Skip to contentSkip to conversation

mm: kmsan: remove runtime checks from kmsan_unpoison_memory()

PATCH Superseded Alexander Potapenko 2 participants 3 messages Jan 24, 2024

Subsystems (2)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Alexander Potapenko

author maintainer
#0 lore raw

PATCH 1/1 mm: kmsan: remove runtime checks from kmsan_unpoison_memory()

Similarly to what's been done in commit ff444efbbb9be ("kmsan: allow using __msan_instrument_asm_store() inside runtime"), it should be safe to call kmsan_unpoison_memory() from within the runtime, as it does not allocate memory or take locks. Remove the redundant runtime checks.

This should fix false positives seen with CONFIG_DEBUG_LIST=y when the non-instrumented lib/stackdepot.c failed to unpoison the memory chunks later checked by the instrumented lib/list_debug.c

Also replace the implementation of kmsan_unpoison_entry_regs() with a call to kmsan_unpoison_memory().

Cc Marco Elver <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Ilya Leoshkevich <[email protected]>
Cc Nicholas Miehlbradt <[email protected]>
---
 mm/kmsan/hooks.c | 36 +++++++++++++-----------------------
 1 file changed, 13 insertions(+), 23 deletions(-)
mm/kmsan/hooks.c +13 -23
index 5d6e2dee5692a..8a990cbf6d670 100644
@@ -359,6 +359,12 @@ void kmsan_handle_dma_sg(struct scatterlist *sg, int nents,
 }
 
 /* Functions from kmsan-checks.h follow. */
+
+/*
+ * To create an origin, kmsan_poison_memory() unwinds the stacks and stores it
+ * into the stack depot. This may cause deadlocks if done from within KMSAN
+ * runtime, therefore we bail out if kmsan_in_runtime().
+ */
 void kmsan_poison_memory(const void *address, size_t size, gfp_t flags)
 {
 	if (!kmsan_enabled || kmsan_in_runtime())
@@ -371,47 +377,31 @@ void kmsan_poison_memory(const void *address, size_t size, gfp_t flags)
 }
 EXPORT_SYMBOL(kmsan_poison_memory);
 
+/*
+ * Unlike kmsan_poison_memory(), this function can be used from within KMSAN
+ * runtime, because it does not trigger allocations or call instrumented code.
+ */
 void kmsan_unpoison_memory(const void *address, size_t size)
 {
 	unsigned long ua_flags;
 
-	if (!kmsan_enabled || kmsan_in_runtime())
+	if (!kmsan_enabled)
 		return;
 
 	ua_flags = user_access_save();
-	kmsan_enter_runtime();
 	/* The users may want to poison/unpoison random memory. */
 	kmsan_internal_unpoison_memory((void *)address, size,
 				       KMSAN_POISON_NOCHECK);
-	kmsan_leave_runtime();
 	user_access_restore(ua_flags);
 }
 EXPORT_SYMBOL(kmsan_unpoison_memory);
 
 /*
- * Version of kmsan_unpoison_memory() that can be called from within the KMSAN
- * runtime.
- *
- * Non-instrumented IRQ entry functions receive struct pt_regs from assembly
- * code. Those regs need to be unpoisoned, otherwise using them will result in
- * false positives.
- * Using kmsan_unpoison_memory() is not an option in entry code, because the
- * return value of in_task() is inconsistent - as a result, certain calls to
- * kmsan_unpoison_memory() are ignored. kmsan_unpoison_entry_regs() ensures that
- * the registers are unpoisoned even if kmsan_in_runtime() is true in the early
- * entry code.
+ * Version of kmsan_unpoison_memory() called from IRQ entry functions.
  */
 void kmsan_unpoison_entry_regs(const struct pt_regs *regs)
 {
-	unsigned long ua_flags;
-
-	if (!kmsan_enabled)
-		return;
-
-	ua_flags = user_access_save();
-	kmsan_internal_unpoison_memory((void *)regs, sizeof(*regs),
-				       KMSAN_POISON_NOCHECK);
-	user_access_restore(ua_flags);
+	kmsan_unpoison_memory((void *)regs, sizeof(*regs);
 }
 
 void kmsan_check_memory(const void *addr, size_t size)
Signature · 1 line
2.43.0.429.g432eaa2c6b-goog
reviewer Tested-by on v1 1/1
#1 lore raw
Quoted 10 lines · Alexander Potapenko' via kasan-dev

Similarly to what's been done in commit ff444efbbb9be ("kmsan: allow using __msan_instrument_asm_store() inside runtime"), it should be safe to call kmsan_unpoison_memory() from within the runtime, as it does not allocate memory or take locks. Remove the redundant runtime checks.

This should fix false positives seen with CONFIG_DEBUG_LIST=y when the non-instrumented lib/stackdepot.c failed to unpoison the memory chunks later checked by the instrumented lib/list_debug.c

Quoted 8 lines

Also replace the implementation of kmsan_unpoison_entry_regs() with a call to kmsan_unpoison_memory().

Cc Marco Elver <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Ilya Leoshkevich <[email protected]>
Cc Nicholas Miehlbradt <[email protected]>

Nice - this fixes the false positives I've seen in testing the new stack depot changes.

But I think this version of the patch wasn't compile-tested, see below.

Quoted 63 lines
---
 mm/kmsan/hooks.c | 36 +++++++++++++-----------------------
 1 file changed, 13 insertions(+), 23 deletions(-)
mm/kmsan/hooks.c +12 -14
index 5d6e2dee5692a..8a990cbf6d670 100644
@@ -359,6 +359,12 @@ void kmsan_handle_dma_sg(struct scatterlist *sg, int nents,
 }
 
 /* Functions from kmsan-checks.h follow. */
+
+/*
+ * To create an origin, kmsan_poison_memory() unwinds the stacks and stores it
+ * into the stack depot. This may cause deadlocks if done from within KMSAN
+ * runtime, therefore we bail out if kmsan_in_runtime().
+ */
 void kmsan_poison_memory(const void *address, size_t size, gfp_t flags)
 {
        if (!kmsan_enabled || kmsan_in_runtime())
@@ -371,47 +377,31 @@ void kmsan_poison_memory(const void *address, size_t size, gfp_t flags)
 }
 EXPORT_SYMBOL(kmsan_poison_memory);
 
+/*
+ * Unlike kmsan_poison_memory(), this function can be used from within KMSAN
+ * runtime, because it does not trigger allocations or call instrumented code.
+ */
 void kmsan_unpoison_memory(const void *address, size_t size)
 {
        unsigned long ua_flags;
 
-       if (!kmsan_enabled || kmsan_in_runtime())
+       if (!kmsan_enabled)
                return;
 
        ua_flags = user_access_save();
-       kmsan_enter_runtime();
        /* The users may want to poison/unpoison random memory. */
        kmsan_internal_unpoison_memory((void *)address, size,
                                       KMSAN_POISON_NOCHECK);
-       kmsan_leave_runtime();
        user_access_restore(ua_flags);
 }
 EXPORT_SYMBOL(kmsan_unpoison_memory);
 
 /*
- * Version of kmsan_unpoison_memory() that can be called from within the KMSAN
- * runtime.
- *
- * Non-instrumented IRQ entry functions receive struct pt_regs from assembly
- * code. Those regs need to be unpoisoned, otherwise using them will result in
- * false positives.
- * Using kmsan_unpoison_memory() is not an option in entry code, because the
- * return value of in_task() is inconsistent - as a result, certain calls to
- * kmsan_unpoison_memory() are ignored. kmsan_unpoison_entry_regs() ensures that
- * the registers are unpoisoned even if kmsan_in_runtime() is true in the early
- * entry code.
+ * Version of kmsan_unpoison_memory() called from IRQ entry functions.
  */
 void kmsan_unpoison_entry_regs(const struct pt_regs *regs)
 {

mm/kmsan/hooks.c void kmsan_poison_memory(const void *address, size_t size, gfp_t flags) · line 404

Quoted 10 lines
mm/kmsan/hooks.c +1 -9
-       unsigned long ua_flags;
-
-       if (!kmsan_enabled)
-               return;
-
-       ua_flags = user_access_save();
-       kmsan_internal_unpoison_memory((void *)regs, sizeof(*regs),
-                                      KMSAN_POISON_NOCHECK);
-       user_access_restore(ua_flags);
+       kmsan_unpoison_memory((void *)regs, sizeof(*regs);

missing ')', probably:

+ kmsan_unpoison_memory((void *)regs, sizeof(*regs));

author maintainer
#2 lore raw
Quoted 22 lines · Marco Elver
Quoted 10 lines · Alexander Potapenko' via kasan-dev

Similarly to what's been done in commit ff444efbbb9be ("kmsan: allow using __msan_instrument_asm_store() inside runtime"), it should be safe to call kmsan_unpoison_memory() from within the runtime, as it does not allocate memory or take locks. Remove the redundant runtime checks.

This should fix false positives seen with CONFIG_DEBUG_LIST=y when the non-instrumented lib/stackdepot.c failed to unpoison the memory chunks later checked by the instrumented lib/list_debug.c

Quoted 8 lines

Also replace the implementation of kmsan_unpoison_entry_regs() with a call to kmsan_unpoison_memory().

Cc Marco Elver <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Ilya Leoshkevich <[email protected]>
Cc Nicholas Miehlbradt <[email protected]>
Quoted 6 lines

missing ')', probably:

+ kmsan_unpoison_memory((void *)regs, sizeof(*regs));

My bad - you are right. Thanks for catching!