Skip to contentSkip to conversation

block/blk-iocost: annotate ioc_pd_stat reads with data_race()

PATCH Applied Tao Cui 3 participants 3 messages Aug 4

Subsystems (3)
lore
Share & tools
lore

Tao Cui

author
#0 lore raw

PATCH v3 1/1 block/blk-iocost: annotate ioc_pd_stat reads with data_race()

From: Tao Cui <[email protected]>

ioc_pd_stat() reads ioc->enabled, ioc->vtime_base_rate, and iocg->last_stat without holding ioc->lock, which trips KCSAN since ioc_adjust_base_vrate() and iocg_flush_stat_upward() write those fields under ioc->lock.

Commit 35198e323001 fixed the same issue in ioc_qos_prfill() and ioc_cost_model_prfill() by adding spin_lock_irq(&ioc->lock). However, those functions read configuration parameters (qos/model) that need synchronized reads. In contrast, ioc_pd_stat() only reads stat values (vrate, usage) where stale reads are harmless, so data_race() is more appropriate — it silences the KCSAN warning without adding lock contention during high-frequency stat reads.

---

Changes in v3:
- Use data_race() instead of spin_lock_irqsave/irqrestore, since these are stat values that don't require synchronized reads — the goal is just to silence KCSAN, not to guarantee consistency. (Tejun Heo)

Changes in v2:
- Use spin_lock_irqsave/irqrestore instead of spin_lock_irq/irq, since the caller (blkcg_print_stat) may already have IRQs disabled.

---
 block/blk-iocost.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
block/blk-iocost.c +6 -6
index 8b2aeba2e1e3..bc47ed034b25 100644
@@ -3093,23 +3093,23 @@ static void ioc_pd_stat(struct blkg_policy_data *pd, struct seq_file *s)
 	struct ioc_gq *iocg = pd_to_iocg(pd);
 	struct ioc *ioc = iocg->ioc;
 
-	if (!ioc->enabled)
+	if (!data_race(ioc->enabled))
 		return;
 
 	if (iocg->level == 0) {
 		unsigned vp10k = DIV64_U64_ROUND_CLOSEST(
-			ioc->vtime_base_rate * 10000,
+			data_race(ioc->vtime_base_rate) * 10000,
 			VTIME_PER_USEC);
 		seq_printf(s, " cost.vrate=%u.%02u", vp10k / 100, vp10k % 100);
 	}
 
-	seq_printf(s, " cost.usage=%llu", iocg->last_stat.usage_us);
+	seq_printf(s, " cost.usage=%llu", data_race(iocg->last_stat.usage_us));
 
 	if (blkcg_debug_stats)
 		seq_printf(s, " cost.wait=%llu cost.indebt=%llu cost.indelay=%llu",
-			iocg->last_stat.wait_us,
-			iocg->last_stat.indebt_us,
-			iocg->last_stat.indelay_us);
+			data_race(iocg->last_stat.wait_us),
+			data_race(iocg->last_stat.indebt_us),
+			data_race(iocg->last_stat.indelay_us));
 }
 
 static u64 ioc_weight_prfill(struct seq_file *sf, struct blkg_policy_data *pd,
Signature · 1 line
2.43.0
maintainer Acked-by on v3 1/1
#1 lore raw
Quoted 15 lines · Tao Cui

From: Tao Cui <[email protected]>

ioc_pd_stat() reads ioc->enabled, ioc->vtime_base_rate, and iocg->last_stat without holding ioc->lock, which trips KCSAN since ioc_adjust_base_vrate() and iocg_flush_stat_upward() write those fields under ioc->lock.

Commit 35198e323001 fixed the same issue in ioc_qos_prfill() and ioc_cost_model_prfill() by adding spin_lock_irq(&ioc->lock). However, those functions read configuration parameters (qos/model) that need synchronized reads. In contrast, ioc_pd_stat() only reads stat values (vrate, usage) where stale reads are harmless, so data_race() is more appropriate — it silences the KCSAN warning without adding lock contention during high-frequency stat reads.

Quoted 1 line

Thanks.

Signature · 1 line
tejun
maintainer Applied
#2 lore raw
In reply to Tao Cui
Quoted 13 lines · Tao Cui

ioc_pd_stat() reads ioc->enabled, ioc->vtime_base_rate, and iocg->last_stat without holding ioc->lock, which trips KCSAN since ioc_adjust_base_vrate() and iocg_flush_stat_upward() write those fields under ioc->lock.

Commit 35198e323001 fixed the same issue in ioc_qos_prfill() and ioc_cost_model_prfill() by adding spin_lock_irq(&ioc->lock). However, those functions read configuration parameters (qos/model) that need synchronized reads. In contrast, ioc_pd_stat() only reads stat values (vrate, usage) where stale reads are harmless, so data_race() is more appropriate — it silences the KCSAN warning without adding lock contention during high-frequency stat reads.

Quoted 1 line

[...]

Applied, thanks!

[1/1] block/blk-iocost: annotate ioc_pd_stat reads with data_race()

      commit: 4d73bf0ca4fbe7f252154ce98d6693c6c198164c

Best regards,

Signature · 1 line
Jens Axboe