Skip to contentSkip to conversation

lib/stackdepot: allow optional init and stack_table allocation by kvmalloc()

PATCH Superseded Vlastimil Babka 4 participants 7 messages Oct 7, 2021 – Oct 12, 2021

Subsystems (7)

v1 of 3 Superseded Latest is v3 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 lib/stackdepot: allow optional init and stack_table allocation by kvmalloc()

Currently, enabling CONFIG_STACKDEPOT means its stack_table will be allocated from memblock, even if stack depot ends up not actually used. The default size of stack_table is 4MB on 32-bit, 8MB on 64-bit.

This is fine for use-cases such as KASAN which is also a config option and has overhead on its own. But it's an issue for functionality that has to be actually enabled on boot (page_owner) or depends on hardware (GPU drivers) and thus the memory might be wasted. This was raised as an issue when trying to add stackdepot support for SLUB's debug object tracking functionality. It's common to build kernels with CONFIG_SLUB_DEBUG and enable slub_debug on boot only when needed, or create specific kmem caches with debugging for testing purposes.

It would thus be more efficient if stackdepot's table was allocated only when actually going to be used. This patch thus makes the allocation (and whole stack_depot_init() call) optional:

- Add a CONFIG_STACKDEPOT_ALWAYS_INIT flag to keep using the current well-defined point of allocation as part of mem_init(). Make CONFIG_KASAN select this flag.
- Other users have to call stack_depot_init() as part of their own init when it's determined that stack depot will actually be used. This may depend on both config and runtime conditions. Convert current users which are page_owner and several in the DRM subsystem. Same will be done for SLUB later.
- Because the init might now be called after the boot-time memblock allocation has given all memory to the buddy allocator, change stack_depot_init() to allocate stack_table with kvmalloc() when memblock is no longer available. Also handle allocation failure by disabling stackdepot (could have theoretically happened even with memblock allocation previously), and don't unnecessarily align the memblock allocation to its own size anymore.

[1] https://lore.kernel.org/all/CAMuHMdW=eoVzM1Re5FVoEN87nKfiLmM2+Ah7eNu2KXEhCvbZyA@mail.gmail.com/

Cc Marco Elver <[email protected]>
Cc Vijayanand Jitta <[email protected]>
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Thomas Zimmermann <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Andrey Ryabinin <[email protected]>
Cc Alexander Potapenko <[email protected]>
Cc Andrey Konovalov <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Cc Oliver Glitta <[email protected]>
Cc Imran Khan <[email protected]>

---
Hi, I'd appreciate review of the DRM parts - namely that I've got correctly that stack_depot_init() is called from the proper init functions and iff stack_depot_save() is going to be used later. Thanks!

 drivers/gpu/drm/drm_dp_mst_topology.c   |  1 +
 drivers/gpu/drm/drm_mm.c                |  4 ++++
 drivers/gpu/drm/i915/intel_runtime_pm.c |  3 +++
 include/linux/stackdepot.h              | 19 ++++++++-------
 init/main.c                             |  3 ++-
 lib/Kconfig                             |  3 +++
 lib/Kconfig.kasan                       |  1 +
 lib/stackdepot.c                        | 32 +++++++++++++++++++++----
 mm/page_owner.c                         |  2 ++
 9 files changed, 53 insertions(+), 15 deletions(-)
drivers/gpu/drm/drm_dp_mst_topology.c +1 -0
index 2d1adab9e360..bbe972d59dae 100644
@@ -5490,6 +5490,7 @@ int drm_dp_mst_topology_mgr_init(struct drm_dp_mst_topology_mgr *mgr,
 	mutex_init(&mgr->probe_lock);
 #if IS_ENABLED(CONFIG_DRM_DEBUG_DP_MST_TOPOLOGY_REFS)
 	mutex_init(&mgr->topology_ref_history_lock);
+	stack_depot_init();
 #endif
 	INIT_LIST_HEAD(&mgr->tx_msg_downq);
 	INIT_LIST_HEAD(&mgr->destroy_port_list);
drivers/gpu/drm/drm_mm.c +4 -0
index 7d1c578388d3..8257f9d4f619 100644
@@ -980,6 +980,10 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
 	add_hole(&mm->head_node);
 
 	mm->scan_active = 0;
+
+#ifdef CONFIG_DRM_DEBUG_MM
+	stack_depot_init();
+#endif
 }
 EXPORT_SYMBOL(drm_mm_init);
 
drivers/gpu/drm/i915/intel_runtime_pm.c +3 -0
index 0d85f3c5c526..806c32ab410b 100644
@@ -68,6 +68,9 @@ static noinline depot_stack_handle_t __save_depot_stack(void)
 static void init_intel_runtime_pm_wakeref(struct intel_runtime_pm *rpm)
 {
 	spin_lock_init(&rpm->debug.lock);
+
+	if (rpm->available)
+		stack_depot_init();
 }
 
 static noinline depot_stack_handle_t
include/linux/stackdepot.h +10 -9
index c34b55a6e554..60ba99a43745 100644
@@ -15,6 +15,16 @@
 
 typedef u32 depot_stack_handle_t;
 
+/*
+ * Every user of stack depot has to call this during its own init when it's
+ * decided that it will be calling stack_depot_save() later.
+ *
+ * The alternative is to select STACKDEPOT_ALWAYS_INIT to have stack depot
+ * enabled as part of mm_init(), for subsystems where it's known at compile time
+ * that stack depot will be used.
+ */
+int stack_depot_init(void);
+
 depot_stack_handle_t __stack_depot_save(unsigned long *entries,
 					unsigned int nr_entries,
 					gfp_t gfp_flags, bool can_alloc);
@@ -30,13 +40,4 @@ int stack_depot_snprint(depot_stack_handle_t handle, char *buf, size_t size,
 
 void stack_depot_print(depot_stack_handle_t stack);
 
-#ifdef CONFIG_STACKDEPOT
-int stack_depot_init(void);
-#else
-static inline int stack_depot_init(void)
-{
-	return 0;
-}
-#endif	/* CONFIG_STACKDEPOT */
-
 #endif
init/main.c +2 -1
index ee4d3e1b3eb9..b6a5833d98f5 100644
@@ -844,7 +844,8 @@ static void __init mm_init(void)
 	init_mem_debugging_and_hardening();
 	kfence_alloc_pool();
 	report_meminit();
-	stack_depot_init();
+	if (IS_ENABLED(CONFIG_STACKDEPOT_ALWAYS_INIT))
+		stack_depot_init();
 	mem_init();
 	mem_init_print_info();
 	/* page_owner must be initialized after buddy is ready */
lib/Kconfig +3 -0
index 5e7165e6a346..df6bcf0a4cc3 100644
@@ -671,6 +671,9 @@ config STACKDEPOT
 	bool
 	select STACKTRACE
 
+config STACKDEPOT_ALWAYS_INIT
+	bool
+
 config STACK_HASH_ORDER
 	int "stack depot hash size (12 => 4KB, 20 => 1024KB)"
 	range 12 20
lib/Kconfig.kasan +1 -0
index cdc842d090db..695deb603c66 100644
@@ -39,6 +39,7 @@ menuconfig KASAN
 		   HAVE_ARCH_KASAN_HW_TAGS
 	depends on (SLUB && SYSFS) || (SLAB && !DEBUG_SLAB)
 	select STACKDEPOT
+	select STACKDEPOT_ALWAYS_INIT
 	help
 	  Enables KASAN (KernelAddressSANitizer) - runtime memory debugger,
 	  designed to find out-of-bounds accesses and use-after-free bugs.
lib/stackdepot.c +27 -5
index b437ae79aca1..a4f449ccd0dc 100644
@@ -23,6 +23,7 @@
 #include <linux/jhash.h>
 #include <linux/kernel.h>
 #include <linux/mm.h>
+#include <linux/mutex.h>
 #include <linux/percpu.h>
 #include <linux/printk.h>
 #include <linux/slab.h>
@@ -145,6 +146,7 @@ depot_alloc_stack(unsigned long *entries, int size, u32 hash, void **prealloc)
 #define STACK_HASH_MASK (STACK_HASH_SIZE - 1)
 #define STACK_HASH_SEED 0x9747b28c
 
+DEFINE_MUTEX(stack_depot_init_mutex);
 static bool stack_depot_disable;
 static struct stack_record **stack_table;
 
@@ -161,18 +163,38 @@ static int __init is_stack_depot_disabled(char *str)
 }
 early_param("stack_depot_disable", is_stack_depot_disabled);
 
-int __init stack_depot_init(void)
+/*
+ * __ref because of memblock_alloc(), which will not be actually called after
+ * the __init code is gone
+ */
+__ref int stack_depot_init(void)
 {
-	if (!stack_depot_disable) {
+	mutex_lock(&stack_depot_init_mutex);
+	if (!stack_depot_disable && stack_table == NULL) {
 		size_t size = (STACK_HASH_SIZE * sizeof(struct stack_record *));
 		int i;
 
-		stack_table = memblock_alloc(size, size);
-		for (i = 0; i < STACK_HASH_SIZE;  i++)
-			stack_table[i] = NULL;
+		if (slab_is_available()) {
+			pr_info("Stack Depot allocating hash table with kvmalloc\n");
+			stack_table = kvmalloc(size, GFP_KERNEL);
+		} else {
+			pr_info("Stack Depot allocating hash table with memblock_alloc\n");
+			stack_table = memblock_alloc(size, SMP_CACHE_BYTES);
+		}
+		if (stack_table) {
+			for (i = 0; i < STACK_HASH_SIZE;  i++)
+				stack_table[i] = NULL;
+		} else {
+			pr_err("Stack Depot failed hash table allocationg, disabling\n");
+			stack_depot_disable = true;
+			mutex_unlock(&stack_depot_init_mutex);
+			return -ENOMEM;
+		}
 	}
+	mutex_unlock(&stack_depot_init_mutex);
 	return 0;
 }
+EXPORT_SYMBOL_GPL(stack_depot_init);
 
 /* Calculate hash for a stack */
 static inline u32 hash_stack(unsigned long *entries, unsigned int size)
mm/page_owner.c +2 -0
index a83f546c06b5..a48607b51a97 100644
@@ -80,6 +80,8 @@ static void init_page_owner(void)
 	if (!page_owner_enabled)
 		return;
 
+	stack_depot_init();
+
 	register_dummy_stack();
 	register_failure_stack();
 	register_early_stack();
Signature · 1 line
2.33.0
reviewer Acked-by on v1 1/1
#1 lore raw
Quoted 35 lines · Vlastimil Babka

Currently, enabling CONFIG_STACKDEPOT means its stack_table will be allocated from memblock, even if stack depot ends up not actually used. The default size of stack_table is 4MB on 32-bit, 8MB on 64-bit.

This is fine for use-cases such as KASAN which is also a config option and has overhead on its own. But it's an issue for functionality that has to be actually enabled on boot (page_owner) or depends on hardware (GPU drivers) and thus the memory might be wasted. This was raised as an issue when trying to add stackdepot support for SLUB's debug object tracking functionality. It's common to build kernels with CONFIG_SLUB_DEBUG and enable slub_debug on boot only when needed, or create specific kmem caches with debugging for testing purposes.

It would thus be more efficient if stackdepot's table was allocated only when actually going to be used. This patch thus makes the allocation (and whole stack_depot_init() call) optional:

- Add a CONFIG_STACKDEPOT_ALWAYS_INIT flag to keep using the current well-defined point of allocation as part of mem_init(). Make CONFIG_KASAN select this flag.
- Other users have to call stack_depot_init() as part of their own init when it's determined that stack depot will actually be used. This may depend on both config and runtime conditions. Convert current users which are page_owner and several in the DRM subsystem. Same will be done for SLUB later.
- Because the init might now be called after the boot-time memblock allocation has given all memory to the buddy allocator, change stack_depot_init() to allocate stack_table with kvmalloc() when memblock is no longer available. Also handle allocation failure by disabling stackdepot (could have theoretically happened even with memblock allocation previously), and don't unnecessarily align the memblock allocation to its own size anymore.

[1] https://lore.kernel.org/all/CAMuHMdW=eoVzM1Re5FVoEN87nKfiLmM2+Ah7eNu2KXEhCvbZyA@mail.gmail.com/

Quoted 15 lines
Cc Marco Elver <[email protected]>
Cc Vijayanand Jitta <[email protected]>
Cc Maarten Lankhorst <[email protected]>
Cc Maxime Ripard <[email protected]>
Cc Thomas Zimmermann <[email protected]>
Cc David Airlie <[email protected]>
Cc Daniel Vetter <[email protected]>
Cc Andrey Ryabinin <[email protected]>
Cc Alexander Potapenko <[email protected]>
Cc Andrey Konovalov <[email protected]>
Cc Dmitry Vyukov <[email protected]>
Cc Geert Uytterhoeven <[email protected]>
Cc Oliver Glitta <[email protected]>
Cc Imran Khan <[email protected]>
Quoted 216 lines

---
Hi, I'd appreciate review of the DRM parts - namely that I've got correctly that stack_depot_init() is called from the proper init functions and iff stack_depot_save() is going to be used later. Thanks!

 drivers/gpu/drm/drm_dp_mst_topology.c   |  1 +
 drivers/gpu/drm/drm_mm.c                |  4 ++++
 drivers/gpu/drm/i915/intel_runtime_pm.c |  3 +++
 include/linux/stackdepot.h              | 19 ++++++++-------
 init/main.c                             |  3 ++-
 lib/Kconfig                             |  3 +++
 lib/Kconfig.kasan                       |  1 +
 lib/stackdepot.c                        | 32 +++++++++++++++++++++----
 mm/page_owner.c                         |  2 ++
 9 files changed, 53 insertions(+), 15 deletions(-)
drivers/gpu/drm/drm_dp_mst_topology.c +1 -0
index 2d1adab9e360..bbe972d59dae 100644
@@ -5490,6 +5490,7 @@ int drm_dp_mst_topology_mgr_init(struct drm_dp_mst_topology_mgr *mgr,
        mutex_init(&mgr->probe_lock);
 #if IS_ENABLED(CONFIG_DRM_DEBUG_DP_MST_TOPOLOGY_REFS)
        mutex_init(&mgr->topology_ref_history_lock);
+       stack_depot_init();
 #endif
        INIT_LIST_HEAD(&mgr->tx_msg_downq);
        INIT_LIST_HEAD(&mgr->destroy_port_list);
drivers/gpu/drm/drm_mm.c +4 -0
index 7d1c578388d3..8257f9d4f619 100644
@@ -980,6 +980,10 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
        add_hole(&mm->head_node);
 
        mm->scan_active = 0;
+
+#ifdef CONFIG_DRM_DEBUG_MM
+       stack_depot_init();
+#endif
 }
 EXPORT_SYMBOL(drm_mm_init);
 
drivers/gpu/drm/i915/intel_runtime_pm.c +3 -0
index 0d85f3c5c526..806c32ab410b 100644
@@ -68,6 +68,9 @@ static noinline depot_stack_handle_t __save_depot_stack(void)
 static void init_intel_runtime_pm_wakeref(struct intel_runtime_pm *rpm)
 {
        spin_lock_init(&rpm->debug.lock);
+
+       if (rpm->available)
+               stack_depot_init();
 }
 
 static noinline depot_stack_handle_t
include/linux/stackdepot.h +10 -9
index c34b55a6e554..60ba99a43745 100644
@@ -15,6 +15,16 @@
 
 typedef u32 depot_stack_handle_t;
 
+/*
+ * Every user of stack depot has to call this during its own init when it's
+ * decided that it will be calling stack_depot_save() later.
+ *
+ * The alternative is to select STACKDEPOT_ALWAYS_INIT to have stack depot
+ * enabled as part of mm_init(), for subsystems where it's known at compile time
+ * that stack depot will be used.
+ */
+int stack_depot_init(void);
+
 depot_stack_handle_t __stack_depot_save(unsigned long *entries,
                                        unsigned int nr_entries,
                                        gfp_t gfp_flags, bool can_alloc);
@@ -30,13 +40,4 @@ int stack_depot_snprint(depot_stack_handle_t handle, char *buf, size_t size,
 
 void stack_depot_print(depot_stack_handle_t stack);
 
-#ifdef CONFIG_STACKDEPOT
-int stack_depot_init(void);
-#else
-static inline int stack_depot_init(void)
-{
-       return 0;
-}
-#endif /* CONFIG_STACKDEPOT */
-
 #endif
init/main.c +2 -1
index ee4d3e1b3eb9..b6a5833d98f5 100644
@@ -844,7 +844,8 @@ static void __init mm_init(void)
        init_mem_debugging_and_hardening();
        kfence_alloc_pool();
        report_meminit();
-       stack_depot_init();
+       if (IS_ENABLED(CONFIG_STACKDEPOT_ALWAYS_INIT))
+               stack_depot_init();
        mem_init();
        mem_init_print_info();
        /* page_owner must be initialized after buddy is ready */
lib/Kconfig +3 -0
index 5e7165e6a346..df6bcf0a4cc3 100644
@@ -671,6 +671,9 @@ config STACKDEPOT
        bool
        select STACKTRACE
 
+config STACKDEPOT_ALWAYS_INIT
+       bool
+
 config STACK_HASH_ORDER
        int "stack depot hash size (12 => 4KB, 20 => 1024KB)"
        range 12 20
lib/Kconfig.kasan +1 -0
index cdc842d090db..695deb603c66 100644
@@ -39,6 +39,7 @@ menuconfig KASAN
                   HAVE_ARCH_KASAN_HW_TAGS
        depends on (SLUB && SYSFS) || (SLAB && !DEBUG_SLAB)
        select STACKDEPOT
+       select STACKDEPOT_ALWAYS_INIT
        help
          Enables KASAN (KernelAddressSANitizer) - runtime memory debugger,
          designed to find out-of-bounds accesses and use-after-free bugs.
lib/stackdepot.c +27 -5
index b437ae79aca1..a4f449ccd0dc 100644
@@ -23,6 +23,7 @@
 #include <linux/jhash.h>
 #include <linux/kernel.h>
 #include <linux/mm.h>
+#include <linux/mutex.h>
 #include <linux/percpu.h>
 #include <linux/printk.h>
 #include <linux/slab.h>
@@ -145,6 +146,7 @@ depot_alloc_stack(unsigned long *entries, int size, u32 hash, void **prealloc)
 #define STACK_HASH_MASK (STACK_HASH_SIZE - 1)
 #define STACK_HASH_SEED 0x9747b28c
 
+DEFINE_MUTEX(stack_depot_init_mutex);
 static bool stack_depot_disable;
 static struct stack_record **stack_table;
 
@@ -161,18 +163,38 @@ static int __init is_stack_depot_disabled(char *str)
 }
 early_param("stack_depot_disable", is_stack_depot_disabled);
 
-int __init stack_depot_init(void)
+/*
+ * __ref because of memblock_alloc(), which will not be actually called after
+ * the __init code is gone
+ */
+__ref int stack_depot_init(void)
 {
-       if (!stack_depot_disable) {
+       mutex_lock(&stack_depot_init_mutex);
+       if (!stack_depot_disable && stack_table == NULL) {
                size_t size = (STACK_HASH_SIZE * sizeof(struct stack_record *));
                int i;
 
-               stack_table = memblock_alloc(size, size);
-               for (i = 0; i < STACK_HASH_SIZE;  i++)
-                       stack_table[i] = NULL;
+               if (slab_is_available()) {
+                       pr_info("Stack Depot allocating hash table with kvmalloc\n");
+                       stack_table = kvmalloc(size, GFP_KERNEL);
+               } else {
+                       pr_info("Stack Depot allocating hash table with memblock_alloc\n");
+                       stack_table = memblock_alloc(size, SMP_CACHE_BYTES);
+               }
+               if (stack_table) {
+                       for (i = 0; i < STACK_HASH_SIZE;  i++)
+                               stack_table[i] = NULL;
+               } else {
+                       pr_err("Stack Depot failed hash table allocationg, disabling\n");
+                       stack_depot_disable = true;
+                       mutex_unlock(&stack_depot_init_mutex);
+                       return -ENOMEM;
+               }
        }
+       mutex_unlock(&stack_depot_init_mutex);
        return 0;
 }
+EXPORT_SYMBOL_GPL(stack_depot_init);
 
 /* Calculate hash for a stack */
 static inline u32 hash_stack(unsigned long *entries, unsigned int size)
mm/page_owner.c +2 -1
index a83f546c06b5..a48607b51a97 100644
@@ -80,6 +80,8 @@ static void init_page_owner(void)
        if (!page_owner_enabled)
                return;
 
+       stack_depot_init();
+
        register_dummy_stack();
        register_failure_stack();
        register_early_stack();
--

2.33.0

--
You received this message because you are subscribed to the Google Groups "kasan-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To view this discussion on the web visit https://groups.google.com/d/msgid/kasan-dev/20211007095815.3563-1-vbabka%40suse.cz.

#2 lore raw
In reply to Vlastimil Babka

On Thu, Oct 07, 2021 at 11:58AM +0200, Vlastimil Babka wrote:
[...]

Quoted 14 lines

- Add a CONFIG_STACKDEPOT_ALWAYS_INIT flag to keep using the current well-defined point of allocation as part of mem_init(). Make CONFIG_KASAN select this flag.
- Other users have to call stack_depot_init() as part of their own init when it's determined that stack depot will actually be used. This may depend on both config and runtime conditions. Convert current users which are page_owner and several in the DRM subsystem. Same will be done for SLUB later.
- Because the init might now be called after the boot-time memblock allocation has given all memory to the buddy allocator, change stack_depot_init() to allocate stack_table with kvmalloc() when memblock is no longer available. Also handle allocation failure by disabling stackdepot (could have theoretically happened even with memblock allocation previously), and don't unnecessarily align the memblock allocation to its own size anymore.

...

Quoted 3 lines

Hi, I'd appreciate review of the DRM parts - namely that I've got correctly that stack_depot_init() is called from the proper init functions and iff stack_depot_save() is going to be used later. Thanks!

For ease of review between stackdepot and DRM changes, I thought it'd be nice to split into 2 patches, but not sure it'll work, because you're changing the semantics of the normal STACKDEPOT.

One option would be to flip it around, and instead have STACKDEPOT_LAZY_INIT, but that seems counter-intuitive if the majority of STACKDEPOT users are LAZY_INIT users.

On the other hand, the lazy initialization mode you're introducing requires an explicit stack_depot_init() call somewhere and isn't as straightforward as before.

Not sure what is best. My intuition tells me STACKDEPOT_LAZY_INIT would be safer as it's a deliberate opt-in to the lazy initialization behaviour.

Preferences?

[...]

drivers/gpu/drm/drm_mm.c void drm_mm_init(struct drm_mm *mm, u64 start, u64 size) · line 986

Quoted 10 lines
drivers/gpu/drm/drm_mm.c +4 -0
@@ -980,6 +980,10 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
 	add_hole(&mm->head_node);
 
 	mm->scan_active = 0;
+
+#ifdef CONFIG_DRM_DEBUG_MM
+	stack_depot_init();
+#endif

DRM_DEBUG_MM implies STACKDEPOT. Not sure what is more readable to drm maintainers, but perhaps it'd be nicer to avoid the #ifdef here, and instead just keep the no-op version of stack_depot_init() in <linux/stackdepot.h>. I don't have a strong preference.

Quoted 41 lines
drivers/gpu/drm/drm_mm.c +0 -0
 }
 EXPORT_SYMBOL(drm_mm_init);
 
drivers/gpu/drm/i915/intel_runtime_pm.c +3 -0
index 0d85f3c5c526..806c32ab410b 100644
@@ -68,6 +68,9 @@ static noinline depot_stack_handle_t __save_depot_stack(void)
 static void init_intel_runtime_pm_wakeref(struct intel_runtime_pm *rpm)
 {
 	spin_lock_init(&rpm->debug.lock);
+
+	if (rpm->available)
+		stack_depot_init();
 }
 
 static noinline depot_stack_handle_t
include/linux/stackdepot.h +10 -0
index c34b55a6e554..60ba99a43745 100644
@@ -15,6 +15,16 @@
 
 typedef u32 depot_stack_handle_t;
 
+/*
+ * Every user of stack depot has to call this during its own init when it's
+ * decided that it will be calling stack_depot_save() later.
+ *
+ * The alternative is to select STACKDEPOT_ALWAYS_INIT to have stack depot
+ * enabled as part of mm_init(), for subsystems where it's known at compile time
+ * that stack depot will be used.
+ */
+int stack_depot_init(void);
+
 depot_stack_handle_t __stack_depot_save(unsigned long *entries,
 					unsigned int nr_entries,
 					gfp_t gfp_flags, bool can_alloc);
@@ -30,13 +40,4 @@ int stack_depot_snprint(depot_stack_handle_t handle, char *buf, size_t size,
 
 void stack_depot_print(depot_stack_handle_t stack);

include/linux/stackdepot.h int stack_depot_snprint(depot_stack_handle_t handle, char *buf, size_t size, · line 41

Quoted 10 lines
include/linux/stackdepot.h +0 -9
 
-#ifdef CONFIG_STACKDEPOT
-int stack_depot_init(void);
-#else
-static inline int stack_depot_init(void)
-{
-	return 0;
-}
-#endif	/* CONFIG_STACKDEPOT */
-

Could we avoid the IS_ENABLED() in init/main.c by adding a wrapper here:

+#ifdef CONFIG_STACKDEPOT_ALWAYS_INIT
+static inline int stack_depot_early_init(void) { return stack_depot_init(); }
+#else
+static inline int stack_depot_early_init(void) { return 0; }
+#endif /* CONFIG_STACKDEPOT_ALWAYS_INIT */

init/main.c static void __init mm_init(void) · line 848

Quoted 12 lines
include/linux/stackdepot.h +0 -0
 #endif
init/main.c +2 -1
index ee4d3e1b3eb9..b6a5833d98f5 100644
@@ -844,7 +844,8 @@ static void __init mm_init(void)
 	init_mem_debugging_and_hardening();
 	kfence_alloc_pool();
 	report_meminit();
-	stack_depot_init();
+	if (IS_ENABLED(CONFIG_STACKDEPOT_ALWAYS_INIT))
+		stack_depot_init();

I'd push the decision of when to call this into <linux/stackdepot.h> via wrapper stack_depot_early_init().

lib/Kconfig config STACKDEPOT · line 675

Quoted 13 lines
init/main.c +0 -0
 	mem_init();
 	mem_init_print_info();
 	/* page_owner must be initialized after buddy is ready */
lib/Kconfig +2 -0
index 5e7165e6a346..df6bcf0a4cc3 100644
@@ -671,6 +671,9 @@ config STACKDEPOT
 	bool
 	select STACKTRACE
 
+config STACKDEPOT_ALWAYS_INIT
+	bool

It looks like every users of STACKDEPOT_ALWAYS_INIT will also select STACKDEPOT, so we could just make this:

+config STACKDEPOT_ALWAYS_INIT
+ bool
+ select STACKDEPOT

And remove the redundant 'select STACKDEPOT' in Kconfig.kasan.

Quoted 12 lines
lib/Kconfig +0 -0
 config STACK_HASH_ORDER
 	int "stack depot hash size (12 => 4KB, 20 => 1024KB)"
 	range 12 20
lib/Kconfig.kasan +1 -0
index cdc842d090db..695deb603c66 100644
@@ -39,6 +39,7 @@ menuconfig KASAN
 		   HAVE_ARCH_KASAN_HW_TAGS
 	depends on (SLUB && SYSFS) || (SLAB && !DEBUG_SLAB)
 	select STACKDEPOT
+	select STACKDEPOT_ALWAYS_INIT

[...]

lib/Kconfig.kasan

Quoted 5 lines
lib/Kconfig.kasan +3 -1
-int __init stack_depot_init(void)
+/*
+ * __ref because of memblock_alloc(), which will not be actually called after
+ * the __init code is gone

The reason is that after __init code is gone, slab_is_available() will be true (might be worth adding to the comment).

lib/Kconfig.kasan

Quoted 7 lines

+ */

lib/Kconfig.kasan +3 -1
+__ref int stack_depot_init(void)
 {
-	if (!stack_depot_disable) {
+	mutex_lock(&stack_depot_init_mutex);
+	if (!stack_depot_disable && stack_table == NULL) {
 		size_t size = (STACK_HASH_SIZE * sizeof(struct stack_record *));

Thanks,
-- Marco

#3 lore raw
Quoted 16 lines · Marco Elver

On Thu, Oct 07, 2021 at 11:58AM +0200, Vlastimil Babka wrote:
[...]

Quoted 6 lines

- Add a CONFIG_STACKDEPOT_ALWAYS_INIT flag to keep using the current well-defined point of allocation as part of mem_init(). Make CONFIG_KASAN select this flag.
- Other users have to call stack_depot_init() as part of their own init when it's determined that stack depot will actually be used. This may depend on both config and runtime conditions. Convert current users which are

Quoted 8 lines

page_owner and several in the DRM subsystem. Same will be done for SLUB later.
- Because the init might now be called after the boot-time memblock allocation has given all memory to the buddy allocator, change stack_depot_init() to allocate stack_table with kvmalloc() when memblock is no longer available. Also handle allocation failure by disabling stackdepot (could have theoretically happened even with memblock allocation previously), and don't unnecessarily align the memblock allocation to its own size anymore.

Quoted 8 lines

...

Quoted 3 lines

Hi, I'd appreciate review of the DRM parts - namely that I've got correctly that stack_depot_init() is called from the proper init functions and iff stack_depot_save() is going to be used later. Thanks!

For ease of review between stackdepot and DRM changes, I thought it'd be nice to split into 2 patches, but not sure it'll work, because you're changing the semantics of the normal STACKDEPOT.

Yeah, that's why it's a single patch. As the DRM parts are clearly separated to their files, I think review should be fine.

Quoted 3 lines

One option would be to flip it around, and instead have STACKDEPOT_LAZY_INIT, but that seems counter-intuitive if the majority of STACKDEPOT users are LAZY_INIT users.

Agree.

Quoted 7 lines

On the other hand, the lazy initialization mode you're introducing requires an explicit stack_depot_init() call somewhere and isn't as straightforward as before.

Not sure what is best. My intuition tells me STACKDEPOT_LAZY_INIT would be safer as it's a deliberate opt-in to the lazy initialization behaviour.

I think it should be fine with ALWAYS_INIT. There are not many stackdepot users being added, and anyone developing a new one will very quickly find out if they forget to call stack_depot_init()?

Quoted 8 lines

Preferences?

[...]

Quoted 5 lines
drivers/gpu/drm/drm_mm.c +0 -0
@@ -980,6 +980,10 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
 	add_hole(&mm->head_node);
 
Quoted 10 lines

drivers/gpu/drm/drm_mm.c void drm_mm_init(struct drm_mm *mm, u64 start, u64 size) · line 986

Quoted 5 lines
drivers/gpu/drm/drm_mm.c +4 -0
 	mm->scan_active = 0;
+
+#ifdef CONFIG_DRM_DEBUG_MM
+	stack_depot_init();
+#endif

DRM_DEBUG_MM implies STACKDEPOT. Not sure what is more readable to drm maintainers, but perhaps it'd be nicer to avoid the #ifdef here, and instead just keep the no-op version of stack_depot_init() in <linux/stackdepot.h>. I don't have a strong preference.

Hm, but in case STACKDEPOT is also selected by something else (e.g. CONFIG_PAGE_OWNER) which uses lazy init but isn't enabled on boot, then without #ifdef CONFIG_DRM_DEBUG_MM above, this code would call a stack_depot_init() (that's not a no-op) even in case it's not going to be using it, so not what we want to achieve.
But it could be changed to use IS_ENABLED() if that's preferred by DRM folks.

BTW it's possible that there won't be any DRM review because this failed to apply:
https://patchwork.freedesktop.org/series/95549/
DRM folks, any hint how to indicate that the base was next-20211001?

Quoted 16 lines

drivers/gpu/drm/drm_mm.c int stack_depot_snprint(depot_stack_handle_t handle, char *buf, size_t size, · line 41

Quoted 13 lines
drivers/gpu/drm/drm_mm.c +0 -9
@@ -30,13 +40,4 @@ int stack_depot_snprint(depot_stack_handle_t handle, char *buf, size_t size,
 
 void stack_depot_print(depot_stack_handle_t stack);
 
-#ifdef CONFIG_STACKDEPOT
-int stack_depot_init(void);
-#else
-static inline int stack_depot_init(void)
-{
-	return 0;
-}
-#endif	/* CONFIG_STACKDEPOT */
-

Could we avoid the IS_ENABLED() in init/main.c by adding a wrapper here:

Review

Quoted 5 lines
(quoted diff) +5 -0
+#ifdef CONFIG_STACKDEPOT_ALWAYS_INIT
+static inline int stack_depot_early_init(void)	{ return stack_depot_init(); }
+#else
+static inline int stack_depot_early_init(void)	{ return 0; }
+#endif	/* CONFIG_STACKDEPOT_ALWAYS_INIT */

We could, but it's a wrapper made for only a single caller...

Quoted 7 lines
Quoted 7 lines
drivers/gpu/drm/drm_mm.c +0 -0
 #endif
init/main.c +0 -0
index ee4d3e1b3eb9..b6a5833d98f5 100644
@@ -844,7 +844,8 @@ static void __init mm_init(void)
 	init_mem_debugging_and_hardening();
Quoted 8 lines

init/main.c static void __init mm_init(void) · line 848

Quoted 5 lines
init/main.c +2 -1
 	kfence_alloc_pool();
 	report_meminit();
-	stack_depot_init();
+	if (IS_ENABLED(CONFIG_STACKDEPOT_ALWAYS_INIT))
+		stack_depot_init();

I'd push the decision of when to call this into <linux/stackdepot.h> via wrapper stack_depot_early_init().

No strong preferrences, if you think it's worth it.

Quoted 21 lines

lib/Kconfig config STACKDEPOT · line 675

Quoted 13 lines
init/main.c +0 -0
 	mem_init();
 	mem_init_print_info();
 	/* page_owner must be initialized after buddy is ready */
lib/Kconfig +2 -0
index 5e7165e6a346..df6bcf0a4cc3 100644
@@ -671,6 +671,9 @@ config STACKDEPOT
 	bool
 	select STACKTRACE
 
+config STACKDEPOT_ALWAYS_INIT
+	bool

It looks like every users of STACKDEPOT_ALWAYS_INIT will also select STACKDEPOT, so we could just make this:

lib/Kconfig +3 -0
+config STACKDEPOT_ALWAYS_INIT
+	bool
+	select STACKDEPOT
Quoted 1 line

And remove the redundant 'select STACKDEPOT' in Kconfig.kasan.

Right, will do, if KConfig resolver doesn't bite me.

Quoted 15 lines
Quoted 12 lines
lib/Kconfig +0 -0
 config STACK_HASH_ORDER
 	int "stack depot hash size (12 => 4KB, 20 => 1024KB)"
 	range 12 20
lib/Kconfig.kasan +1 -0
index cdc842d090db..695deb603c66 100644
@@ -39,6 +39,7 @@ menuconfig KASAN
 		   HAVE_ARCH_KASAN_HW_TAGS
 	depends on (SLUB && SYSFS) || (SLAB && !DEBUG_SLAB)
 	select STACKDEPOT
+	select STACKDEPOT_ALWAYS_INIT

[...]

Quoted 1 line
Quoted 7 lines

lib/Kconfig.kasan

Quoted 4 lines
lib/Kconfig.kasan +3 -1
-int __init stack_depot_init(void)
+/*
+ * __ref because of memblock_alloc(), which will not be actually called after
+ * the __init code is gone

The reason is that after __init code is gone, slab_is_available() will be true (might be worth adding to the comment).

OK

Thanks for the review!

#4 lore raw

On Mon, 11 Oct 2021 at 19:02, Vlastimil Babka <[email protected]> wrote:
[...]

Quoted 11 lines
Quoted 7 lines

On the other hand, the lazy initialization mode you're introducing requires an explicit stack_depot_init() call somewhere and isn't as straightforward as before.

Not sure what is best. My intuition tells me STACKDEPOT_LAZY_INIT would be safer as it's a deliberate opt-in to the lazy initialization behaviour.

I think it should be fine with ALWAYS_INIT. There are not many stackdepot users being added, and anyone developing a new one will very quickly find out if they forget to call stack_depot_init()?

I think that's fine.

Quoted 19 lines
Quoted 8 lines

Preferences?

[...]

Quoted 5 lines
drivers/gpu/drm/drm_mm.c +0 -0
@@ -980,6 +980,10 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
     add_hole(&mm->head_node);
 
Quoted 10 lines

drivers/gpu/drm/drm_mm.c void drm_mm_init(struct drm_mm *mm, u64 start, u64 size) · line 986

Quoted 5 lines
drivers/gpu/drm/drm_mm.c +4 -0
     mm->scan_active = 0;
+
+#ifdef CONFIG_DRM_DEBUG_MM
+    stack_depot_init();
+#endif

DRM_DEBUG_MM implies STACKDEPOT. Not sure what is more readable to drm maintainers, but perhaps it'd be nicer to avoid the #ifdef here, and instead just keep the no-op version of stack_depot_init() in <linux/stackdepot.h>. I don't have a strong preference.

Quoted 6 lines

Hm, but in case STACKDEPOT is also selected by something else (e.g. CONFIG_PAGE_OWNER) which uses lazy init but isn't enabled on boot, then without #ifdef CONFIG_DRM_DEBUG_MM above, this code would call a stack_depot_init() (that's not a no-op) even in case it's not going to be using it, so not what we want to achieve.
But it could be changed to use IS_ENABLED() if that's preferred by DRM folks.

You're right -- but I'll leave this to DRM folks.

Quoted 5 lines

BTW it's possible that there won't be any DRM review because this failed to apply:
https://patchwork.freedesktop.org/series/95549/
DRM folks, any hint how to indicate that the base was next-20211001?

[...]

Quoted 24 lines

Review

Quoted 5 lines
(quoted diff) +5 -0
+#ifdef CONFIG_STACKDEPOT_ALWAYS_INIT
+static inline int stack_depot_early_init(void)       { return stack_depot_init(); }
+#else
+static inline int stack_depot_early_init(void)       { return 0; }
+#endif       /* CONFIG_STACKDEPOT_ALWAYS_INIT */

We could, but it's a wrapper made for only a single caller...

Quoted 7 lines
Quoted 7 lines

#endif

init/main.c +0 -0
index ee4d3e1b3eb9..b6a5833d98f5 100644
@@ -844,7 +844,8 @@ static void __init mm_init(void)
     init_mem_debugging_and_hardening();
Quoted 8 lines

init/main.c static void __init mm_init(void) · line 848

Quoted 5 lines
init/main.c +2 -1
     kfence_alloc_pool();
     report_meminit();
-    stack_depot_init();
+    if (IS_ENABLED(CONFIG_STACKDEPOT_ALWAYS_INIT))
+            stack_depot_init();

I'd push the decision of when to call this into <linux/stackdepot.h> via wrapper stack_depot_early_init().

Quoted 1 line

No strong preferrences, if you think it's worth it.

All the other *init() functions seem to follow the same idiom as there are barely any IS_ENABLED() in init/main.c. So I think it's just about being consistent with the rest.

Thanks,
-- Marco

#5 lore raw
Quoted 35 lines · Marco Elver

On Mon, 11 Oct 2021 at 19:02, Vlastimil Babka <[email protected]> wrote:
[...]

Quoted 11 lines
Quoted 7 lines

On the other hand, the lazy initialization mode you're introducing requires an explicit stack_depot_init() call somewhere and isn't as straightforward as before.

Not sure what is best. My intuition tells me STACKDEPOT_LAZY_INIT would be safer as it's a deliberate opt-in to the lazy initialization behaviour.

I think it should be fine with ALWAYS_INIT. There are not many stackdepot users being added, and anyone developing a new one will very quickly find out if they forget to call stack_depot_init()?

I think that's fine.

Quoted 19 lines
Quoted 18 lines

Preferences?

[...]

drivers/gpu/drm/drm_mm.c void drm_mm_init(struct drm_mm *mm, u64 start, u64 size) · line 986

Quoted 10 lines
drivers/gpu/drm/drm_mm.c +4 -0
@@ -980,6 +980,10 @@ void drm_mm_init(struct drm_mm *mm, u64 start, u64 size)
     add_hole(&mm->head_node);
 
     mm->scan_active = 0;
+
+#ifdef CONFIG_DRM_DEBUG_MM
+    stack_depot_init();
+#endif

DRM_DEBUG_MM implies STACKDEPOT. Not sure what is more readable to drm maintainers, but perhaps it'd be nicer to avoid the #ifdef here, and instead just keep the no-op version of stack_depot_init() in <linux/stackdepot.h>. I don't have a strong preference.

Quoted 8 lines
Quoted 6 lines

Hm, but in case STACKDEPOT is also selected by something else (e.g. CONFIG_PAGE_OWNER) which uses lazy init but isn't enabled on boot, then without #ifdef CONFIG_DRM_DEBUG_MM above, this code would call a stack_depot_init() (that's not a no-op) even in case it's not going to be using it, so not what we want to achieve.
But it could be changed to use IS_ENABLED() if that's preferred by DRM folks.

You're right -- but I'll leave this to DRM folks.

Ah, the file only includes stackdepot.h in a #ifdef CONFIG_DRM_DEBUG_MM section so I will keep the #ifdef here for a minimal change, unless requested otherwise.

#6 lore raw

[Intel-gfx] ✗ Fi.CI.BUILD: failure for lib/stackdepot: allow optional init and stack_table allocation by kvmalloc()

In reply to Vlastimil Babka
== Series Details ==

== Series Details ==

Series: lib/stackdepot: allow optional init and stack_table allocation by kvmalloc()
URL : https://patchwork.freedesktop.org/series/95549/
State : failure

== Summary ==

Applying: lib/stackdepot: allow optional init and stack_table allocation by kvmalloc()
error: sha1 information is lacking or useless (include/linux/stackdepot.h).
error: could not build fake ancestor
hint: Use 'git am --show-current-patch=diff' to see the failed patch
Patch failed at 0001 lib/stackdepot: allow optional init and stack_table allocation by kvmalloc() When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".