hw-breakpoints: Modify breakpoints without unregistering them

Currently, when ptrace needs to modify a breakpoint, like disabling
it, changing its address, type or len, it calls
modify_user_hw_breakpoint(). This latter will perform the heavy and
racy task of unregistering the old breakpoint and registering a new
one.

This is racy as someone else might steal the reserved breakpoint
slot under us, which is undesired as the breakpoint is only
supposed to be modified, sometimes in the middle of a debugging
workflow. We don't want our slot to be stolen in the middle.

So instead of unregistering/registering the breakpoint, just
disable it while we modify its breakpoint fields and re-enable it
after if necessary.

Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Prasad <prasad@linux.vnet.ibm.com>
LKML-Reference: <1260347148-5519-1-git-send-regression-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

diff --git a/arch/x86/kernel/ptrace.c b/arch/x86/kernel/ptrace.c
index b361d28061d0f4e364a1e234b101c4dbcded94cf..7079ddaf0731e949d99940f7a036d82f3b613954 100644 (file)
--- a/arch/x86/kernel/ptrace.c
+++ b/arch/x86/kernel/ptrace.c
@@ -595,7 +595,7 @@ static unsigned long ptrace_get_dr7(struct perf_event *bp[])
        return dr7;
 }
 
-static struct perf_event *
+static int
 ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
                         struct task_struct *tsk, int disabled)
 {
@@ -609,11 +609,11 @@ ptrace_modify_breakpoint(struct perf_event *bp, int len, int type,
         * written the address register first
         */
        if (!bp)
-               return ERR_PTR(-EINVAL);
+               return -EINVAL;
 
        err = arch_bp_generic_fields(len, type, &gen_len, &gen_type);
        if (err)
-               return ERR_PTR(err);
+               return err;
 
        attr = bp->attr;
        attr.bp_len = gen_len;
@@ -658,28 +658,17 @@ restore:
                                if (!second_pass)
                                        continue;
 
-                               thread->ptrace_bps[i] = NULL;
-                               bp = ptrace_modify_breakpoint(bp, len, type,
+                               rc = ptrace_modify_breakpoint(bp, len, type,
                                                              tsk, 1);
-                               if (IS_ERR(bp)) {
-                                       rc = PTR_ERR(bp);
-                                       thread->ptrace_bps[i] = NULL;
+                               if (rc)
                                        break;
-                               }
-                               thread->ptrace_bps[i] = bp;
                        }
                        continue;
                }
 
-               bp = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
-
-               /* Incorrect bp, or we have a bug in bp API */
-               if (IS_ERR(bp)) {
-                       rc = PTR_ERR(bp);
-                       thread->ptrace_bps[i] = NULL;
+               rc = ptrace_modify_breakpoint(bp, len, type, tsk, 0);
+               if (rc)
                        break;
-               }
-               thread->ptrace_bps[i] = bp;
        }
        /*
         * Make a second pass to free the remaining unused breakpoints
@@ -737,26 +726,32 @@ static int ptrace_set_breakpoint_addr(struct task_struct *tsk, int nr,
                attr.disabled = 1;
 
                bp = register_user_hw_breakpoint(&attr, ptrace_triggered, tsk);
+
+               /*
+                * CHECKME: the previous code returned -EIO if the addr wasn't
+                * a valid task virtual addr. The new one will return -EINVAL in
+                *  this case.
+                * -EINVAL may be what we want for in-kernel breakpoints users,
+                * but -EIO looks better for ptrace, since we refuse a register
+                * writing for the user. And anyway this is the previous
+                * behaviour.
+                */
+               if (IS_ERR(bp))
+                       return PTR_ERR(bp);
+
+               t->ptrace_bps[nr] = bp;
        } else {
+               int err;
+
                bp = t->ptrace_bps[nr];
-               t->ptrace_bps[nr] = NULL;
 
                attr = bp->attr;
                attr.bp_addr = addr;
-               bp = modify_user_hw_breakpoint(bp, &attr);
+               err = modify_user_hw_breakpoint(bp, &attr);
+               if (err)
+                       return err;
        }
-       /*
-        * CHECKME: the previous code returned -EIO if the addr wasn't a
-        * valid task virtual addr. The new one will return -EINVAL in this
-        * case.
-        * -EINVAL may be what we want for in-kernel breakpoints users, but
-        * -EIO looks better for ptrace, since we refuse a register writing
-        * for the user. And anyway this is the previous behaviour.
-        */
-       if (IS_ERR(bp))
-               return PTR_ERR(bp);
 
-       t->ptrace_bps[nr] = bp;
 
        return 0;
 }

diff --git a/include/linux/hw_breakpoint.h b/include/linux/hw_breakpoint.h
index 42da1ce19ec0d8809839f35576ee7c8c2a865ff9..69f07a9f1277bceb76fddc3b4ff2d223b6787368 100644 (file)
--- a/include/linux/hw_breakpoint.h
+++ b/include/linux/hw_breakpoint.h
@@ -55,7 +55,7 @@ register_user_hw_breakpoint(struct perf_event_attr *attr,
                            struct task_struct *tsk);
 
 /* FIXME: only change from the attr, and don't unregister */
-extern struct perf_event *
+extern int
 modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr);
 
 /*
@@ -91,7 +91,7 @@ static inline struct perf_event *
 register_user_hw_breakpoint(struct perf_event_attr *attr,
                            perf_overflow_handler_t triggered,
                            struct task_struct *tsk)    { return NULL; }
-static inline struct perf_event *
+static inline int
 modify_user_hw_breakpoint(struct perf_event *bp,
                          struct perf_event_attr *attr) { return NULL; }
 static inline struct perf_event *

diff --git a/include/linux/perf_event.h b/include/linux/perf_event.h
index bf3329413e181681bec5bf8d775df5d52654d91f..64a53f74c9a91d140331863bbc4a1e0423fbb7ee 100644 (file)
--- a/include/linux/perf_event.h
+++ b/include/linux/perf_event.h
@@ -872,6 +872,8 @@ extern void perf_output_copy(struct perf_output_handle *handle,
                             const void *buf, unsigned int len);
 extern int perf_swevent_get_recursion_context(void);
 extern void perf_swevent_put_recursion_context(int rctx);
+extern void perf_event_enable(struct perf_event *event);
+extern void perf_event_disable(struct perf_event *event);
 #else
 static inline void
 perf_event_task_sched_in(struct task_struct *task, int cpu)            { }
@@ -902,7 +904,8 @@ static inline void perf_event_fork(struct task_struct *tsk)         { }
 static inline void perf_event_init(void)                               { }
 static inline int  perf_swevent_get_recursion_context(void)  { return -1; }
 static inline void perf_swevent_put_recursion_context(int rctx)                { }
-
+static inline void perf_event_enable(struct perf_event *event)         { }
+static inline void perf_event_disable(struct perf_event *event)                { }
 #endif
 
 #define perf_output_put(handle, x) \

diff --git a/kernel/hw_breakpoint.c b/kernel/hw_breakpoint.c
index 03a0773ac2b2c89dabe4ddd7aeb371fec94fb0d1..366eedf949c0a47380af9050463dae76431aa063 100644 (file)
--- a/kernel/hw_breakpoint.c
+++ b/kernel/hw_breakpoint.c
@@ -320,18 +320,40 @@ EXPORT_SYMBOL_GPL(register_user_hw_breakpoint);
  * @triggered: callback to trigger when we hit the breakpoint
  * @tsk: pointer to 'task_struct' of the process to which the address belongs
  */
-struct perf_event *
-modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr)
+int modify_user_hw_breakpoint(struct perf_event *bp, struct perf_event_attr *attr)
 {
-       /*
-        * FIXME: do it without unregistering
-        * - We don't want to lose our slot
-        * - If the new bp is incorrect, don't lose the older one
-        */
-       unregister_hw_breakpoint(bp);
+       u64 old_addr = bp->attr.bp_addr;
+       int old_type = bp->attr.bp_type;
+       int old_len = bp->attr.bp_len;
+       int err = 0;
+
+       perf_event_disable(bp);
+
+       bp->attr.bp_addr = attr->bp_addr;
+       bp->attr.bp_type = attr->bp_type;
+       bp->attr.bp_len = attr->bp_len;
+
+       if (attr->disabled)
+               goto end;
 
-       return perf_event_create_kernel_counter(attr, -1, bp->ctx->task->pid,
-                                               bp->overflow_handler);
+       err = arch_validate_hwbkpt_settings(bp, bp->ctx->task);
+       if (!err)
+               perf_event_enable(bp);
+
+       if (err) {
+               bp->attr.bp_addr = old_addr;
+               bp->attr.bp_type = old_type;
+               bp->attr.bp_len = old_len;
+               if (!bp->attr.disabled)
+                       perf_event_enable(bp);
+
+               return err;
+       }
+
+end:
+       bp->attr.disabled = attr->disabled;
+
+       return 0;
 }
 EXPORT_SYMBOL_GPL(modify_user_hw_breakpoint);
 

diff --git a/kernel/perf_event.c b/kernel/perf_event.c
index fd43ff4ac860077150b0ecfdd0566e1a99c50082..3b0cf86eee845bbec95c497a9037d0ae458bdf48 100644 (file)
--- a/kernel/perf_event.c
+++ b/kernel/perf_event.c
@@ -567,7 +567,7 @@ static void __perf_event_disable(void *info)
  * is the current context on this CPU and preemption is disabled,
  * hence we can't get into perf_event_task_sched_out for this context.
  */
-static void perf_event_disable(struct perf_event *event)
+void perf_event_disable(struct perf_event *event)
 {
        struct perf_event_context *ctx = event->ctx;
        struct task_struct *task = ctx->task;
@@ -971,7 +971,7 @@ static void __perf_event_enable(void *info)
  * perf_event_for_each_child or perf_event_for_each as described
  * for perf_event_disable.
  */
-static void perf_event_enable(struct perf_event *event)
+void perf_event_enable(struct perf_event *event)
 {
        struct perf_event_context *ctx = event->ctx;
        struct task_struct *task = ctx->task;