Re: [PATCH] XArray: Make xa_lock_init macro
From: Waiman Long
Date: Mon Oct 23 2023 - 21:27:26 EST
On 10/23/23 04:49, Stanislaw Gruszka wrote:
On Mon, Oct 02, 2023 at 10:25:35AM +0200, Stanislaw Gruszka wrote:
Make xa_init_flags() macro to avoid false positive lockdep splats.
Friendly ping. The subject should be changed to mention xa_init_flags(),
but anything else should be done here to get it apply ?
Regards
Stanislaw
When spin_lock_init() is used inside initialization function (like
in xa_init_flags()) which can be called many times, lockdep assign
the same key to different locks.
For example this splat is seen with intel_vpu driver which uses
two xarrays and has two separate xa_init_flags() calls:
[ 1139.148679] WARNING: inconsistent lock state
[ 1139.152941] 6.6.0-hardening.1+ #2 Tainted: G OE
[ 1139.158758] --------------------------------
[ 1139.163024] inconsistent {HARDIRQ-ON-W} -> {IN-HARDIRQ-W} usage.
[ 1139.169018] kworker/10:1/109 [HC1[1]:SC0[0]:HE0:SE1] takes:
[ 1139.174576] ffff888137237150 (&xa->xa_lock#18){?.+.}-{2:2}, at: ivpu_mmu_user_context_mark_invalid+0x1c/0x80 [intel_vpu]
[ 1139.185438] {HARDIRQ-ON-W} state was registered at:
[ 1139.190305] lock_acquire+0x1a3/0x4a0
[ 1139.194055] _raw_spin_lock+0x2c/0x40
[ 1139.197800] ivpu_submit_ioctl+0xf0b/0x3520 [intel_vpu]
[ 1139.203114] drm_ioctl_kernel+0x201/0x3f0 [drm]
[ 1139.207791] drm_ioctl+0x47d/0xa20 [drm]
[ 1139.211846] __x64_sys_ioctl+0x12e/0x1a0
[ 1139.215849] do_syscall_64+0x59/0x90
[ 1139.219509] entry_SYSCALL_64_after_hwframe+0x6e/0xd8
[ 1139.224636] irq event stamp: 45500
[ 1139.228037] hardirqs last enabled at (45499): [<ffffffff92ef0314>] _raw_spin_unlock_irq+0x24/0x50
[ 1139.236961] hardirqs last disabled at (45500): [<ffffffff92eadf8f>] common_interrupt+0xf/0x90
[ 1139.245457] softirqs last enabled at (44956): [<ffffffff92ef3430>] __do_softirq+0x4c0/0x712
[ 1139.253862] softirqs last disabled at (44461): [<ffffffff907df310>] irq_exit_rcu+0xa0/0xd0
[ 1139.262098]
other info that might help us debug this:
[ 1139.268604] Possible unsafe locking scenario:
[ 1139.274505] CPU0
[ 1139.276955] ----
[ 1139.279403] lock(&xa->xa_lock#18);
[ 1139.282978] <Interrupt>
[ 1139.285601] lock(&xa->xa_lock#18);
[ 1139.289345]
*** DEADLOCK ***
Lockdep falsely identified xa_lock from two different xarrays as the same
lock and report deadlock. More detailed description of the problem
is provided in commit c21f11d182c2 ("drm: fix drmm_mutex_init()")
Signed-off-by: Stanislaw Gruszka <stanislaw.gruszka@xxxxxxxxxxxxxxx>
---
include/linux/xarray.h | 17 +++++++----------
1 file changed, 7 insertions(+), 10 deletions(-)
diff --git a/include/linux/xarray.h b/include/linux/xarray.h
index cb571dfcf4b1..409d9d739ee9 100644
--- a/include/linux/xarray.h
+++ b/include/linux/xarray.h
@@ -375,12 +375,12 @@ void xa_destroy(struct xarray *);
*
* Context: Any context.
*/
-static inline void xa_init_flags(struct xarray *xa, gfp_t flags)
-{
- spin_lock_init(&xa->xa_lock);
- xa->xa_flags = flags;
- xa->xa_head = NULL;
-}
+#define xa_init_flags(_xa, _flags) \
+do { \
+ spin_lock_init(&(_xa)->xa_lock);\
+ (_xa)->xa_flags = (_flags); \
+ (_xa)->xa_head = NULL; \
+} while (0)
/**
* xa_init() - Initialise an empty XArray.
@@ -390,10 +390,7 @@ static inline void xa_init_flags(struct xarray *xa, gfp_t flags)
*
* Context: Any context.
*/
-static inline void xa_init(struct xarray *xa)
-{
- xa_init_flags(xa, 0);
-}
+#define xa_init(xa) xa_init_flags(xa, 0)
/**
* xa_empty() - Determine if an array has any present entries.
--
2.25.1
LGTM. However, it is up to Matthew to take it or not as he is the XArray
maintainer.
Acked-by: Waiman Long <longman@xxxxxxxxxx>