sh: Abstracted SH-4A UBC support on hw-breakpoint core.

This is the next big chunk of hw_breakpoint support. This decouples
the SH-4A support from the core and moves it out in to its own stub,
following many of the conventions established with the perf events
layering.

In addition to extending SH-4A support to encapsulate the remainder
of the UBC channels, clock framework support for handling the UBC
interface clock is added as well, allowing for dynamic clock gating.

This also fixes up a regression introduced by the SIGTRAP handling that
broke the ksym_tracer, to the extent that the current support works well
with all of the ksym_tracer/ptrace/kgdb. The kprobes singlestep code will
follow in turn.

With this in place, the remaining UBC variants (SH-2A and SH-4) can now
be trivially plugged in.
Signed-off-by: Paul Mundt <lethal@linux-sh.org>

arch/sh/include/asm/hw_breakpoint.h

@@ -3,7 +3,6 @@
 
 #include <linux/kdebug.h>
 #include <linux/types.h>
-#include <asm/ubc.h>
 
 #ifdef __KERNEL__
 #define __ARCH_HW_BREAKPOINT_H
@@ -11,7 +10,6 @@
 struct arch_hw_breakpoint {
 	char		*name; /* Contains name of the symbol to set bkpt */
 	unsigned long	address;
-	unsigned long	asid;
 	u16		len;
 	u16		type;
 };
@@ -27,13 +25,28 @@ enum {
 	SH_BREAKPOINT_LEN_8	= (1 << 14),
 };
 
-/* Total number of available UBC channels */
-#define HBP_NUM		1 /* XXX */
+struct sh_ubc {
+	const char	*name;
+	unsigned int	num_events;
+	unsigned int	trap_nr;
+	void		(*enable)(struct arch_hw_breakpoint *, int);
+	void		(*disable)(struct arch_hw_breakpoint *, int);
+	void		(*enable_all)(unsigned long);
+	void		(*disable_all)(void);
+	unsigned long	(*active_mask)(void);
+	unsigned long	(*triggered_mask)(void);
+	void		(*clear_triggered_mask)(unsigned long);
+	struct clk	*clk;	/* optional interface clock / MSTP bit */
+};
 
 struct perf_event;
 struct task_struct;
 struct pmu;
 
+/* Maximum number of UBC channels */
+#define HBP_NUM		2
+
+/* arch/sh/kernel/hw_breakpoint.c */
 extern int arch_check_va_in_userspace(unsigned long va, u16 hbp_len);
 extern int arch_validate_hwbkpt_settings(struct perf_event *bp,
 					 struct task_struct *tsk);
@@ -46,6 +59,7 @@ void hw_breakpoint_pmu_read(struct perf_event *bp);
 void hw_breakpoint_pmu_unthrottle(struct perf_event *bp);
 
 extern void arch_fill_perf_breakpoint(struct perf_event *bp);
+extern int register_sh_ubc(struct sh_ubc *);
 
 extern struct pmu perf_ops_bp;
 

arch/sh/include/asm/processor_32.h

@@ -14,7 +14,7 @@
 #include <asm/page.h>
 #include <asm/types.h>
 #include <asm/ptrace.h>
-#include <asm/ubc.h>
+#include <asm/hw_breakpoint.h>
 
 /*
  * Default implementation of macro that returns current
@@ -102,7 +102,7 @@ struct thread_struct {
 	unsigned long pc;
 
 	/* Save middle states of ptrace breakpoints */
-	struct perf_event	*ptrace_bps[NR_UBC_CHANNELS];
+	struct perf_event	*ptrace_bps[HBP_NUM];
 
 	/* floating point info */
 	union sh_fpu_union fpu;

arch/sh/kernel/cpu/sh4a/Makefile

@@ -41,7 +41,8 @@ pinmux-$(CONFIG_CPU_SUBTYPE_SH7757)	:= pinmux-sh7757.o
 pinmux-$(CONFIG_CPU_SUBTYPE_SH7785)	:= pinmux-sh7785.o
 pinmux-$(CONFIG_CPU_SUBTYPE_SH7786)	:= pinmux-sh7786.o
 
-obj-y				+= $(clock-y)
-obj-$(CONFIG_SMP)		+= $(smp-y)
-obj-$(CONFIG_GENERIC_GPIO)	+= $(pinmux-y)
-obj-$(CONFIG_PERF_EVENTS)	+= perf_event.o
+obj-y					+= $(clock-y)
+obj-$(CONFIG_SMP)			+= $(smp-y)
+obj-$(CONFIG_GENERIC_GPIO)		+= $(pinmux-y)
+obj-$(CONFIG_PERF_EVENTS)		+= perf_event.o
+obj-$(CONFIG_HAVE_HW_BREAKPOINT)	+= ubc.o

arch/sh/kernel/cpu/sh4a/ubc.c

+/*
+ * arch/sh/kernel/cpu/sh4a/ubc.c
+ *
+ * On-chip UBC support for SH-4A CPUs.
+ *
+ * Copyright (C) 2009 - 2010  Paul Mundt
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+#include <linux/init.h>
+#include <linux/err.h>
+#include <linux/clk.h>
+#include <linux/io.h>
+#include <asm/hw_breakpoint.h>
+
+#define UBC_CBR(idx)	(0xff200000 + (0x20 * idx))
+#define UBC_CRR(idx)	(0xff200004 + (0x20 * idx))
+#define UBC_CAR(idx)	(0xff200008 + (0x20 * idx))
+#define UBC_CAMR(idx)	(0xff20000c + (0x20 * idx))
+
+#define UBC_CCMFR	0xff200600
+#define UBC_CBCR	0xff200620
+
+/* CRR */
+#define UBC_CRR_PCB	(1 << 1)
+#define UBC_CRR_BIE	(1 << 0)
+
+/* CBR */
+#define UBC_CBR_CE	(1 << 0)
+
+static struct sh_ubc sh4a_ubc;
+
+static void sh4a_ubc_enable(struct arch_hw_breakpoint *info, int idx)
+{
+	__raw_writel(UBC_CBR_CE | info->len | info->type, UBC_CBR(idx));
+	__raw_writel(info->address, UBC_CAR(idx));
+}
+
+static void sh4a_ubc_disable(struct arch_hw_breakpoint *info, int idx)
+{
+	__raw_writel(0, UBC_CBR(idx));
+	__raw_writel(0, UBC_CAR(idx));
+}
+
+static void sh4a_ubc_enable_all(unsigned long mask)
+{
+	int i;
+
+	for (i = 0; i < sh4a_ubc.num_events; i++)
+		if (mask & (1 << i))
+			__raw_writel(__raw_readl(UBC_CBR(i)) | UBC_CBR_CE,
+				     UBC_CBR(i));
+}
+
+static void sh4a_ubc_disable_all(void)
+{
+	int i;
+
+	for (i = 0; i < sh4a_ubc.num_events; i++)
+		__raw_writel(__raw_readl(UBC_CBR(i)) & ~UBC_CBR_CE,
+			     UBC_CBR(i));
+}
+
+static unsigned long sh4a_ubc_active_mask(void)
+{
+	unsigned long active = 0;
+	int i;
+
+	for (i = 0; i < sh4a_ubc.num_events; i++)
+		if (__raw_readl(UBC_CBR(i)) & UBC_CBR_CE)
+			active |= (1 << i);
+
+	return active;
+}
+
+static unsigned long sh4a_ubc_triggered_mask(void)
+{
+	return __raw_readl(UBC_CCMFR);
+}
+
+static void sh4a_ubc_clear_triggered_mask(unsigned long mask)
+{
+	__raw_writel(__raw_readl(UBC_CCMFR) & ~mask, UBC_CCMFR);
+}
+
+static struct sh_ubc sh4a_ubc = {
+	.name			= "SH-4A",
+	.num_events		= 2,
+	.trap_nr		= 0x1e0,
+	.enable			= sh4a_ubc_enable,
+	.disable		= sh4a_ubc_disable,
+	.enable_all		= sh4a_ubc_enable_all,
+	.disable_all		= sh4a_ubc_disable_all,
+	.active_mask		= sh4a_ubc_active_mask,
+	.triggered_mask		= sh4a_ubc_triggered_mask,
+	.clear_triggered_mask	= sh4a_ubc_clear_triggered_mask,
+};
+
+static int __init sh4a_ubc_init(void)
+{
+	struct clk *ubc_iclk = clk_get(NULL, "ubc0");
+	int i;
+
+	/*
+	 * The UBC MSTP bit is optional, as not all platforms will have
+	 * it. Just ignore it if we can't find it.
+	 */
+	if (IS_ERR(ubc_iclk))
+		ubc_iclk = NULL;
+
+	clk_enable(ubc_iclk);
+
+	__raw_writel(0, UBC_CBCR);
+
+	for (i = 0; i < sh4a_ubc.num_events; i++) {
+		__raw_writel(0, UBC_CAMR(i));
+		__raw_writel(0, UBC_CBR(i));
+
+		__raw_writel(UBC_CRR_BIE | UBC_CRR_PCB, UBC_CRR(i));
+
+		/* dummy read for write posting */
+		(void)__raw_readl(UBC_CRR(i));
+	}
+
+	clk_disable(ubc_iclk);
+
+	sh4a_ubc.clk = ubc_iclk;
+
+	return register_sh_ubc(&sh4a_ubc);
+}
+arch_initcall(sh4a_ubc_init);

arch/sh/kernel/hw_breakpoint.c

@@ -3,7 +3,7 @@
  *
  * Unified kernel/user-space hardware breakpoint facility for the on-chip UBC.
  *
- * Copyright (C) 2009  Paul Mundt
+ * Copyright (C) 2009 - 2010  Paul Mundt
  *
  * This file is subject to the terms and conditions of the GNU General Public
  * License.  See the file "COPYING" in the main directory of this archive
@@ -18,38 +18,24 @@
 #include <linux/kprobes.h>
 #include <linux/kdebug.h>
 #include <linux/io.h>
+#include <linux/clk.h>
 #include <asm/hw_breakpoint.h>
 #include <asm/mmu_context.h>
 #include <asm/ptrace.h>
 
-struct ubc_context {
-	unsigned long pc;
-	unsigned long state;
-};
-
-/* Per cpu ubc channel state */
-static DEFINE_PER_CPU(struct ubc_context, ubc_ctx[HBP_NUM]);
-
 /*
  * Stores the breakpoints currently in use on each breakpoint address
  * register for each cpus
  */
 static DEFINE_PER_CPU(struct perf_event *, bp_per_reg[HBP_NUM]);
 
-static int __init ubc_init(void)
-{
-	__raw_writel(0, UBC_CAMR0);
-	__raw_writel(0, UBC_CBR0);
-	__raw_writel(0, UBC_CBCR);
-
-	__raw_writel(UBC_CRR_BIE | UBC_CRR_PCB, UBC_CRR0);
-
-	/* dummy read for write posting */
-	(void)__raw_readl(UBC_CRR0);
+/*
+ * A dummy placeholder for early accesses until the CPUs get a chance to
+ * register their UBCs later in the boot process.
+ */
+static struct sh_ubc ubc_dummy = { .num_events = 0 };
 
-	return 0;
-}
-arch_initcall(ubc_init);
+static struct sh_ubc *sh_ubc __read_mostly = &ubc_dummy;
 
 /*
  * Install a perf counter breakpoint.
@@ -62,10 +48,9 @@ arch_initcall(ubc_init);
 int arch_install_hw_breakpoint(struct perf_event *bp)
 {
 	struct arch_hw_breakpoint *info = counter_arch_bp(bp);
-	struct ubc_context *ubc_ctx;
 	int i;
 
-	for (i = 0; i < HBP_NUM; i++) {
+	for (i = 0; i < sh_ubc->num_events; i++) {
 		struct perf_event **slot = &__get_cpu_var(bp_per_reg[i]);
 
 		if (!*slot) {
@@ -74,16 +59,11 @@ int arch_install_hw_breakpoint(struct perf_event *bp)
 		}
 	}
 
-	if (WARN_ONCE(i == HBP_NUM, "Can't find any breakpoint slot"))
+	if (WARN_ONCE(i == sh_ubc->num_events, "Can't find any breakpoint slot"))
 		return -EBUSY;
 
-	ubc_ctx = &__get_cpu_var(ubc_ctx[i]);
-
-	ubc_ctx->pc = info->address;
-	ubc_ctx->state = info->len | info->type;
-
-	__raw_writel(UBC_CBR_CE | ubc_ctx->state, UBC_CBR0);
-	__raw_writel(ubc_ctx->pc, UBC_CAR0);
+	clk_enable(sh_ubc->clk);
+	sh_ubc->enable(info, i);
 
 	return 0;
 }
@@ -100,10 +80,9 @@ int arch_install_hw_breakpoint(struct perf_event *bp)
 void arch_uninstall_hw_breakpoint(struct perf_event *bp)
 {
 	struct arch_hw_breakpoint *info = counter_arch_bp(bp);
-	struct ubc_context *ubc_ctx;
 	int i;
 
-	for (i = 0; i < HBP_NUM; i++) {
+	for (i = 0; i < sh_ubc->num_events; i++) {
 		struct perf_event **slot = &__get_cpu_var(bp_per_reg[i]);
 
 		if (*slot == bp) {
@@ -112,15 +91,11 @@ void arch_uninstall_hw_breakpoint(struct perf_event *bp)
 		}
 	}
 
-	if (WARN_ONCE(i == HBP_NUM, "Can't find any breakpoint slot"))
+	if (WARN_ONCE(i == sh_ubc->num_events, "Can't find any breakpoint slot"))
 		return;
 
-	ubc_ctx = &__get_cpu_var(ubc_ctx[i]);
-	ubc_ctx->pc = 0;
-	ubc_ctx->state &= ~(info->len | info->type);
-
-	__raw_writel(ubc_ctx->pc, UBC_CBR0);
-	__raw_writel(ubc_ctx->state, UBC_CAR0);
+	sh_ubc->disable(info, i);
+	clk_disable(sh_ubc->clk);
 }
 
 static int get_hbp_len(u16 hbp_len)
@@ -182,10 +157,8 @@ static int arch_store_info(struct perf_event *bp)
 	 */
 	if (info->name)
 		info->address = (unsigned long)kallsyms_lookup_name(info->name);
-	if (info->address) {
-		info->asid = get_asid();
+	if (info->address)
 		return 0;
-	}
 
 	return -EINVAL;
 }
@@ -335,7 +308,7 @@ void flush_ptrace_hw_breakpoint(struct task_struct *tsk)
 	int i;
 	struct thread_struct *t = &tsk->thread;
 
-	for (i = 0; i < HBP_NUM; i++) {
+	for (i = 0; i < sh_ubc->num_events; i++) {
 		unregister_hw_breakpoint(t->ptrace_bps[i]);
 		t->ptrace_bps[i] = NULL;
 	}
@@ -345,13 +318,32 @@ static int __kprobes hw_breakpoint_handler(struct die_args *args)
 {
 	int cpu, i, rc = NOTIFY_STOP;
 	struct perf_event *bp;
-	unsigned long val;
+	unsigned int cmf, resume_mask;
+
+	/*
+	 * Do an early return if none of the channels triggered.
+	 */
+	cmf = sh_ubc->triggered_mask();
+	if (unlikely(!cmf))
+		return NOTIFY_DONE;
+
+	/*
+	 * By default, resume all of the active channels.
+	 */
+	resume_mask = sh_ubc->active_mask();
 
-	val = __raw_readl(UBC_CBR0);
-	__raw_writel(val & ~UBC_CBR_CE, UBC_CBR0);
+	/*
+	 * Disable breakpoints during exception handling.
+	 */
+	sh_ubc->disable_all();
 
 	cpu = get_cpu();
-	for (i = 0; i < HBP_NUM; i++) {
+	for (i = 0; i < sh_ubc->num_events; i++) {
+		unsigned long event_mask = (1 << i);
+
+		if (likely(!(cmf & event_mask)))
+			continue;
+
 		/*
 		 * The counter may be concurrently released but that can only
 		 * occur from a call_rcu() path. We can then safely fetch
@@ -361,24 +353,52 @@ static int __kprobes hw_breakpoint_handler(struct die_args *args)
 		rcu_read_lock();
 
 		bp = per_cpu(bp_per_reg[i], cpu);
-		if (bp) {
+		if (bp)
 			rc = NOTIFY_DONE;
-		} else {
+
+		/*
+		 * Reset the condition match flag to denote completion of
+		 * exception handling.
+		 */
+		sh_ubc->clear_triggered_mask(event_mask);
+
+		/*
+		 * bp can be NULL due to concurrent perf counter
+		 * removing.
+		 */
+		if (!bp) {
 			rcu_read_unlock();
 			break;
 		}
 
+		/*
+		 * Don't restore the channel if the breakpoint is from
+		 * ptrace, as it always operates in one-shot mode.
+		 */
+		if (bp->overflow_handler == ptrace_triggered)
+			resume_mask &= ~(1 << i);
+
 		perf_bp_event(bp, args->regs);
 
+		/* Deliver the signal to userspace */
+		if (arch_check_va_in_userspace(bp->attr.bp_addr,
+					       bp->attr.bp_len)) {
+			siginfo_t info;
+
+			info.si_signo = args->signr;
+			info.si_errno = notifier_to_errno(rc);
+			info.si_code = TRAP_HWBKPT;
+
+			force_sig_info(args->signr, &info, current);
+		}
+
 		rcu_read_unlock();
 	}
 
-	if (bp && bp->overflow_handler != ptrace_triggered) {
-		struct arch_hw_breakpoint *info = counter_arch_bp(bp);
+	if (cmf == 0)
+		rc = NOTIFY_DONE;
 
-		__raw_writel(UBC_CBR_CE | info->len | info->type, UBC_CBR0);
-		__raw_writel(info->address, UBC_CAR0);
-	}
+	sh_ubc->enable_all(resume_mask);
 
 	put_cpu();
 
@@ -388,19 +408,9 @@ static int __kprobes hw_breakpoint_handler(struct die_args *args)
 BUILD_TRAP_HANDLER(breakpoint)
 {
 	unsigned long ex = lookup_exception_vector();
-	siginfo_t info;
-	int err;
 	TRAP_HANDLER_DECL;
 
-	err = notify_die(DIE_BREAKPOINT, "breakpoint", regs, 0, ex, SIGTRAP);
-	if (err == NOTIFY_STOP)
-		return;
-
-	/* Deliver the signal to userspace */
-	info.si_signo = SIGTRAP;
-	info.si_errno = 0;
-	info.si_code = TRAP_HWBKPT;
-	force_sig_info(SIGTRAP, &info, current);
+	notify_die(DIE_BREAKPOINT, "breakpoint", regs, 0, ex, SIGTRAP);
 }
 
 /*
@@ -417,8 +427,12 @@ int __kprobes hw_breakpoint_exceptions_notify(struct notifier_block *unused,
 	/*
 	 * If the breakpoint hasn't been triggered by the UBC, it's
 	 * probably from a debugger, so don't do anything more here.
+	 *
+	 * This also permits the UBC interface clock to remain off for
+	 * non-UBC breakpoints, as we don't need to check the triggered
+	 * or active channel masks.
 	 */
-	if (args->trapnr != 0x1e0)
+	if (args->trapnr != sh_ubc->trap_nr)
 		return NOTIFY_DONE;
 
 	return hw_breakpoint_handler(data);
@@ -433,3 +447,17 @@ void hw_breakpoint_pmu_unthrottle(struct perf_event *bp)
 {
 	/* TODO */
 }
+
+int register_sh_ubc(struct sh_ubc *ubc)
+{
+	/* Bail if it's already assigned */
+	if (sh_ubc != &ubc_dummy)
+		return -EBUSY;
+	sh_ubc = ubc;
+
+	pr_info("HW Breakpoints: %s UBC support registered\n", ubc->name);
+
+	WARN_ON(ubc->num_events > HBP_NUM);
+
+	return 0;
+}