[PATCH] kfence: free instead of ignore pool from kmemleak

[Marco Elver]

Due to recent changes to kmemleak and how memblock allocated memory is
stored in the phys object tree of kmemleak, 07313a2b29ed ("mm: kfence:
apply kmemleak_ignore_phys on early allocated pool") tried to fix KFENCE
compatibility.

KFENCE's memory can't simply be ignored, but must be freed completely
due to it being handed out on slab allocations, and the slab post-alloc
hook attempting to insert the object to the kmemleak object tree.

Without this fix, reports like the below will appear during boot, and
kmemleak is effectively rendered useless when KFENCE is enabled:

 | kmemleak: Cannot insert 0xffffff806e24f000 into the object search tree (overlaps existing)
 | CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.19.0-v8-0815+ #5
 | Hardware name: Raspberry Pi Compute Module 4 Rev 1.0 (DT)
 | Call trace:
 |  dump_backtrace.part.0+0x1dc/0x1ec
 |  show_stack+0x24/0x80
 |  dump_stack_lvl+0x8c/0xb8
 |  dump_stack+0x1c/0x38
 |  create_object.isra.0+0x490/0x4b0
 |  kmemleak_alloc+0x3c/0x50
 |  kmem_cache_alloc+0x2f8/0x450
 |  __proc_create+0x18c/0x400
 |  proc_create_reg+0x54/0xd0
 |  proc_create_seq_private+0x94/0x120
 |  init_mm_internals+0x1d8/0x248
 |  kernel_init_freeable+0x188/0x388
 |  kernel_init+0x30/0x150
 |  ret_from_fork+0x10/0x20
 | kmemleak: Kernel memory leak detector disabled
 | kmemleak: Object 0xffffff806e24d000 (size 2097152):
 | kmemleak:   comm "swapper", pid 0, jiffies 4294892296
 | kmemleak:   min_count = -1
 | kmemleak:   count = 0
 | kmemleak:   flags = 0x5
 | kmemleak:   checksum = 0
 | kmemleak:   backtrace:
 |      kmemleak_alloc_phys+0x94/0xb0
 |      memblock_alloc_range_nid+0x1c0/0x20c
 |      memblock_alloc_internal+0x88/0x100
 |      memblock_alloc_try_nid+0x148/0x1ac
 |      kfence_alloc_pool+0x44/0x6c
 |      mm_init+0x28/0x98
 |      start_kernel+0x178/0x3e8
 |      __primary_switched+0xc4/0xcc

Reported-by: Max Schulze <max.schulze@xxxxxxxxx>
Fixes: 07313a2b29ed ("mm: kfence: apply kmemleak_ignore_phys on early allocated pool")
Fixes: 0c24e061196c ("mm: kmemleak: add rbtree and store physical address for objects allocated with PA")
Signed-off-by: Marco Elver <elver@xxxxxxxxxx>
Cc: Will Deacon <will@xxxxxxxxxx>
Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
Cc: Yee Lee <yee.lee@xxxxxxxxxxxx>
---

Note: This easily reproduces on v5.19, but on 6.0-rc1 the issue is
hidden by yet more kmemleak changes, but properly freeing the pool is
the correct thing to do either way, given the post-alloc slab hooks.
---
 mm/kfence/core.c | 11 ++++++-----
 1 file changed, 6 insertions(+), 5 deletions(-)

diff --git a/mm/kfence/core.c b/mm/kfence/core.c
index c252081b11df..9e52f2b87374 100644
--- a/mm/kfence/core.c
+++ b/mm/kfence/core.c
@@ -617,12 +617,13 @@ static bool __init kfence_init_pool_early(void)
 
 	if (!addr) {
 		/*
-		 * The pool is live and will never be deallocated from this point on.
-		 * Ignore the pool object from the kmemleak phys object tree, as it would
-		 * otherwise overlap with allocations returned by kfence_alloc(), which
-		 * are registered with kmemleak through the slab post-alloc hook.
+		 * The pool is live and will never be deallocated from this
+		 * point on. Remove the pool object from the kmemleak phys
+		 * object tree, as it would otherwise overlap with allocations
+		 * returned by kfence_alloc(), which are registered with
+		 * kmemleak through the slab post-alloc hook.
 		 */
-		kmemleak_ignore_phys(__pa(__kfence_pool));
+		kmemleak_free_part_phys(__pa(__kfence_pool), KFENCE_POOL_SIZE);
 		return true;
 	}
 
-- 
2.37.1.595.g718a3a8f04-goog