Hi,

When collecting userland stacktraces with dt(4), the probe may run in an
interrupt routine at IPL_VM or higher. Resolving pagefaults is a very
bad idea in that case. It might panic or corrupt UVM data structures.
Userland stack frames cannot be trusted, a bogus frame pointer may force
us into a page fault.

The easiest solution is to simply disable pagefault handling when
walking userland stacks.


 - Christian

---
 sys/arch/amd64/amd64/db_trace.c | 2 ++
 sys/arch/i386/i386/db_trace.c   | 2 ++
 2 files changed, 4 insertions(+)

diff --git a/sys/arch/amd64/amd64/db_trace.c b/sys/arch/amd64/amd64/db_trace.c
index 9c2ec0fc7d07..8a22b510976d 100644
--- a/sys/arch/amd64/amd64/db_trace.c
+++ b/sys/arch/amd64/amd64/db_trace.c
@@ -302,6 +302,7 @@ stacktrace_save_utrace(struct stacktrace *st)
 	KASSERT(INKERNEL(frame));
 	f = *frame;
 
+	curcpu()->ci_inatomic++;
 	while (st->st_count < STACKTRACE_MAX) {
 		if (f.f_retaddr != 0 && !INKERNEL(f.f_retaddr))
 			st->st_pc[st->st_count++] = f.f_retaddr;
@@ -322,6 +323,7 @@ stacktrace_save_utrace(struct stacktrace *st)
 		if (copyin(frame, &f, sizeof(f)) != 0)
 			break;
 	}
+	curcpu()->ci_inatomic--;
 }
 
 vaddr_t
diff --git a/sys/arch/i386/i386/db_trace.c b/sys/arch/i386/i386/db_trace.c
index 69b55f2f4618..c021d8a7a5fa 100644
--- a/sys/arch/i386/i386/db_trace.c
+++ b/sys/arch/i386/i386/db_trace.c
@@ -301,6 +301,7 @@ stacktrace_save_utrace(struct stacktrace *st)
 	KASSERT(INKERNEL(frame));
 	f = *frame;
 
+	curcpu()->ci_inatomic++;
 	while (st->st_count < STACKTRACE_MAX) {
 		if (f.f_retaddr != 0 && !INKERNEL(f.f_retaddr))
 			st->st_pc[st->st_count++] = f.f_retaddr;
@@ -321,6 +322,7 @@ stacktrace_save_utrace(struct stacktrace *st)
 		if (copyin(frame, &f, sizeof(f)) != 0)
 			break;
 	}
+	curcpu()->ci_inatomic--;
 }
 
 vaddr_t
-- 
2.34.1

• smime.p7s (5K)