(bp) || !core_has_mismatch_bps()),
 			"overflow handler required but none found")) {
 		ret = -EINVAL;
-		goto out;
 	}
 out:
 	return ret;
Signature · 1 line
1.7.0.4
#4 lore raw

PATCH 4/11 ARM: hw_breakpoint: disable preemption during debug exception handling

In reply to Will Deacon

On ARM, debug exceptions occur in the form of data or prefetch aborts. One difference is that debug exceptions require access to per-cpu banked registers and data structures which are not saved in the low-level exception code. For kernels built with CONFIG_PREEMPT, there is an unlikely scenario that the debug handler ends up running on a different CPU from the one that originally signalled the event, resulting in random data being read from the wrong registers.

This patch adds a debug_entry macro to the low-level exception handling code which checks whether the taken exception is a debug exception. If it is, the preempt count for the faulting process is incremented. After the debug handler has finished, the count is decremented.

---
 arch/arm/kernel/entry-armv.S    |    4 ++++
 arch/arm/kernel/entry-header.S  |   19 +++++++++++++++++++
 arch/arm/kernel/hw_breakpoint.c |   18 +++++++++++++-----
 3 files changed, 36 insertions(+), 5 deletions(-)
arch/arm/kernel/entry-armv.S +4 -0
index c09e357..34bbef0 100644
@@ -198,6 +198,7 @@ __dabt_svc:
 	@
 	@ set desired IRQ state, then call main handler
 	@
+	debug_entry r1
 	msr	cpsr_c, r9
 	mov	r2, sp
 	bl	do_DataAbort
@@ -324,6 +325,7 @@ __pabt_svc:
 #else
 	bl	CPU_PABORT_HANDLER
 #endif
+	debug_entry r1
 	msr	cpsr_c, r9			@ Maybe enable interrupts
 	mov	r2, sp				@ regs
 	bl	do_PrefetchAbort		@ call abort handler
@@ -439,6 +441,7 @@ __dabt_usr:
 	@
 	@ IRQs on, then call the main handler
 	@
+	debug_entry r1
 	enable_irq
 	mov	r2, sp
 	adr	lr, BSYM(ret_from_exception)
@@ -703,6 +706,7 @@ __pabt_usr:
 #else
 	bl	CPU_PABORT_HANDLER
 #endif
+	debug_entry r1
 	enable_irq				@ Enable interrupts
 	mov	r2, sp				@ regs
 	bl	do_PrefetchAbort		@ call abort handler
arch/arm/kernel/entry-header.S +19 -0
index d93f976..ae94649 100644
@@ -165,6 +165,25 @@
 	.endm
 #endif	/* !CONFIG_THUMB2_KERNEL */
 
+	@
+	@ Debug exceptions are taken as prefetch or data aborts.
+	@ We must disable preemption during the handler so that
+	@ we can access the debug registers safely.
+	@
+	.macro	debug_entry, fsr
+#if defined(CONFIG_HAVE_HW_BREAKPOINT) && defined(CONFIG_PREEMPT)
+	ldr	r4, =0x40f		@ mask out fsr.fs
+	and	r5, r4, \fsr
+	cmp	r5, #2			@ debug exception
+	bne	1f
+	get_thread_info r10
+	ldr	r6, [r10, #TI_PREEMPT]	@ get preempt count
+	add	r11, r6, #1		@ increment it
+	str	r11, [r10, #TI_PREEMPT]
+1:
+#endif
+	.endm
+
 /*
  * These are the registers used in the syscall handler, and allow us to
  * have in theory up to 7 arguments to a function - r0 to r6.
arch/arm/kernel/hw_breakpoint.c +13 -5
index d37ed35..eeba380 100644
@@ -24,6 +24,7 @@
 #define pr_fmt(fmt) "hw-breakpoint: " fmt
 
 #include <linux/errno.h>
+#include <linux/hardirq.h>
 #include <linux/perf_event.h>
 #include <linux/hw_breakpoint.h>
 #include <linux/smp.h>
@@ -736,14 +737,17 @@ unlock:
 
 /*
  * Called from either the Data Abort Handler [watchpoint] or the
- * Prefetch Abort Handler [breakpoint].
+ * Prefetch Abort Handler [breakpoint] with preemption disabled.
  */
 static int hw_breakpoint_pending(unsigned long addr, unsigned int fsr,
 				 struct pt_regs *regs)
 {
-	int ret = 1; /* Unhandled fault. */
+	int ret = 0;
 	u32 dscr;
 
+	/* We must be called with preemption disabled. */
+	WARN_ON(preemptible());
+
 	/* We only handle watchpoints and hardware breakpoints. */
 	ARM_DBG_READ(c1, 0, dscr);
 
@@ -758,11 +762,15 @@ static int hw_breakpoint_pending(unsigned long addr, unsigned int fsr,
 		watchpoint_handler(addr, regs);
 		break;
 	default:
-		goto out;
+		ret = 1; /* Unhandled fault. */
 	}
 
-	ret = 0;
-out:
+	/*
+	 * Re-enable preemption after it was disabled in the
+	 * low-level exception handling code.
+	 */
+	preempt_enable();
+
 	return ret;
 }
 
Signature · 1 line
1.7.0.4
#5 lore raw

PATCH 5/11 ARM: hw_breakpoint: don't advertise reserved breakpoints

In reply to Will Deacon

To permit handling of watchpoint exceptions without signalling a debugger, it is necessary to reserve breakpoint registers for in-kernel use only.

This patch ensures that we record and subtract the number of reserved breakpoints from the number of usable breakpoint registers that we advertise to userspace via the ptrace API.

---
 arch/arm/kernel/hw_breakpoint.c |  206 ++++++++++++++++++++++-----------------
 1 files changed, 117 insertions(+), 89 deletions(-)
arch/arm/kernel/hw_breakpoint.c +117 -89
index eeba380..b16c456 100644
@@ -45,6 +45,7 @@ static DEFINE_PER_CPU(struct perf_event *, wp_on_reg[ARM_MAX_WRP]);
 
 /* Number of BRP/WRP registers on this CPU. */
 static int core_num_brps;
+static int core_num_reserved_brps;
 static int core_num_wrps;
 
 /* Debug architecture version. */
@@ -53,87 +54,6 @@ static u8 debug_arch;
 /* Maximum supported watchpoint length. */
 static u8 max_watchpoint_len;
 
-/* Determine number of BRP registers available. */
-static int get_num_brps(void)
-{
-	u32 didr;
-	ARM_DBG_READ(c0, 0, didr);
-	return ((didr >> 24) & 0xf) + 1;
-}
-
-/* Determine number of WRP registers available. */
-static int get_num_wrps(void)
-{
-	/*
-	 * FIXME: When a watchpoint fires, the only way to work out which
-	 * watchpoint it was is by disassembling the faulting instruction
-	 * and working out the address of the memory access.
-	 *
-	 * Furthermore, we can only do this if the watchpoint was precise
-	 * since imprecise watchpoints prevent us from calculating register
-	 * based addresses.
-	 *
-	 * For the time being, we only report 1 watchpoint register so we
-	 * always know which watchpoint fired. In the future we can either
-	 * add a disassembler and address generation emulator, or we can
-	 * insert a check to see if the DFAR is set on watchpoint exception
-	 * entry [the ARM ARM states that the DFAR is UNKNOWN, but
-	 * experience shows that it is set on some implementations].
-	 */
-
-#if 0
-	u32 didr, wrps;
-	ARM_DBG_READ(c0, 0, didr);
-	return ((didr >> 28) & 0xf) + 1;
-#endif
-
-	return 1;
-}
-
-int hw_breakpoint_slots(int type)
-{
-	/*
-	 * We can be called early, so don't rely on
-	 * our static variables being initialised.
-	 */
-	switch (type) {
-	case TYPE_INST:
-		return get_num_brps();
-	case TYPE_DATA:
-		return get_num_wrps();
-	default:
-		pr_warning("unknown slot type: %d\n", type);
-		return 0;
-	}
-}
-
-/* Determine debug architecture. */
-static u8 get_debug_arch(void)
-{
-	u32 didr;
-
-	/* Do we implement the extended CPUID interface? */
-	if (((read_cpuid_id() >> 16) & 0xf) != 0xf) {
-		pr_warning("CPUID feature registers not supported. "
-				"Assuming v6 debug is present.\n");
-		return ARM_DEBUG_ARCH_V6;
-	}
-
-	ARM_DBG_READ(c0, 0, didr);
-	return (didr >> 16) & 0xf;
-}
-
-/* Does this core support mismatch breakpoints? */
-static int core_has_mismatch_bps(void)
-{
-	return debug_arch >= ARM_DEBUG_ARCH_V7_ECP14 && core_num_brps > 1;
-}
-
-u8 arch_get_debug_arch(void)
-{
-	return debug_arch;
-}
-
 #define READ_WB_REG_CASE(OP2, M, VAL)		\
 	case ((OP2 << 4) + M):			\
 		ARM_DBG_READ(c ## M, OP2, VAL); \
@@ -211,6 +131,111 @@ static void write_wb_reg(int n, u32 val)
 	isb();
 }
 
+/* Determine debug architecture. */
+static u8 get_debug_arch(void)
+{
+	u32 didr;
+
+	/* Do we implement the extended CPUID interface? */
+	if (((read_cpuid_id() >> 16) & 0xf) != 0xf) {
+		pr_warning("CPUID feature registers not supported. "
+				"Assuming v6 debug is present.\n");
+		return ARM_DEBUG_ARCH_V6;
+	}
+
+	ARM_DBG_READ(c0, 0, didr);
+	return (didr >> 16) & 0xf;
+}
+
+u8 arch_get_debug_arch(void)
+{
+	return debug_arch;
+}
+
+/* Determine number of BRP register available. */
+static int get_num_brp_resources(void)
+{
+	u32 didr;
+	ARM_DBG_READ(c0, 0, didr);
+	return ((didr >> 24) & 0xf) + 1;
+}
+
+/* Does this core support mismatch breakpoints? */
+static int core_has_mismatch_brps(void)
+{
+	return (get_debug_arch() >= ARM_DEBUG_ARCH_V7_ECP14 &&
+		get_num_brp_resources() > 1);
+}
+
+/* Determine number of usable WRPs available. */
+static int get_num_wrps(void)
+{
+	/*
+	 * FIXME: When a watchpoint fires, the only way to work out which
+	 * watchpoint it was is by disassembling the faulting instruction
+	 * and working out the address of the memory access.
+	 *
+	 * Furthermore, we can only do this if the watchpoint was precise
+	 * since imprecise watchpoints prevent us from calculating register
+	 * based addresses.
+	 *
+	 * Providing we have more than 1 breakpoint register, we only report
+	 * a single watchpoint register for the time being. This way, we always
+	 * know which watchpoint fired. In the future we can either add a
+	 * disassembler and address generation emulator, or we can insert a
+	 * check to see if the DFAR is set on watchpoint exception entry
+	 * [the ARM ARM states that the DFAR is UNKNOWN, but experience shows
+	 * that it is set on some implementations].
+	 */
+
+#if 0
+	int wrps;
+	u32 didr;
+	ARM_DBG_READ(c0, 0, didr);
+	wrps = ((didr >> 28) & 0xf) + 1;
+#endif
+	int wrps = 1;
+
+	if (core_has_mismatch_brps() && wrps >= get_num_brp_resources())
+		wrps = get_num_brp_resources() - 1;
+
+	return wrps;
+}
+
+/* We reserve one breakpoint for each watchpoint. */
+static int get_num_reserved_brps(void)
+{
+	if (core_has_mismatch_brps())
+		return get_num_wrps();
+	return 0;
+}
+
+/* Determine number of usable BRPs available. */
+static int get_num_brps(void)
+{
+	int brps = get_num_brp_resources();
+	if (core_has_mismatch_brps())
+		brps -= get_num_reserved_brps();
+	return brps;
+}
+
+int hw_breakpoint_slots(int type)
+{
+	/*
+	 * We can be called early, so don't rely on
+	 * our static variables being initialised.
+	 */
+	switch (type) {
+	case TYPE_INST:
+		return get_num_brps();
+	case TYPE_DATA:
+		return get_num_wrps();
+	default:
+		pr_warning("unknown slot type: %d\n", type);
+		return 0;
+	}
+}
+
 /*
  * In order to access the breakpoint/watchpoint control registers,
  * we must be running in debug monitor mode. Unfortunately, we can
@@ -326,7 +351,7 @@ int arch_install_hw_breakpoint(struct perf_event *bp)
 		ctrl_base = ARM_BASE_BCR;
 		val_base = ARM_BASE_BVR;
 		slots = __get_cpu_var(bp_on_reg);
-		max_slots = core_num_brps - 1;
+		max_slots = core_num_brps;
 
 		if (bp_is_single_step(bp)) {
 			info->ctrl.mismatch = 1;
@@ -377,7 +402,7 @@ void arch_uninstall_hw_breakpoint(struct perf_event *bp)
 		/* Breakpoint */
 		base = ARM_BASE_BCR;
 		slots = __get_cpu_var(bp_on_reg);
-		max_slots = core_num_brps - 1;
+		max_slots = core_num_brps;
 
 		if (bp_is_single_step(bp)) {
 			i = max_slots;
@@ -611,7 +636,7 @@ int arch_validate_hwbkpt_settings(struct perf_event *bp)
 	 * we can use the mismatch feature as a poor-man's hardware single-step.
 	 */
 	if (WARN_ONCE(!bp->overflow_handler &&
-		(arch_check_bp_in_kernelspace(bp) || !core_has_mismatch_bps()),
+		(arch_check_bp_in_kernelspace(bp) || !core_has_mismatch_brps()),
 			"overflow handler required but none found")) {
 		ret = -EINVAL;
 	}
@@ -698,7 +723,7 @@ static void breakpoint_handler(unsigned long unknown, struct pt_regs *regs)
 	/* The exception entry code places the amended lr in the PC. */
 	addr = regs->ARM_pc;
 
-	for (i = 0; i < core_num_brps; ++i) {
+	for (i = 0; i < core_num_brps + core_num_reserved_brps; ++i) {
 		rcu_read_lock();
 
 		bp = slots[i];
@@ -801,7 +826,8 @@ static void reset_ctrl_regs(void *unused)
 	if (enable_monitor_mode())
 		return;
 
-	for (i = 0; i < core_num_brps; ++i) {
+	/* We must also reset any reserved registers. */
+	for (i = 0; i < core_num_brps + core_num_reserved_brps; ++i) {
 		write_wb_reg(ARM_BASE_BCR + i, 0UL);
 		write_wb_reg(ARM_BASE_BVR + i, 0UL);
 	}
@@ -839,13 +865,15 @@ static int __init arch_hw_breakpoint_init(void)
 
 	/* Determine how many BRPs/WRPs are available. */
 	core_num_brps = get_num_brps();
+	core_num_reserved_brps = get_num_reserved_brps();
 	core_num_wrps = get_num_wrps();
 
 	pr_info("found %d breakpoint and %d watchpoint registers.\n",
-			core_num_brps, core_num_wrps);
+		core_num_brps + core_num_reserved_brps, core_num_wrps);
 
-	if (core_has_mismatch_bps())
-		pr_info("1 breakpoint reserved for watchpoint single-step.\n");
+	if (core_num_reserved_brps)
+		pr_info("%d breakpoint(s) reserved for watchpoint "
+				"single-step.\n", core_num_reserved_brps);
 
 	ARM_DBG_READ(c1, 0, dscr);
 	if (dscr & ARM_DSCR_HDBGEN) {
Signature · 1 line
1.7.0.4
#6 lore raw

PATCH 6/11 ARM: hw_breakpoint: do not allocate new breakpoints with preemption disabled

In reply to Will Deacon

The watchpoint single-stepping code calls register_user_hw_breakpoint to register a mismatch breakpoint for stepping over the watchpoint. This is performed with preemption disabled, which is unsafe as we may end up scheduling whilst in_atomic(). Furthermore, using the perf API is rather overkill since we are already in the hw-breakpoint backend and only require access to reserved breakpoints anyway.

This patch reworks the watchpoint stepping code so that we don't require another perf_event for the mismatch breakpoint. Instead, we hold a separate arch_hw_breakpoint_ctrl struct inside the watchpoint which is used exclusively for stepping. We can check whether or not stepping is enabled when installing or uninstalling the watchpoint and operate on the breakpoint accordingly.

---
 arch/arm/include/asm/hw_breakpoint.h |    2 +-
 arch/arm/kernel/hw_breakpoint.c      |  134 +++++++++++++++++++--------------
 2 files changed, 78 insertions(+), 58 deletions(-)
arch/arm/include/asm/hw_breakpoint.h +1 -1
index 4d8ae9d..881429d 100644
@@ -20,7 +20,7 @@ struct arch_hw_breakpoint_ctrl {
 struct arch_hw_breakpoint {
 	u32	address;
 	u32	trigger;
-	struct perf_event *suspended_wp;
+	struct arch_hw_breakpoint_ctrl step_ctrl;
 	struct arch_hw_breakpoint_ctrl ctrl;
 };
 
arch/arm/kernel/hw_breakpoint.c +77 -57

236 lines not on this page · raw message

Signature · 1 line
1.7.0.4
#7 lore raw

PATCH 7/11 ARM: hw_breakpoint: unify single-stepping code for watchpoints and breakpoints

In reply to Will Deacon

The single-stepping code is currently different depending on whether we are stepping over a breakpoint or a watchpoint. There is no good reason for this, so let's sort it out.

This patch adds functions for enabling/disabling single-step for a particular hw_breakpoint and integrates this with the exception handling code.

---
 arch/arm/include/asm/hw_breakpoint.h |    4 +-
 arch/arm/kernel/hw_breakpoint.c      |   81 +++++++++++++++++-----------------
 2 files changed, 42 insertions(+), 43 deletions(-)
arch/arm/include/asm/hw_breakpoint.h +2 -2
index 881429d..f389b27 100644
@@ -20,8 +20,8 @@ struct arch_hw_breakpoint_ctrl {
 struct arch_hw_breakpoint {
 	u32	address;
 	u32	trigger;
-	struct arch_hw_breakpoint_ctrl step_ctrl;
-	struct arch_hw_breakpoint_ctrl ctrl;
+	struct	arch_hw_breakpoint_ctrl step_ctrl;
+	struct	arch_hw_breakpoint_ctrl ctrl;
 };
 
 static inline u32 encode_ctrl_reg(struct arch_hw_breakpoint_ctrl ctrl)
arch/arm/kernel/hw_breakpoint.c +40 -41

135 lines not on this page · raw message

Signature · 1 line
1.7.0.4
#8 lore raw

PATCH 8/11 ARM: hw_breakpoint: disallow per-cpu breakpoints without overflow handler

In reply to Will Deacon

Single-stepping a breakpoint requires us to disable it temporarily so that we don't get stuck in a recursive debug trap. With per-cpu breakpoints this presents a problem where an interrupt can be taken before the single-step has completed and a new task is eventually scheduled. This new task will not hit the breakpoint because it will have been disabled during the previous handling code.

This patch disallows per-cpu breakpoints on ARM when an overflow handler is not present. A similar effect can be created by placing breakpoints on a shell and then running applications there.

---
 arch/arm/kernel/hw_breakpoint.c |    6 ++++--
 1 files changed, 4 insertions(+), 2 deletions(-)
arch/arm/kernel/hw_breakpoint.c +4 -2
index 36cd768..eef1b1e 100644
@@ -622,10 +622,12 @@ int arch_validate_hwbkpt_settings(struct perf_event *bp)
 	 * Currently we rely on an overflow handler to take
 	 * care of single-stepping the breakpoint when it fires.
 	 * In the case of userspace breakpoints on a core with V7 debug,
-	 * we can use the mismatch feature as a poor-man's hardware single-step.
+	 * we can use the mismatch feature as a poor-man's hardware
+	 * single-step, but this only works for per-task breakpoints.
 	 */
 	if (WARN_ONCE(!bp->overflow_handler &&
-		(arch_check_bp_in_kernelspace(bp) || !core_has_mismatch_brps()),
+		(arch_check_bp_in_kernelspace(bp) || !core_has_mismatch_brps()
+		 || !bp->hw.bp_target),
 			"overflow handler required but none found")) {
 		ret = -EINVAL;
 	}
Signature · 1 line
1.7.0.4
#9 lore raw

PATCH 9/11 ARM: ptrace: fix style issue with hw_breakpoint interface

In reply to Will Deacon

This patch fixes a trivial style issue in ptrace.c.

---
 arch/arm/kernel/ptrace.c |    4 ++--
 1 files changed, 2 insertions(+), 2 deletions(-)
arch/arm/kernel/ptrace.c +2 -2
index 3e97483..19c6816 100644
@@ -1060,8 +1060,8 @@ static int ptrace_sethbpregs(struct task_struct *tsk, long num,
 			goto out;
 
 		if ((gen_type & implied_type) != gen_type) {
-				ret = -EINVAL;
-				goto out;
+			ret = -EINVAL;
+			goto out;
 		}
 
 		attr.bp_len	= gen_len;
Signature · 1 line
1.7.0.4
#10 lore raw

PATCH 10/11 ARM: hw_breakpoint: fix warnings generated by sparse

In reply to Will Deacon

sparse doesn't like per-cpu accesses such as:

static DEFINE_PER_CPU(struct perf_event *, foo[MAXLEN]); struct perf_event **bar = __get_cpu_var(foo);

and shouts quite loudly about it:

| warning: incorrect type in assignment (different modifiers)
|    expected struct perf_event **slots
|    got struct perf_event *[noderef] *<noident>

This patch adds casts to these sorts of assignments in hw_breakpoint.c in order to silence the warnings.

---
 arch/arm/kernel/hw_breakpoint.c |   20 +++++++++++++-------
 1 files changed, 13 insertions(+), 7 deletions(-)
arch/arm/kernel/hw_breakpoint.c +13 -7
index eef1b1e..56ed9a6 100644
@@ -337,7 +337,7 @@ int arch_install_hw_breakpoint(struct perf_event *bp)
 		/* Breakpoint */
 		ctrl_base = ARM_BASE_BCR;
 		val_base = ARM_BASE_BVR;
-		slots = __get_cpu_var(bp_on_reg);
+		slots = (struct perf_event **)__get_cpu_var(bp_on_reg);
 		max_slots = core_num_brps;
 		if (info->step_ctrl.enabled) {
 			/* Override the breakpoint data with the step data. */
@@ -357,7 +357,7 @@ int arch_install_hw_breakpoint(struct perf_event *bp)
 			ctrl_base = ARM_BASE_WCR;
 			val_base = ARM_BASE_WVR;
 		}
-		slots = __get_cpu_var(wp_on_reg);
+		slots = (struct perf_event **)__get_cpu_var(wp_on_reg);
 		max_slots = core_num_wrps;
 	}
 
@@ -394,7 +394,7 @@ void arch_uninstall_hw_breakpoint(struct perf_event *bp)
 	if (info->ctrl.type == ARM_BREAKPOINT_EXECUTE) {
 		/* Breakpoint */
 		base = ARM_BASE_BCR;
-		slots = __get_cpu_var(bp_on_reg);
+		slots = (struct perf_event **)__get_cpu_var(bp_on_reg);
 		max_slots = core_num_brps;
 	} else {
 		/* Watchpoint */
@@ -402,7 +402,7 @@ void arch_uninstall_hw_breakpoint(struct perf_event *bp)
 			base = ARM_BASE_BCR + core_num_brps;
 		else
 			base = ARM_BASE_WCR;
-		slots = __get_cpu_var(wp_on_reg);
+		slots = (struct perf_event **)__get_cpu_var(wp_on_reg);
 		max_slots = core_num_wrps;
 	}
 
@@ -662,9 +662,11 @@ static void disable_single_step(struct perf_event *bp)
 static void watchpoint_handler(unsigned long unknown, struct pt_regs *regs)
 {
 	int i;
-	struct perf_event *wp, **slots = __get_cpu_var(wp_on_reg);
+	struct perf_event *wp, **slots;
 	struct arch_hw_breakpoint *info;
 
+	slots = (struct perf_event **)__get_cpu_var(wp_on_reg);
+
 	/* Without a disassembler, we can only handle 1 watchpoint. */
 	BUG_ON(core_num_wrps > 1);
 
@@ -703,9 +705,11 @@ static void watchpoint_handler(unsigned long unknown, struct pt_regs *regs)
 static void watchpoint_single_step_handler(unsigned long pc)
 {
 	int i;
-	struct perf_event *wp, **slots = __get_cpu_var(wp_on_reg);
+	struct perf_event *wp, **slots;
 	struct arch_hw_breakpoint *info;
 
+	slots = (struct perf_event **)__get_cpu_var(wp_on_reg);
+
 	for (i = 0; i < core_num_reserved_brps; ++i) {
 		rcu_read_lock();
 
@@ -734,10 +738,12 @@ static void breakpoint_handler(unsigned long unknown, struct pt_regs *regs)
 {
 	int i;
 	u32 ctrl_reg, val, addr;
-	struct perf_event *bp, **slots = __get_cpu_var(bp_on_reg);
+	struct perf_event *bp, **slots;
 	struct arch_hw_breakpoint *info;
 	struct arch_hw_breakpoint_ctrl ctrl;
 
+	slots = (struct perf_event **)__get_cpu_var(bp_on_reg);
+
 	/* The exception entry code places the amended lr in the PC. */
 	addr = regs->ARM_pc;
 
Signature · 1 line
1.7.0.4
#12 lore raw

PATCH 10/11 ARM: hw_breakpoint: fix warnings generated by sparse

Hi Russell,

Quoted 12 lines · Will Deacon
Quoted 4 lines

This patch adds casts to these sorts of assignments in hw_breakpoint.c in order to silence the warnings.

Please change this to:

I don't need more - and different - email addresses in git for get_maintainer.pl etc to spam me with.

Sure thing, I'll update it. I just used the email address from which you reported the problem.

Will

#13 lore raw

PATCH 11/11 ARM: hw_breakpoint: do not fail initcall if monitor mode is disabled

In reply to Will Deacon

The debug registers can only be manipulated from software if monitor debug mode is enabled. On some cores, this can never be enabled (i.e. the corresponding bit the the DSCR is RAZ/WI).

This patch ensures we can handle this hardware configuration and fail gracefully, rather than blow up the kernel during boot.

---
 arch/arm/kernel/hw_breakpoint.c |   22 +++++++++++-----------
 1 files changed, 11 insertions(+), 11 deletions(-)
arch/arm/kernel/hw_breakpoint.c +11 -11
index 56ed9a6..da647d9 100644
@@ -225,6 +225,9 @@ int hw_breakpoint_slots(int type)
 	 * We can be called early, so don't rely on
 	 * our static variables being initialised.
 	 */
+	if (enable_monitor_mode())
+		return 0;
+
 	switch (type) {
 	case TYPE_INST:
 		return get_num_brps();
@@ -272,10 +275,8 @@ static int enable_monitor_mode(void)
 
 	/* Check that the write made it through. */
 	ARM_DBG_READ(c1, 0, dscr);
-	if (WARN_ONCE(!(dscr & ARM_DSCR_MDBGEN),
-				"failed to enable monitor mode.")) {
+	if (!(dscr & ARM_DSCR_MDBGEN))
 		ret = -EPERM;
-	}
 
 out:
 	return ret;
@@ -294,9 +295,6 @@ static u8 get_max_wp_len(void)
 	if (debug_arch < ARM_DEBUG_ARCH_V7_ECP14)
 		goto out;
 
-	if (enable_monitor_mode())
-		goto out;
-
 	memset(&ctrl, 0, sizeof(ctrl));
 	ctrl.len = ARM_BREAKPOINT_LEN_8;
 	ctrl_reg = encode_ctrl_reg(ctrl);
@@ -879,15 +877,18 @@ static struct notifier_block __cpuinitdata dbg_reset_nb = {
 
 static int __init arch_hw_breakpoint_init(void)
 {
-	int ret = 0;
 	u32 dscr;
 
 	debug_arch = get_debug_arch();
 
 	if (debug_arch > ARM_DEBUG_ARCH_V7_ECP14) {
 		pr_info("debug architecture 0x%x unsupported.\n", debug_arch);
-		ret = -ENODEV;
-		goto out;
+		return 0;
+	}
+
+	if (enable_monitor_mode()) {
+		pr_warning("unable to activate monitor debug mode.\n");
+		return 0;
 	}
 
 	/* Determine how many BRPs/WRPs are available. */
@@ -928,8 +929,7 @@ static int __init arch_hw_breakpoint_init(void)
 
 	/* Register hotplug notifier. */
 	register_cpu_notifier(&dbg_reset_nb);
-out:
-	return ret;
+	return 0;
 }
 arch_initcall(arch_hw_breakpoint_init);
 
Signature · 1 line
1.7.0.4