mirror of https://lore.kernel.org/lkml/
 help / color / mirror / Atom feed
* [PATCH net v5] libcxgb: use raw_cpu_ptr() in ppm_get_cpu_entries()
@ 2026-11-19  9:40 Haoran Wang
  2026-11-19  9:41 ` netdev-bot+sinfo
  0 siblings, 1 reply; 2+ messages in thread
From: Haoran Wang @ 2026-11-19  9:40 UTC (permalink / raw)
  To: David Laight
  Cc: Sebastian Andrzej Siewior, netdev, linux-kernel, linux-rt-devel,
	andrew+netdev, davem, edumazet, kuba, pabeni, rostedt,
	wanghr2026

ppm_get_cpu_entries() must not acquire a sleeping lock while preemption
is disabled: get_cpu() disables preemption, and spin_lock_bh() may sleep
on PREEMPT_RT kernels.  Use raw_cpu_ptr() instead, which reads the
current CPU's per-CPU pointer without touching preempt_count.

The task is then free to migrate before taking pool->lock, so it may end
up allocating from a different CPU's pool.  That is harmless: each
per-CPU pool is just a shard of the global page pod (ppod) index space,
and every read-modify-write of a pool's bitmap and next index happens
under pool->lock.  The free path, ppm_unmark_entries(), re-derives the
owning pool from the global index rather than from the running CPU and
takes that same lock, so any thread can service any shard.  Derive the
global index base from pool->base so it always matches the pool that
was locked.

Found by static analysis and not triggered.  The bug is only reachable
on PREEMPT_RT kernels, where spin_lock_bh() is a sleeping lock: there it
would trigger a "scheduling while atomic" BUG under
CONFIG_DEBUG_ATOMIC_SLEEP, or a deadlock if pool->lock is contended.

Fixes: 47da7e62b391 ("cxgb4: add iSCSI DDP page pod manager")
Signed-off-by: Haoran Wang <wanghr2026@mail.zgclab.edu.cn>

---
Changes since v4:
- Reword the commit message (review).
- pr_debug(): drop the running CPU that David suggested in v3; it is of
  limited use, so keep only the pool-owning CPU.
- Add a Fixes: tag and record how the issue was found (netdev bot).

Changes since v3:
- Explain in the commit message why a foreign CPU's pool is harmless and
  trim the PREEMPT_RT problem description (review).
- pr_debug(): label the pool-owning CPU and also print the running CPU.

Changes since v2:
- Keep the pool-owning CPU in the pr_debug() output, derived from
  pool->base (suggested by Sebastian).
- Reword the commit message; spell PREEMPT_RT.

Changes since v1:
- Read the global-index base from pool->base (initialised at pool
  allocation) instead of a cpu captured before the lock.

 drivers/net/ethernet/chelsio/libcxgb/libcxgb_ppm.c |   14 ++++++--------
 1 file changed, 6 insertions(+), 8 deletions(-)

diff --git a/drivers/net/ethernet/chelsio/libcxgb/libcxgb_ppm.c b/drivers/net/ethernet/chelsio/libcxgb/libcxgb_ppm.c
index 2e3973a32..837bdf412 100644
--- a/drivers/net/ethernet/chelsio/libcxgb/libcxgb_ppm.c
+++ b/drivers/net/ethernet/chelsio/libcxgb/libcxgb_ppm.c
@@ -119,16 +119,13 @@ static int ppm_get_cpu_entries(struct cxgbi_ppm *ppm, unsigned int count,
 			       unsigned long caller_data)
 {
 	struct cxgbi_ppm_pool *pool;
-	unsigned int cpu;
 	int i;
 
 	if (!ppm->pool)
 		return -EINVAL;
 
-	cpu = get_cpu();
-	pool = per_cpu_ptr(ppm->pool, cpu);
+	pool = raw_cpu_ptr(ppm->pool);
 	spin_lock_bh(&pool->lock);
-	put_cpu();
 
 	i = ppm_find_unused_entries(pool->bmap, ppm->pool_index_max,
 				    pool->next, count, 0);
@@ -144,11 +141,11 @@ static int ppm_get_cpu_entries(struct cxgbi_ppm *ppm, unsigned int count,
 
 	spin_unlock_bh(&pool->lock);
 
-	pr_debug("%s: cpu %u, idx %d + %d (%d), next %u.\n",
-		 __func__, cpu, i, count, i + cpu * ppm->pool_index_max,
-		pool->next);
+	pr_debug("%s: pool cpu %u, idx %d + %d (%d), next %u.\n",
+		 __func__, pool->base / ppm->pool_index_max,
+		 i, count, i + pool->base, pool->next);
 
-	i += cpu * ppm->pool_index_max;
+	i += pool->base;
 	ppm_mark_entries(ppm, i, count, caller_data);
 
 	return i;
@@ -375,6 +372,7 @@ ppm_alloc_cpu_pool(unsigned int *total, unsigned int *pcpu_ppmax)
 
 		memset(ppool, 0, alloc_sz);
 		spin_lock_init(&ppool->lock);
+		ppool->base = count;
 		count += ppmax;
 	}
 


^ permalink raw reply	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-19  9:42 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-11-19  9:40 [PATCH net v5] libcxgb: use raw_cpu_ptr() in ppm_get_cpu_entries() Haoran Wang
2026-11-19  9:41 ` netdev-bot+sinfo

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox

all inboxes | Powered by JetHome®