[PATCH v7 net-next 00/15] locking: Introduce nested-BH locking.

[PATCH v7 net-next 00/15] locking: Introduce nested-BH locking.

[Sebastian Andrzej Siewior]

Disabling bottoms halves acts as per-CPU BKL. On PREEMPT_RT code within
local_bh_disable() section remains preemtible. As a result high prior
tasks (or threaded interrupts) will be blocked by lower-prio task (or
threaded interrupts) which are long running which includes softirq
sections.

The proposed way out is to introduce explicit per-CPU locks for
resources which are protected by local_bh_disable() and use those only
on PREEMPT_RT so there is no additional overhead for !PREEMPT_RT builds.

The series introduces the infrastructure and converts large parts of
networking which is largest stake holder here. Once this done the
per-CPU lock from local_bh_disable() on PREEMPT_RT can be lifted.

Performance testing. Baseline is net-next as of commit 93bda33046e7a
("Merge branch'net-constify-ctl_table-arguments-of-utility-functions'")
plus v6.10-rc1. A 10GiG link is used between two hosts. The command
   xdp-bench redirect-cpu --cpu 3 --remote-action drop eth1 -e

was invoked on the receiving side with a ixgbe. The sending side uses
pktgen_sample03_burst_single_flow.sh on i40e.

Baseline:
| eth1->?                 9,018,604 rx/s                  0 err,drop/s
|   receive total         9,018,604 pkt/s                 0 drop/s                0 error/s
|     cpu:7               9,018,604 pkt/s                 0 drop/s                0 error/s
|   enqueue to cpu 3      9,018,602 pkt/s                 0 drop/s             7.00 bulk-avg
|     cpu:7->3            9,018,602 pkt/s                 0 drop/s             7.00 bulk-avg
|   kthread total         9,018,606 pkt/s                 0 drop/s          214,698 sched
|     cpu:3               9,018,606 pkt/s                 0 drop/s          214,698 sched
|     xdp_stats                   0 pass/s        9,018,606 drop/s                0 redir/s
|       cpu:3                     0 pass/s        9,018,606 drop/s                0 redir/s
|   redirect_err                  0 error/s
|   xdp_exception                 0 hit/s

perf top --sort cpu,symbol --no-children:
|   18.14%  007  [k] bpf_prog_4f0ffbb35139c187_cpumap_l4_hash
|   13.29%  007  [k] ixgbe_poll
|   12.66%  003  [k] cpu_map_kthread_run
|    7.23%  003  [k] page_frag_free
|    6.76%  007  [k] xdp_do_redirect
|    3.76%  007  [k] cpu_map_redirect
|    3.13%  007  [k] bq_flush_to_queue
|    2.51%  003  [k] xdp_return_frame
|    1.93%  007  [k] try_to_wake_up
|    1.78%  007  [k] _raw_spin_lock
|    1.74%  007  [k] cpu_map_enqueue
|    1.56%  003  [k] bpf_prog_57cd311f2e27366b_cpumap_drop

With this series applied:
| eth1->?                10,329,340 rx/s                  0 err,drop/s
|   receive total        10,329,340 pkt/s                 0 drop/s                0 error/s
|     cpu:6              10,329,340 pkt/s                 0 drop/s                0 error/s
|   enqueue to cpu 3     10,329,338 pkt/s                 0 drop/s             8.00 bulk-avg
|     cpu:6->3           10,329,338 pkt/s                 0 drop/s             8.00 bulk-avg
|   kthread total        10,329,321 pkt/s                 0 drop/s           96,297 sched
|     cpu:3              10,329,321 pkt/s                 0 drop/s           96,297 sched
|     xdp_stats                   0 pass/s       10,329,321 drop/s                0 redir/s
|       cpu:3                     0 pass/s       10,329,321 drop/s                0 redir/s
|   redirect_err                  0 error/s
|   xdp_exception                 0 hit/s

perf top --sort cpu,symbol --no-children:
|   20.90%  006  [k] bpf_prog_4f0ffbb35139c187_cpumap_l4_hash
|   12.62%  006  [k] ixgbe_poll
|    9.82%  003  [k] page_frag_free
|    8.73%  003  [k] cpu_map_bpf_prog_run_xdp
|    6.63%  006  [k] xdp_do_redirect
|    4.94%  003  [k] cpu_map_kthread_run
|    4.28%  006  [k] cpu_map_redirect
|    4.03%  006  [k] bq_flush_to_queue
|    3.01%  003  [k] xdp_return_frame
|    1.95%  006  [k] _raw_spin_lock
|    1.94%  003  [k] bpf_prog_57cd311f2e27366b_cpumap_drop

This diff appears to be noise.

v6…v7 https://lore.kernel.org/all/20240612170303.3896084-1-bigeasy@linutronix.de/:
- The softnet_data::xmit (08/15) is moved completely for RT and not
  limited to one member. During the review it has been noticed that the
  `more' member requires same protection.
  After some cleanup this might also be considered conditionally for !RT.

v5…v6 https://lore.kernel.org/all/20240607070427.1379327-1-bigeasy@linutronix.de/:
- bpf_redirect_info and the lists (cpu, dev, xsk map_flush_list) is now
  initialized on first usage instead during setup time
  (bpf_net_ctx_set()). bpf_redirect_info::kern_flags is used as
  status to note which member require an init.
  The bpf_redirect_info::kern_flags is moved to the end of the struct
  (after nh) in order to be excluded from the memset() which clears
  everything upto the the nh member.
  This whole lazy init performed to save cycles and not to waste them to
  memset bpf_redirect_info and init the three lists on each
  net_rx_action()/ NAPI invocation even if BPF/redirect is not used.
  Suggested by Jesper Dangaard Brouer.

- Collect Acked-by/Review-by from PeterZ/ tglx

v4…v5 https://lore.kernel.org/all/20240604154425.878636-1-bigeasy@linutronix.de/:
- Remove the guard() notation as well as __free() within the patches.
  Patch #1 and #2 add the guard definition for local_lock_nested_bh()
  but it remains unused with the series.
  The __free() notation for bpf_net_ctx_clear has been removed entirely.

- Collect Toke's Reviewed-by.

v3…v4 https://lore.kernel.org/all/20240529162927.403425-1-bigeasy@linutronix.de/:
- Removed bpf_clear_redirect_map(), moved the comment to the caller.
  Suggested by Toke.

- The bpf_redirect_info structure is memset() each time it is assigned.
  Suggested by Toke.

- The bpf_net_ctx_set() in __napi_busy_loop() has been moved from the
  top of the function to begin/ end of the BH-disabled section. This has
  been done to remain in sync with other call sites.
  After adding the memset() I've been looking at the perf-numbers in my
  test-case and I haven't noticed an impact, the numbers are in the same
  range with and without the change. Therefore I kept the numbers from
  previous posting.

- Collected Alexei's Acked-by.

v2…v3 https://lore.kernel.org/all/20240503182957.1042122-1-bigeasy@linutronix.de/:
- WARN checks checks for bpf_net_ctx_get() have been dropped and all
  NULL checks around it. This means bpf_net_ctx_get_ri() assumes the
  context has been set and will segfault if it is not the case.
  Suggested by Alexei and Jesper. This should always work or always
  segfault.

- It has been suggested by Toke to embed struct bpf_net_context into
  task_struct instead just a pointer to it. This would increase the size
  of task_struct by 112 bytes instead just eight and Alexei didn't like
  it due to the size impact with 1m threads. It is a pointer again.

v1…v2 https://lore.kernel.org/all/20231215171020.687342-1-bigeasy@linutronix.de/:
- Jakub complained about touching networking drivers to make the
  additional locking work. Alexei complained about the additional
  locking within the XDP/eBFP case.
  This led to a change in how the per-CPU variables are accessed for the
  XDP/eBPF case. On PREEMPT_RT the variables are now stored on stack and
  the task pointer to the structure is saved in the task_struct while
  keeping every for !RT unchanged. This was proposed as a RFC in
  	v1: https://lore.kernel.org/all/20240213145923.2552753-1-bigeasy@linutronix.de/

  and then updated

        v2: https://lore.kernel.org/all/20240229183109.646865-1-bigeasy@linutronix.de/
	  - Renamed the container struct from xdp_storage to bpf_net_context.
            Suggested by Toke Høiland-Jørgensen.
	  - Use the container struct also on !PREEMPT_RT builds. Store the
	    pointer to the on-stack struct in a per-CPU variable. Suggested by
            Toke Høiland-Jørgensen.

  This reduces the initial queue from 24 to 15 patches.

- There were complains about the scoped_guard() which shifts the whole
  block and makes it harder to review because the whole gets removed and
  added again. The usage has been replaced with local_lock_nested_bh()+
  its unlock counterpart.

Sebastian

[PATCH v7 net-next 01/15] locking/local_lock: Introduce guard definition for local_lock.

[Sebastian Andrzej Siewior]

Introduce lock guard definition for local_lock_t. There are no users
yet.

Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 include/linux/local_lock.h | 11 +++++++++++
 1 file changed, 11 insertions(+)

diff --git a/include/linux/local_lock.h b/include/linux/local_lock.h
index e55010fa73296..82366a37f4474 100644
--- a/include/linux/local_lock.h
+++ b/include/linux/local_lock.h
@@ -51,4 +51,15 @@
 #define local_unlock_irqrestore(lock, flags)			\
 	__local_unlock_irqrestore(lock, flags)
 
+DEFINE_GUARD(local_lock, local_lock_t __percpu*,
+	     local_lock(_T),
+	     local_unlock(_T))
+DEFINE_GUARD(local_lock_irq, local_lock_t __percpu*,
+	     local_lock_irq(_T),
+	     local_unlock_irq(_T))
+DEFINE_LOCK_GUARD_1(local_lock_irqsave, local_lock_t __percpu,
+		    local_lock_irqsave(_T->lock, _T->flags),
+		    local_unlock_irqrestore(_T->lock, _T->flags),
+		    unsigned long flags)
+
 #endif
-- 
2.45.2

[PATCH v7 net-next 02/15] locking/local_lock: Add local nested BH locking infrastructure.

[Sebastian Andrzej Siewior]

Add local_lock_nested_bh() locking. It is based on local_lock_t and the
naming follows the preempt_disable_nested() example.

For !PREEMPT_RT + !LOCKDEP it is a per-CPU annotation for locking
assumptions based on local_bh_disable(). The macro is optimized away
during compilation.
For !PREEMPT_RT + LOCKDEP the local_lock_nested_bh() is reduced to
the usual lock-acquire plus lockdep_assert_in_softirq() - ensuring that
BH is disabled.

For PREEMPT_RT local_lock_nested_bh() acquires the specified per-CPU
lock. It does not disable CPU migration because it relies on
local_bh_disable() disabling CPU migration.
With LOCKDEP it performans the usual lockdep checks as with !PREEMPT_RT.
Due to include hell the softirq check has been moved spinlock.c.

The intention is to use this locking in places where locking of a per-CPU
variable relies on BH being disabled. Instead of treating disabled
bottom halves as a big per-CPU lock, PREEMPT_RT can use this to reduce
the locking scope to what actually needs protecting.
A side effect is that it also documents the protection scope of the
per-CPU variables.

Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 include/linux/local_lock.h          | 10 ++++++++++
 include/linux/local_lock_internal.h | 31 +++++++++++++++++++++++++++++
 include/linux/lockdep.h             |  3 +++
 kernel/locking/spinlock.c           |  8 ++++++++
 4 files changed, 52 insertions(+)

diff --git a/include/linux/local_lock.h b/include/linux/local_lock.h
index 82366a37f4474..091dc0b6bdfb9 100644
--- a/include/linux/local_lock.h
+++ b/include/linux/local_lock.h
@@ -62,4 +62,14 @@ DEFINE_LOCK_GUARD_1(local_lock_irqsave, local_lock_t __percpu,
 		    local_unlock_irqrestore(_T->lock, _T->flags),
 		    unsigned long flags)
 
+#define local_lock_nested_bh(_lock)				\
+	__local_lock_nested_bh(_lock)
+
+#define local_unlock_nested_bh(_lock)				\
+	__local_unlock_nested_bh(_lock)
+
+DEFINE_GUARD(local_lock_nested_bh, local_lock_t __percpu*,
+	     local_lock_nested_bh(_T),
+	     local_unlock_nested_bh(_T))
+
 #endif
diff --git a/include/linux/local_lock_internal.h b/include/linux/local_lock_internal.h
index 975e33b793a77..8dd71fbbb6d2b 100644
--- a/include/linux/local_lock_internal.h
+++ b/include/linux/local_lock_internal.h
@@ -62,6 +62,17 @@ do {								\
 	local_lock_debug_init(lock);				\
 } while (0)
 
+#define __spinlock_nested_bh_init(lock)				\
+do {								\
+	static struct lock_class_key __key;			\
+								\
+	debug_check_no_locks_freed((void *)lock, sizeof(*lock));\
+	lockdep_init_map_type(&(lock)->dep_map, #lock, &__key,  \
+			      0, LD_WAIT_CONFIG, LD_WAIT_INV,	\
+			      LD_LOCK_NORMAL);			\
+	local_lock_debug_init(lock);				\
+} while (0)
+
 #define __local_lock(lock)					\
 	do {							\
 		preempt_disable();				\
@@ -98,6 +109,15 @@ do {								\
 		local_irq_restore(flags);			\
 	} while (0)
 
+#define __local_lock_nested_bh(lock)				\
+	do {							\
+		lockdep_assert_in_softirq();			\
+		local_lock_acquire(this_cpu_ptr(lock));	\
+	} while (0)
+
+#define __local_unlock_nested_bh(lock)				\
+	local_lock_release(this_cpu_ptr(lock))
+
 #else /* !CONFIG_PREEMPT_RT */
 
 /*
@@ -138,4 +158,15 @@ typedef spinlock_t local_lock_t;
 
 #define __local_unlock_irqrestore(lock, flags)	__local_unlock(lock)
 
+#define __local_lock_nested_bh(lock)				\
+do {								\
+	lockdep_assert_in_softirq_func();			\
+	spin_lock(this_cpu_ptr(lock));				\
+} while (0)
+
+#define __local_unlock_nested_bh(lock)				\
+do {								\
+	spin_unlock(this_cpu_ptr((lock)));			\
+} while (0)
+
 #endif /* CONFIG_PREEMPT_RT */
diff --git a/include/linux/lockdep.h b/include/linux/lockdep.h
index 08b0d1d9d78b7..3f5a551579cc9 100644
--- a/include/linux/lockdep.h
+++ b/include/linux/lockdep.h
@@ -600,6 +600,8 @@ do {									\
 		     (!in_softirq() || in_irq() || in_nmi()));		\
 } while (0)
 
+extern void lockdep_assert_in_softirq_func(void);
+
 #else
 # define might_lock(lock) do { } while (0)
 # define might_lock_read(lock) do { } while (0)
@@ -613,6 +615,7 @@ do {									\
 # define lockdep_assert_preemption_enabled() do { } while (0)
 # define lockdep_assert_preemption_disabled() do { } while (0)
 # define lockdep_assert_in_softirq() do { } while (0)
+# define lockdep_assert_in_softirq_func() do { } while (0)
 #endif
 
 #ifdef CONFIG_PROVE_RAW_LOCK_NESTING
diff --git a/kernel/locking/spinlock.c b/kernel/locking/spinlock.c
index 8475a0794f8c5..438c6086d540e 100644
--- a/kernel/locking/spinlock.c
+++ b/kernel/locking/spinlock.c
@@ -413,3 +413,11 @@ notrace int in_lock_functions(unsigned long addr)
 	&& addr < (unsigned long)__lock_text_end;
 }
 EXPORT_SYMBOL(in_lock_functions);
+
+#if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_PREEMPT_RT)
+void notrace lockdep_assert_in_softirq_func(void)
+{
+	lockdep_assert_in_softirq();
+}
+EXPORT_SYMBOL(lockdep_assert_in_softirq_func);
+#endif
-- 
2.45.2

[PATCH v7 net-next 03/15] net: Use __napi_alloc_frag_align() instead of open coding it.

[Sebastian Andrzej Siewior]

The else condition within __netdev_alloc_frag_align() is an open coded
__napi_alloc_frag_align().

Use __napi_alloc_frag_align() instead of open coding it.
Move fragsz assignment before page_frag_alloc_align() invocation because
__napi_alloc_frag_align() also contains this statement.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 net/core/skbuff.c | 8 ++------
 1 file changed, 2 insertions(+), 6 deletions(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index c8ac79851cd67..656b298255c5f 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -318,19 +318,15 @@ void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
 {
 	void *data;
 
-	fragsz = SKB_DATA_ALIGN(fragsz);
 	if (in_hardirq() || irqs_disabled()) {
 		struct page_frag_cache *nc = this_cpu_ptr(&netdev_alloc_cache);
 
+		fragsz = SKB_DATA_ALIGN(fragsz);
 		data = __page_frag_alloc_align(nc, fragsz, GFP_ATOMIC,
 					       align_mask);
 	} else {
-		struct napi_alloc_cache *nc;
-
 		local_bh_disable();
-		nc = this_cpu_ptr(&napi_alloc_cache);
-		data = __page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC,
-					       align_mask);
+		data = __napi_alloc_frag_align(fragsz, align_mask);
 		local_bh_enable();
 	}
 	return data;
-- 
2.45.2

[PATCH v7 net-next 04/15] net: Use nested-BH locking for napi_alloc_cache.

[Sebastian Andrzej Siewior]

napi_alloc_cache is a per-CPU variable and relies on disabled BH for its
locking. Without per-CPU locking in local_bh_disable() on PREEMPT_RT
this data structure requires explicit locking.

Add a local_lock_t to the data structure and use local_lock_nested_bh()
for locking. This change adds only lockdep coverage and does not alter
the functional behaviour for !PREEMPT_RT.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 net/core/skbuff.c | 29 ++++++++++++++++++++++++-----
 1 file changed, 24 insertions(+), 5 deletions(-)

diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 656b298255c5f..0c13cfa20658f 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -277,6 +277,7 @@ static void *page_frag_alloc_1k(struct page_frag_1k *nc, gfp_t gfp_mask)
 #endif
 
 struct napi_alloc_cache {
+	local_lock_t bh_lock;
 	struct page_frag_cache page;
 	struct page_frag_1k page_small;
 	unsigned int skb_count;
@@ -284,7 +285,9 @@ struct napi_alloc_cache {
 };
 
 static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
-static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
+static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache) = {
+	.bh_lock = INIT_LOCAL_LOCK(bh_lock),
+};
 
 /* Double check that napi_get_frags() allocates skbs with
  * skb->head being backed by slab, not a page fragment.
@@ -306,11 +309,16 @@ void napi_get_frags_check(struct napi_struct *napi)
 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask)
 {
 	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
+	void *data;
 
 	fragsz = SKB_DATA_ALIGN(fragsz);
 
-	return __page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC,
+	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
+	data = __page_frag_alloc_align(&nc->page, fragsz, GFP_ATOMIC,
 				       align_mask);
+	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
+	return data;
+
 }
 EXPORT_SYMBOL(__napi_alloc_frag_align);
 
@@ -338,16 +346,20 @@ static struct sk_buff *napi_skb_cache_get(void)
 	struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
 	struct sk_buff *skb;
 
+	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
 	if (unlikely(!nc->skb_count)) {
 		nc->skb_count = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache,
 						      GFP_ATOMIC,
 						      NAPI_SKB_CACHE_BULK,
 						      nc->skb_cache);
-		if (unlikely(!nc->skb_count))
+		if (unlikely(!nc->skb_count)) {
+			local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
 			return NULL;
+		}
 	}
 
 	skb = nc->skb_cache[--nc->skb_count];
+	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
 	kasan_mempool_unpoison_object(skb, kmem_cache_size(net_hotdata.skbuff_cache));
 
 	return skb;
@@ -740,9 +752,13 @@ struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
 		pfmemalloc = nc->pfmemalloc;
 	} else {
 		local_bh_disable();
+		local_lock_nested_bh(&napi_alloc_cache.bh_lock);
+
 		nc = this_cpu_ptr(&napi_alloc_cache.page);
 		data = page_frag_alloc(nc, len, gfp_mask);
 		pfmemalloc = nc->pfmemalloc;
+
+		local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
 		local_bh_enable();
 	}
 
@@ -806,11 +822,11 @@ struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len)
 		goto skb_success;
 	}
 
-	nc = this_cpu_ptr(&napi_alloc_cache);
-
 	if (sk_memalloc_socks())
 		gfp_mask |= __GFP_MEMALLOC;
 
+	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
+	nc = this_cpu_ptr(&napi_alloc_cache);
 	if (NAPI_HAS_SMALL_PAGE_FRAG && len <= SKB_WITH_OVERHEAD(1024)) {
 		/* we are artificially inflating the allocation size, but
 		 * that is not as bad as it may look like, as:
@@ -832,6 +848,7 @@ struct sk_buff *napi_alloc_skb(struct napi_struct *napi, unsigned int len)
 		data = page_frag_alloc(&nc->page, len, gfp_mask);
 		pfmemalloc = nc->page.pfmemalloc;
 	}
+	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
 
 	if (unlikely(!data))
 		return NULL;
@@ -1429,6 +1446,7 @@ static void napi_skb_cache_put(struct sk_buff *skb)
 	if (!kasan_mempool_poison_object(skb))
 		return;
 
+	local_lock_nested_bh(&napi_alloc_cache.bh_lock);
 	nc->skb_cache[nc->skb_count++] = skb;
 
 	if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
@@ -1440,6 +1458,7 @@ static void napi_skb_cache_put(struct sk_buff *skb)
 				     nc->skb_cache + NAPI_SKB_CACHE_HALF);
 		nc->skb_count = NAPI_SKB_CACHE_HALF;
 	}
+	local_unlock_nested_bh(&napi_alloc_cache.bh_lock);
 }
 
 void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason)
-- 
2.45.2

[PATCH v7 net-next 05/15] net/tcp_sigpool: Use nested-BH locking for sigpool_scratch.

[Sebastian Andrzej Siewior]

sigpool_scratch is a per-CPU variable and relies on disabled BH for its
locking. Without per-CPU locking in local_bh_disable() on PREEMPT_RT
this data structure requires explicit locking.

Make a struct with a pad member (original sigpool_scratch) and a
local_lock_t and use local_lock_nested_bh() for locking. This change
adds only lockdep coverage and does not alter the functional behaviour
for !PREEMPT_RT.

Cc: David Ahern <dsahern@kernel.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 net/ipv4/tcp_sigpool.c | 17 +++++++++++++----
 1 file changed, 13 insertions(+), 4 deletions(-)

diff --git a/net/ipv4/tcp_sigpool.c b/net/ipv4/tcp_sigpool.c
index 8512cb09ebc09..d8a4f192873a2 100644
--- a/net/ipv4/tcp_sigpool.c
+++ b/net/ipv4/tcp_sigpool.c
@@ -10,7 +10,14 @@
 #include <net/tcp.h>
 
 static size_t __scratch_size;
-static DEFINE_PER_CPU(void __rcu *, sigpool_scratch);
+struct sigpool_scratch {
+	local_lock_t bh_lock;
+	void __rcu *pad;
+};
+
+static DEFINE_PER_CPU(struct sigpool_scratch, sigpool_scratch) = {
+	.bh_lock = INIT_LOCAL_LOCK(bh_lock),
+};
 
 struct sigpool_entry {
 	struct crypto_ahash	*hash;
@@ -72,7 +79,7 @@ static int sigpool_reserve_scratch(size_t size)
 			break;
 		}
 
-		old_scratch = rcu_replace_pointer(per_cpu(sigpool_scratch, cpu),
+		old_scratch = rcu_replace_pointer(per_cpu(sigpool_scratch.pad, cpu),
 					scratch, lockdep_is_held(&cpool_mutex));
 		if (!cpu_online(cpu) || !old_scratch) {
 			kfree(old_scratch);
@@ -93,7 +100,7 @@ static void sigpool_scratch_free(void)
 	int cpu;
 
 	for_each_possible_cpu(cpu)
-		kfree(rcu_replace_pointer(per_cpu(sigpool_scratch, cpu),
+		kfree(rcu_replace_pointer(per_cpu(sigpool_scratch.pad, cpu),
 					  NULL, lockdep_is_held(&cpool_mutex)));
 	__scratch_size = 0;
 }
@@ -277,7 +284,8 @@ int tcp_sigpool_start(unsigned int id, struct tcp_sigpool *c) __cond_acquires(RC
 	/* Pairs with tcp_sigpool_reserve_scratch(), scratch area is
 	 * valid (allocated) until tcp_sigpool_end().
 	 */
-	c->scratch = rcu_dereference_bh(*this_cpu_ptr(&sigpool_scratch));
+	local_lock_nested_bh(&sigpool_scratch.bh_lock);
+	c->scratch = rcu_dereference_bh(*this_cpu_ptr(&sigpool_scratch.pad));
 	return 0;
 }
 EXPORT_SYMBOL_GPL(tcp_sigpool_start);
@@ -286,6 +294,7 @@ void tcp_sigpool_end(struct tcp_sigpool *c) __releases(RCU_BH)
 {
 	struct crypto_ahash *hash = crypto_ahash_reqtfm(c->req);
 
+	local_unlock_nested_bh(&sigpool_scratch.bh_lock);
 	rcu_read_unlock_bh();
 	ahash_request_free(c->req);
 	crypto_free_ahash(hash);
-- 
2.45.2

[PATCH v7 net-next 06/15] net/ipv4: Use nested-BH locking for ipv4_tcp_sk.

[Sebastian Andrzej Siewior]

ipv4_tcp_sk is a per-CPU variable and relies on disabled BH for its
locking. Without per-CPU locking in local_bh_disable() on PREEMPT_RT
this data structure requires explicit locking.

Make a struct with a sock member (original ipv4_tcp_sk) and a
local_lock_t and use local_lock_nested_bh() for locking. This change
adds only lockdep coverage and does not alter the functional behaviour
for !PREEMPT_RT.

Cc: David Ahern <dsahern@kernel.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 include/net/sock.h  |  5 +++++
 net/ipv4/tcp_ipv4.c | 15 +++++++++++----
 2 files changed, 16 insertions(+), 4 deletions(-)

diff --git a/include/net/sock.h b/include/net/sock.h
index b30ea0c342a65..cce23ac4d5148 100644
--- a/include/net/sock.h
+++ b/include/net/sock.h
@@ -544,6 +544,11 @@ struct sock {
 	netns_tracker		ns_tracker;
 };
 
+struct sock_bh_locked {
+	struct sock *sock;
+	local_lock_t bh_lock;
+};
+
 enum sk_pacing {
 	SK_PACING_NONE		= 0,
 	SK_PACING_NEEDED	= 1,
diff --git a/net/ipv4/tcp_ipv4.c b/net/ipv4/tcp_ipv4.c
index de0c8f43448ab..318bdd9b8f27c 100644
--- a/net/ipv4/tcp_ipv4.c
+++ b/net/ipv4/tcp_ipv4.c
@@ -93,7 +93,9 @@ static int tcp_v4_md5_hash_hdr(char *md5_hash, const struct tcp_md5sig_key *key,
 struct inet_hashinfo tcp_hashinfo;
 EXPORT_SYMBOL(tcp_hashinfo);
 
-static DEFINE_PER_CPU(struct sock *, ipv4_tcp_sk);
+static DEFINE_PER_CPU(struct sock_bh_locked, ipv4_tcp_sk) = {
+	.bh_lock = INIT_LOCAL_LOCK(bh_lock),
+};
 
 static u32 tcp_v4_init_seq(const struct sk_buff *skb)
 {
@@ -882,7 +884,9 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb,
 	arg.tos = ip_hdr(skb)->tos;
 	arg.uid = sock_net_uid(net, sk && sk_fullsock(sk) ? sk : NULL);
 	local_bh_disable();
-	ctl_sk = this_cpu_read(ipv4_tcp_sk);
+	local_lock_nested_bh(&ipv4_tcp_sk.bh_lock);
+	ctl_sk = this_cpu_read(ipv4_tcp_sk.sock);
+
 	sock_net_set(ctl_sk, net);
 	if (sk) {
 		ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
@@ -907,6 +911,7 @@ static void tcp_v4_send_reset(const struct sock *sk, struct sk_buff *skb,
 	sock_net_set(ctl_sk, &init_net);
 	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
 	__TCP_INC_STATS(net, TCP_MIB_OUTRSTS);
+	local_unlock_nested_bh(&ipv4_tcp_sk.bh_lock);
 	local_bh_enable();
 
 #ifdef CONFIG_TCP_MD5SIG
@@ -1002,7 +1007,8 @@ static void tcp_v4_send_ack(const struct sock *sk,
 	arg.tos = tos;
 	arg.uid = sock_net_uid(net, sk_fullsock(sk) ? sk : NULL);
 	local_bh_disable();
-	ctl_sk = this_cpu_read(ipv4_tcp_sk);
+	local_lock_nested_bh(&ipv4_tcp_sk.bh_lock);
+	ctl_sk = this_cpu_read(ipv4_tcp_sk.sock);
 	sock_net_set(ctl_sk, net);
 	ctl_sk->sk_mark = (sk->sk_state == TCP_TIME_WAIT) ?
 			   inet_twsk(sk)->tw_mark : READ_ONCE(sk->sk_mark);
@@ -1017,6 +1023,7 @@ static void tcp_v4_send_ack(const struct sock *sk,
 
 	sock_net_set(ctl_sk, &init_net);
 	__TCP_INC_STATS(net, TCP_MIB_OUTSEGS);
+	local_unlock_nested_bh(&ipv4_tcp_sk.bh_lock);
 	local_bh_enable();
 }
 
@@ -3615,7 +3622,7 @@ void __init tcp_v4_init(void)
 
 		sk->sk_clockid = CLOCK_MONOTONIC;
 
-		per_cpu(ipv4_tcp_sk, cpu) = sk;
+		per_cpu(ipv4_tcp_sk.sock, cpu) = sk;
 	}
 	if (register_pernet_subsys(&tcp_sk_ops))
 		panic("Failed to create the TCP control socket.\n");
-- 
2.45.2

[PATCH v7 net-next 07/15] netfilter: br_netfilter: Use nested-BH locking for brnf_frag_data_storage.

[Sebastian Andrzej Siewior]

brnf_frag_data_storage is a per-CPU variable and relies on disabled BH
for its locking. Without per-CPU locking in local_bh_disable() on
PREEMPT_RT this data structure requires explicit locking.

Add a local_lock_t to the data structure and use local_lock_nested_bh()
for locking. This change adds only lockdep coverage and does not alter
the functional behaviour for !PREEMPT_RT.

Cc: Florian Westphal <fw@strlen.de>
Cc: Jozsef Kadlecsik <kadlec@netfilter.org>
Cc: Nikolay Aleksandrov <razor@blackwall.org>
Cc: Pablo Neira Ayuso <pablo@netfilter.org>
Cc: Roopa Prabhu <roopa@nvidia.com>
Cc: bridge@lists.linux.dev
Cc: coreteam@netfilter.org
Cc: netfilter-devel@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 net/bridge/br_netfilter_hooks.c | 20 ++++++++++++++++----
 1 file changed, 16 insertions(+), 4 deletions(-)

diff --git a/net/bridge/br_netfilter_hooks.c b/net/bridge/br_netfilter_hooks.c
index bf30c50b56895..3c9f6538990ea 100644
--- a/net/bridge/br_netfilter_hooks.c
+++ b/net/bridge/br_netfilter_hooks.c
@@ -137,6 +137,7 @@ static inline bool is_pppoe_ipv6(const struct sk_buff *skb,
 #define NF_BRIDGE_MAX_MAC_HEADER_LENGTH (PPPOE_SES_HLEN + ETH_HLEN)
 
 struct brnf_frag_data {
+	local_lock_t bh_lock;
 	char mac[NF_BRIDGE_MAX_MAC_HEADER_LENGTH];
 	u8 encap_size;
 	u8 size;
@@ -144,7 +145,9 @@ struct brnf_frag_data {
 	__be16 vlan_proto;
 };
 
-static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage);
+static DEFINE_PER_CPU(struct brnf_frag_data, brnf_frag_data_storage) = {
+	.bh_lock = INIT_LOCAL_LOCK(bh_lock),
+};
 
 static void nf_bridge_info_free(struct sk_buff *skb)
 {
@@ -850,6 +853,7 @@ static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff
 {
 	struct nf_bridge_info *nf_bridge = nf_bridge_info_get(skb);
 	unsigned int mtu, mtu_reserved;
+	int ret;
 
 	mtu_reserved = nf_bridge_mtu_reduction(skb);
 	mtu = skb->dev->mtu;
@@ -882,6 +886,7 @@ static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff
 
 		IPCB(skb)->frag_max_size = nf_bridge->frag_max_size;
 
+		local_lock_nested_bh(&brnf_frag_data_storage.bh_lock);
 		data = this_cpu_ptr(&brnf_frag_data_storage);
 
 		if (skb_vlan_tag_present(skb)) {
@@ -897,7 +902,9 @@ static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff
 		skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 						 data->size);
 
-		return br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit);
+		ret = br_nf_ip_fragment(net, sk, skb, br_nf_push_frag_xmit);
+		local_unlock_nested_bh(&brnf_frag_data_storage.bh_lock);
+		return ret;
 	}
 	if (IS_ENABLED(CONFIG_NF_DEFRAG_IPV6) &&
 	    skb->protocol == htons(ETH_P_IPV6)) {
@@ -909,6 +916,7 @@ static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff
 
 		IP6CB(skb)->frag_max_size = nf_bridge->frag_max_size;
 
+		local_lock_nested_bh(&brnf_frag_data_storage.bh_lock);
 		data = this_cpu_ptr(&brnf_frag_data_storage);
 		data->encap_size = nf_bridge_encap_header_len(skb);
 		data->size = ETH_HLEN + data->encap_size;
@@ -916,8 +924,12 @@ static int br_nf_dev_queue_xmit(struct net *net, struct sock *sk, struct sk_buff
 		skb_copy_from_linear_data_offset(skb, -data->size, data->mac,
 						 data->size);
 
-		if (v6ops)
-			return v6ops->fragment(net, sk, skb, br_nf_push_frag_xmit);
+		if (v6ops) {
+			ret = v6ops->fragment(net, sk, skb, br_nf_push_frag_xmit);
+			local_unlock_nested_bh(&brnf_frag_data_storage.bh_lock);
+			return ret;
+		}
+		local_unlock_nested_bh(&brnf_frag_data_storage.bh_lock);
 
 		kfree_skb(skb);
 		return -EMSGSIZE;
-- 
2.45.2

[PATCH v7 net-next 08/15] net: softnet_data: Make xmit per task.

[Sebastian Andrzej Siewior]

Softirq is preemptible on PREEMPT_RT. Without a per-CPU lock in
local_bh_disable() there is no guarantee that only one device is
transmitting at a time.
With preemption and multiple senders it is possible that the per-CPU
`recursion' counter gets incremented by different threads and exceeds
XMIT_RECURSION_LIMIT leading to a false positive recursion alert.
The `more' member is subject to similar problems if set by one thread
for one driver and wrongly used by another driver within another thread.

Instead of adding a lock to protect the per-CPU variable it is simpler
to make xmit per-task. Sending and receiving skbs happens always
in thread context anyway.

Having a lock to protected the per-CPU counter would block/ serialize two
sending threads needlessly. It would also require a recursive lock to
ensure that the owner can increment the counter further.

Make the softnet_data.xmit a task_struct member on PREEMPT_RT. Add
needed wrapper.

Cc: Ben Segall <bsegall@google.com>
Cc: Daniel Bristot de Oliveira <bristot@redhat.com>
Cc: Dietmar Eggemann <dietmar.eggemann@arm.com>
Cc: Juri Lelli <juri.lelli@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Valentin Schneider <vschneid@redhat.com>
Cc: Vincent Guittot <vincent.guittot@linaro.org>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 include/linux/netdevice.h      | 42 +++++++++++++++++++++++++---------
 include/linux/netdevice_xmit.h | 13 +++++++++++
 include/linux/sched.h          |  5 +++-
 net/core/dev.c                 | 14 ++++++++++++
 net/core/dev.h                 | 18 +++++++++++++++
 5 files changed, 80 insertions(+), 12 deletions(-)
 create mode 100644 include/linux/netdevice_xmit.h

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 85111502cf8f1..96b233a0bc2b3 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -43,6 +43,7 @@
 
 #include <linux/netdev_features.h>
 #include <linux/neighbour.h>
+#include <linux/netdevice_xmit.h>
 #include <uapi/linux/netdevice.h>
 #include <uapi/linux/if_bonding.h>
 #include <uapi/linux/pkt_cls.h>
@@ -3223,13 +3224,7 @@ struct softnet_data {
 	struct sk_buff_head	xfrm_backlog;
 #endif
 	/* written and read only by owning cpu: */
-	struct {
-		u16 recursion;
-		u8  more;
-#ifdef CONFIG_NET_EGRESS
-		u8  skip_txqueue;
-#endif
-	} xmit;
+	struct netdev_xmit xmit;
 #ifdef CONFIG_RPS
 	/* input_queue_head should be written by cpu owning this struct,
 	 * and only read by other cpus. Worth using a cache line.
@@ -3257,10 +3252,18 @@ struct softnet_data {
 
 DECLARE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
 
+#ifndef CONFIG_PREEMPT_RT
 static inline int dev_recursion_level(void)
 {
 	return this_cpu_read(softnet_data.xmit.recursion);
 }
+#else
+static inline int dev_recursion_level(void)
+{
+	return current->net_xmit.recursion;
+}
+
+#endif
 
 void __netif_schedule(struct Qdisc *q);
 void netif_schedule_queue(struct netdev_queue *txq);
@@ -4875,18 +4878,35 @@ static inline ktime_t netdev_get_tstamp(struct net_device *dev,
 	return hwtstamps->hwtstamp;
 }
 
-static inline netdev_tx_t __netdev_start_xmit(const struct net_device_ops *ops,
-					      struct sk_buff *skb, struct net_device *dev,
-					      bool more)
+#ifndef CONFIG_PREEMPT_RT
+static inline void netdev_xmit_set_more(bool more)
 {
 	__this_cpu_write(softnet_data.xmit.more, more);
-	return ops->ndo_start_xmit(skb, dev);
 }
 
 static inline bool netdev_xmit_more(void)
 {
 	return __this_cpu_read(softnet_data.xmit.more);
 }
+#else
+static inline void netdev_xmit_set_more(bool more)
+{
+	current->net_xmit.more = more;
+}
+
+static inline bool netdev_xmit_more(void)
+{
+	return current->net_xmit.more;
+}
+#endif
+
+static inline netdev_tx_t __netdev_start_xmit(const struct net_device_ops *ops,
+					      struct sk_buff *skb, struct net_device *dev,
+					      bool more)
+{
+	netdev_xmit_set_more(more);
+	return ops->ndo_start_xmit(skb, dev);
+}
 
 static inline netdev_tx_t netdev_start_xmit(struct sk_buff *skb, struct net_device *dev,
 					    struct netdev_queue *txq, bool more)
diff --git a/include/linux/netdevice_xmit.h b/include/linux/netdevice_xmit.h
new file mode 100644
index 0000000000000..38325e0702968
--- /dev/null
+++ b/include/linux/netdevice_xmit.h
@@ -0,0 +1,13 @@
+/* SPDX-License-Identifier: GPL-2.0-or-later */
+#ifndef _LINUX_NETDEVICE_XMIT_H
+#define _LINUX_NETDEVICE_XMIT_H
+
+struct netdev_xmit {
+	u16 recursion;
+	u8  more;
+#ifdef CONFIG_NET_EGRESS
+	u8  skip_txqueue;
+#endif
+};
+
+#endif
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 61591ac6eab6d..5187486c25222 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -36,6 +36,7 @@
 #include <linux/signal_types.h>
 #include <linux/syscall_user_dispatch_types.h>
 #include <linux/mm_types_task.h>
+#include <linux/netdevice_xmit.h>
 #include <linux/task_io_accounting.h>
 #include <linux/posix-timers_types.h>
 #include <linux/restart_block.h>
@@ -975,7 +976,9 @@ struct task_struct {
 	/* delay due to memory thrashing */
 	unsigned                        in_thrashing:1;
 #endif
-
+#ifdef CONFIG_PREEMPT_RT
+	struct netdev_xmit		net_xmit;
+#endif
 	unsigned long			atomic_flags; /* Flags requiring atomic access. */
 
 	struct restart_block		restart_block;
diff --git a/net/core/dev.c b/net/core/dev.c
index c361a7b69da86..c15b0215a66b7 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -3940,6 +3940,7 @@ netdev_tx_queue_mapping(struct net_device *dev, struct sk_buff *skb)
 	return netdev_get_tx_queue(dev, netdev_cap_txqueue(dev, qm));
 }
 
+#ifndef CONFIG_PREEMPT_RT
 static bool netdev_xmit_txqueue_skipped(void)
 {
 	return __this_cpu_read(softnet_data.xmit.skip_txqueue);
@@ -3950,6 +3951,19 @@ void netdev_xmit_skip_txqueue(bool skip)
 	__this_cpu_write(softnet_data.xmit.skip_txqueue, skip);
 }
 EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
+
+#else
+static bool netdev_xmit_txqueue_skipped(void)
+{
+	return current->net_xmit.skip_txqueue;
+}
+
+void netdev_xmit_skip_txqueue(bool skip)
+{
+	current->net_xmit.skip_txqueue = skip;
+}
+EXPORT_SYMBOL_GPL(netdev_xmit_skip_txqueue);
+#endif
 #endif /* CONFIG_NET_EGRESS */
 
 #ifdef CONFIG_NET_XGRESS
diff --git a/net/core/dev.h b/net/core/dev.h
index b7b518bc2be55..4984dd9b334bc 100644
--- a/net/core/dev.h
+++ b/net/core/dev.h
@@ -150,6 +150,8 @@ struct napi_struct *napi_by_id(unsigned int napi_id);
 void kick_defer_list_purge(struct softnet_data *sd, unsigned int cpu);
 
 #define XMIT_RECURSION_LIMIT	8
+
+#ifndef CONFIG_PREEMPT_RT
 static inline bool dev_xmit_recursion(void)
 {
 	return unlikely(__this_cpu_read(softnet_data.xmit.recursion) >
@@ -165,5 +167,21 @@ static inline void dev_xmit_recursion_dec(void)
 {
 	__this_cpu_dec(softnet_data.xmit.recursion);
 }
+#else
+static inline bool dev_xmit_recursion(void)
+{
+	return unlikely(current->net_xmit.recursion > XMIT_RECURSION_LIMIT);
+}
+
+static inline void dev_xmit_recursion_inc(void)
+{
+	current->net_xmit.recursion++;
+}
+
+static inline void dev_xmit_recursion_dec(void)
+{
+	current->net_xmit.recursion--;
+}
+#endif
 
 #endif
-- 
2.45.2

[PATCH v7 net-next 09/15] dev: Remove PREEMPT_RT ifdefs from backlog_lock.*().

[Sebastian Andrzej Siewior]

The backlog_napi locking (previously RPS) relies on explicit locking if
either RPS or backlog NAPI is enabled. If both are disabled then locking
was achieved by disabling interrupts except on PREEMPT_RT. PREEMPT_RT
was excluded because the needed synchronisation was already provided
local_bh_disable().

Since the introduction of backlog NAPI and making it mandatory for
PREEMPT_RT the ifdef within backlog_lock.*() is obsolete and can be
removed.

Remove the ifdefs in backlog_lock.*().

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 net/core/dev.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/net/core/dev.c b/net/core/dev.c
index c15b0215a66b7..2745001d09eb4 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -229,7 +229,7 @@ static inline void backlog_lock_irq_save(struct softnet_data *sd,
 {
 	if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
 		spin_lock_irqsave(&sd->input_pkt_queue.lock, *flags);
-	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
+	else
 		local_irq_save(*flags);
 }
 
@@ -237,7 +237,7 @@ static inline void backlog_lock_irq_disable(struct softnet_data *sd)
 {
 	if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
 		spin_lock_irq(&sd->input_pkt_queue.lock);
-	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
+	else
 		local_irq_disable();
 }
 
@@ -246,7 +246,7 @@ static inline void backlog_unlock_irq_restore(struct softnet_data *sd,
 {
 	if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
 		spin_unlock_irqrestore(&sd->input_pkt_queue.lock, *flags);
-	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
+	else
 		local_irq_restore(*flags);
 }
 
@@ -254,7 +254,7 @@ static inline void backlog_unlock_irq_enable(struct softnet_data *sd)
 {
 	if (IS_ENABLED(CONFIG_RPS) || use_backlog_threads())
 		spin_unlock_irq(&sd->input_pkt_queue.lock);
-	else if (!IS_ENABLED(CONFIG_PREEMPT_RT))
+	else
 		local_irq_enable();
 }
 
-- 
2.45.2

[PATCH v7 net-next 10/15] dev: Use nested-BH locking for softnet_data.process_queue.

[Sebastian Andrzej Siewior]

softnet_data::process_queue is a per-CPU variable and relies on disabled
BH for its locking. Without per-CPU locking in local_bh_disable() on
PREEMPT_RT this data structure requires explicit locking.

softnet_data::input_queue_head can be updated lockless. This is fine
because this value is only update CPU local by the local backlog_napi
thread.

Add a local_lock_t to softnet_data and use local_lock_nested_bh() for locking
of process_queue. This change adds only lockdep coverage and does not
alter the functional behaviour for !PREEMPT_RT.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 include/linux/netdevice.h |  1 +
 net/core/dev.c            | 12 +++++++++++-
 2 files changed, 12 insertions(+), 1 deletion(-)

diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index 96b233a0bc2b3..f725129c9f6f4 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -3202,6 +3202,7 @@ static inline bool dev_has_header(const struct net_device *dev)
 struct softnet_data {
 	struct list_head	poll_list;
 	struct sk_buff_head	process_queue;
+	local_lock_t		process_queue_bh_lock;
 
 	/* stats */
 	unsigned int		processed;
diff --git a/net/core/dev.c b/net/core/dev.c
index 2745001d09eb4..42ac76524e84e 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -449,7 +449,9 @@ static RAW_NOTIFIER_HEAD(netdev_chain);
  *	queue in the local softnet handler.
  */
 
-DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data);
+DEFINE_PER_CPU_ALIGNED(struct softnet_data, softnet_data) = {
+	.process_queue_bh_lock = INIT_LOCAL_LOCK(process_queue_bh_lock),
+};
 EXPORT_PER_CPU_SYMBOL(softnet_data);
 
 /* Page_pool has a lockless array/stack to alloc/recycle pages.
@@ -5949,6 +5951,7 @@ static void flush_backlog(struct work_struct *work)
 	}
 	backlog_unlock_irq_enable(sd);
 
+	local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
 	skb_queue_walk_safe(&sd->process_queue, skb, tmp) {
 		if (skb->dev->reg_state == NETREG_UNREGISTERING) {
 			__skb_unlink(skb, &sd->process_queue);
@@ -5956,6 +5959,7 @@ static void flush_backlog(struct work_struct *work)
 			rps_input_queue_head_incr(sd);
 		}
 	}
+	local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
 	local_bh_enable();
 }
 
@@ -6077,7 +6081,9 @@ static int process_backlog(struct napi_struct *napi, int quota)
 	while (again) {
 		struct sk_buff *skb;
 
+		local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
 		while ((skb = __skb_dequeue(&sd->process_queue))) {
+			local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
 			rcu_read_lock();
 			__netif_receive_skb(skb);
 			rcu_read_unlock();
@@ -6086,7 +6092,9 @@ static int process_backlog(struct napi_struct *napi, int quota)
 				return work;
 			}
 
+			local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
 		}
+		local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
 
 		backlog_lock_irq_disable(sd);
 		if (skb_queue_empty(&sd->input_pkt_queue)) {
@@ -6101,8 +6109,10 @@ static int process_backlog(struct napi_struct *napi, int quota)
 			napi->state &= NAPIF_STATE_THREADED;
 			again = false;
 		} else {
+			local_lock_nested_bh(&softnet_data.process_queue_bh_lock);
 			skb_queue_splice_tail_init(&sd->input_pkt_queue,
 						   &sd->process_queue);
+			local_unlock_nested_bh(&softnet_data.process_queue_bh_lock);
 		}
 		backlog_unlock_irq_enable(sd);
 	}
-- 
2.45.2

[PATCH v7 net-next 11/15] lwt: Don't disable migration prio invoking BPF.

[Sebastian Andrzej Siewior]

There is no need to explicitly disable migration if bottom halves are
also disabled. Disabling BH implies disabling migration.

Remove migrate_disable() and rely solely on disabling BH to remain on
the same CPU.

Cc: bpf@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 net/core/lwt_bpf.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)

diff --git a/net/core/lwt_bpf.c b/net/core/lwt_bpf.c
index 4a0797f0a154b..a94943681e5aa 100644
--- a/net/core/lwt_bpf.c
+++ b/net/core/lwt_bpf.c
@@ -40,10 +40,9 @@ static int run_lwt_bpf(struct sk_buff *skb, struct bpf_lwt_prog *lwt,
 {
 	int ret;
 
-	/* Migration disable and BH disable are needed to protect per-cpu
-	 * redirect_info between BPF prog and skb_do_redirect().
+	/* Disabling BH is needed to protect per-CPU bpf_redirect_info between
+	 * BPF prog and skb_do_redirect().
 	 */
-	migrate_disable();
 	local_bh_disable();
 	bpf_compute_data_pointers(skb);
 	ret = bpf_prog_run_save_cb(lwt->prog, skb);
@@ -78,7 +77,6 @@ static int run_lwt_bpf(struct sk_buff *skb, struct bpf_lwt_prog *lwt,
 	}
 
 	local_bh_enable();
-	migrate_enable();
 
 	return ret;
 }
-- 
2.45.2

[PATCH v7 net-next 12/15] seg6: Use nested-BH locking for seg6_bpf_srh_states.

[Sebastian Andrzej Siewior]

The access to seg6_bpf_srh_states is protected by disabling preemption.
Based on the code, the entry point is input_action_end_bpf() and
every other function (the bpf helper functions bpf_lwt_seg6_*()), that
is accessing seg6_bpf_srh_states, should be called from within
input_action_end_bpf().

input_action_end_bpf() accesses seg6_bpf_srh_states first at the top of
the function and then disables preemption. This looks wrong because if
preemption needs to be disabled as part of the locking mechanism then
the variable shouldn't be accessed beforehand.

Looking at how it is used via test_lwt_seg6local.sh then
input_action_end_bpf() is always invoked from softirq context. If this
is always the case then the preempt_disable() statement is superfluous.
If this is not always invoked from softirq then disabling only
preemption is not sufficient.

Replace the preempt_disable() statement with nested-BH locking. This is
not an equivalent replacement as it assumes that the invocation of
input_action_end_bpf() always occurs in softirq context and thus the
preempt_disable() is superfluous.
Add a local_lock_t the data structure and use local_lock_nested_bh() for
locking. Add lockdep_assert_held() to ensure the lock is held while the
per-CPU variable is referenced in the helper functions.

Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andrii Nakryiko <andrii@kernel.org>
Cc: David Ahern <dsahern@kernel.org>
Cc: Hao Luo <haoluo@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: KP Singh <kpsingh@kernel.org>
Cc: Martin KaFai Lau <martin.lau@linux.dev>
Cc: Song Liu <song@kernel.org>
Cc: Stanislav Fomichev <sdf@google.com>
Cc: Yonghong Song <yonghong.song@linux.dev>
Cc: bpf@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 include/net/seg6_local.h |  1 +
 net/core/filter.c        |  3 +++
 net/ipv6/seg6_local.c    | 22 ++++++++++++++--------
 3 files changed, 18 insertions(+), 8 deletions(-)

diff --git a/include/net/seg6_local.h b/include/net/seg6_local.h
index 3fab9dec2ec45..888c1ce6f5272 100644
--- a/include/net/seg6_local.h
+++ b/include/net/seg6_local.h
@@ -19,6 +19,7 @@ extern int seg6_lookup_nexthop(struct sk_buff *skb, struct in6_addr *nhaddr,
 extern bool seg6_bpf_has_valid_srh(struct sk_buff *skb);
 
 struct seg6_bpf_srh_state {
+	local_lock_t bh_lock;
 	struct ipv6_sr_hdr *srh;
 	u16 hdrlen;
 	bool valid;
diff --git a/net/core/filter.c b/net/core/filter.c
index 7c46ecba3b01b..ba1a739a9bedc 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -6450,6 +6450,7 @@ BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset,
 	void *srh_tlvs, *srh_end, *ptr;
 	int srhoff = 0;
 
+	lockdep_assert_held(&srh_state->bh_lock);
 	if (srh == NULL)
 		return -EINVAL;
 
@@ -6506,6 +6507,7 @@ BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb,
 	int hdroff = 0;
 	int err;
 
+	lockdep_assert_held(&srh_state->bh_lock);
 	switch (action) {
 	case SEG6_LOCAL_ACTION_END_X:
 		if (!seg6_bpf_has_valid_srh(skb))
@@ -6582,6 +6584,7 @@ BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset,
 	int srhoff = 0;
 	int ret;
 
+	lockdep_assert_held(&srh_state->bh_lock);
 	if (unlikely(srh == NULL))
 		return -EINVAL;
 
diff --git a/net/ipv6/seg6_local.c b/net/ipv6/seg6_local.c
index 24e2b4b494cb0..c4828c6620f07 100644
--- a/net/ipv6/seg6_local.c
+++ b/net/ipv6/seg6_local.c
@@ -1380,7 +1380,9 @@ static int input_action_end_b6_encap(struct sk_buff *skb,
 	return err;
 }
 
-DEFINE_PER_CPU(struct seg6_bpf_srh_state, seg6_bpf_srh_states);
+DEFINE_PER_CPU(struct seg6_bpf_srh_state, seg6_bpf_srh_states) = {
+	.bh_lock	= INIT_LOCAL_LOCK(bh_lock),
+};
 
 bool seg6_bpf_has_valid_srh(struct sk_buff *skb)
 {
@@ -1388,6 +1390,7 @@ bool seg6_bpf_has_valid_srh(struct sk_buff *skb)
 		this_cpu_ptr(&seg6_bpf_srh_states);
 	struct ipv6_sr_hdr *srh = srh_state->srh;
 
+	lockdep_assert_held(&srh_state->bh_lock);
 	if (unlikely(srh == NULL))
 		return false;
 
@@ -1408,8 +1411,7 @@ bool seg6_bpf_has_valid_srh(struct sk_buff *skb)
 static int input_action_end_bpf(struct sk_buff *skb,
 				struct seg6_local_lwt *slwt)
 {
-	struct seg6_bpf_srh_state *srh_state =
-		this_cpu_ptr(&seg6_bpf_srh_states);
+	struct seg6_bpf_srh_state *srh_state;
 	struct ipv6_sr_hdr *srh;
 	int ret;
 
@@ -1420,10 +1422,14 @@ static int input_action_end_bpf(struct sk_buff *skb,
 	}
 	advance_nextseg(srh, &ipv6_hdr(skb)->daddr);
 
-	/* preempt_disable is needed to protect the per-CPU buffer srh_state,
-	 * which is also accessed by the bpf_lwt_seg6_* helpers
+	/* The access to the per-CPU buffer srh_state is protected by running
+	 * always in softirq context (with disabled BH). On PREEMPT_RT the
+	 * required locking is provided by the following local_lock_nested_bh()
+	 * statement. It is also accessed by the bpf_lwt_seg6_* helpers via
+	 * bpf_prog_run_save_cb().
 	 */
-	preempt_disable();
+	local_lock_nested_bh(&seg6_bpf_srh_states.bh_lock);
+	srh_state = this_cpu_ptr(&seg6_bpf_srh_states);
 	srh_state->srh = srh;
 	srh_state->hdrlen = srh->hdrlen << 3;
 	srh_state->valid = true;
@@ -1446,15 +1452,15 @@ static int input_action_end_bpf(struct sk_buff *skb,
 
 	if (srh_state->srh && !seg6_bpf_has_valid_srh(skb))
 		goto drop;
+	local_unlock_nested_bh(&seg6_bpf_srh_states.bh_lock);
 
-	preempt_enable();
 	if (ret != BPF_REDIRECT)
 		seg6_lookup_nexthop(skb, NULL, 0);
 
 	return dst_input(skb);
 
 drop:
-	preempt_enable();
+	local_unlock_nested_bh(&seg6_bpf_srh_states.bh_lock);
 	kfree_skb(skb);
 	return -EINVAL;
 }
-- 
2.45.2

[PATCH v7 net-next 13/15] net: Use nested-BH locking for bpf_scratchpad.

[Sebastian Andrzej Siewior]

bpf_scratchpad is a per-CPU variable and relies on disabled BH for its
locking. Without per-CPU locking in local_bh_disable() on PREEMPT_RT
this data structure requires explicit locking.

Add a local_lock_t to the data structure and use local_lock_nested_bh()
for locking. This change adds only lockdep coverage and does not alter
the functional behaviour for !PREEMPT_RT.

Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andrii Nakryiko <andrii@kernel.org>
Cc: Hao Luo <haoluo@google.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: KP Singh <kpsingh@kernel.org>
Cc: Martin KaFai Lau <martin.lau@linux.dev>
Cc: Song Liu <song@kernel.org>
Cc: Stanislav Fomichev <sdf@google.com>
Cc: Yonghong Song <yonghong.song@linux.dev>
Cc: bpf@vger.kernel.org
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 net/core/filter.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/net/core/filter.c b/net/core/filter.c
index ba1a739a9bedc..fbcfd563dccfd 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -1658,9 +1658,12 @@ struct bpf_scratchpad {
 		__be32 diff[MAX_BPF_STACK / sizeof(__be32)];
 		u8     buff[MAX_BPF_STACK];
 	};
+	local_lock_t	bh_lock;
 };
 
-static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp);
+static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp) = {
+	.bh_lock	= INIT_LOCAL_LOCK(bh_lock),
+};
 
 static inline int __bpf_try_make_writable(struct sk_buff *skb,
 					  unsigned int write_len)
@@ -2016,6 +2019,7 @@ BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
 	struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp);
 	u32 diff_size = from_size + to_size;
 	int i, j = 0;
+	__wsum ret;
 
 	/* This is quite flexible, some examples:
 	 *
@@ -2029,12 +2033,15 @@ BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size,
 		     diff_size > sizeof(sp->diff)))
 		return -EINVAL;
 
+	local_lock_nested_bh(&bpf_sp.bh_lock);
 	for (i = 0; i < from_size / sizeof(__be32); i++, j++)
 		sp->diff[j] = ~from[i];
 	for (i = 0; i <   to_size / sizeof(__be32); i++, j++)
 		sp->diff[j] = to[i];
 
-	return csum_partial(sp->diff, diff_size, seed);
+	ret = csum_partial(sp->diff, diff_size, seed);
+	local_unlock_nested_bh(&bpf_sp.bh_lock);
+	return ret;
 }
 
 static const struct bpf_func_proto bpf_csum_diff_proto = {
-- 
2.45.2

[PATCH v7 net-next 14/15] net: Reference bpf_redirect_info via task_struct on PREEMPT_RT.

[Sebastian Andrzej Siewior]

The XDP redirect process is two staged:
- bpf_prog_run_xdp() is invoked to run a eBPF program which inspects the
  packet and makes decisions. While doing that, the per-CPU variable
  bpf_redirect_info is used.

- Afterwards xdp_do_redirect() is invoked and accesses bpf_redirect_info
  and it may also access other per-CPU variables like xskmap_flush_list.

At the very end of the NAPI callback, xdp_do_flush() is invoked which
does not access bpf_redirect_info but will touch the individual per-CPU
lists.

The per-CPU variables are only used in the NAPI callback hence disabling
bottom halves is the only protection mechanism. Users from preemptible
context (like cpu_map_kthread_run()) explicitly disable bottom halves
for protections reasons.
Without locking in local_bh_disable() on PREEMPT_RT this data structure
requires explicit locking.

PREEMPT_RT has forced-threaded interrupts enabled and every
NAPI-callback runs in a thread. If each thread has its own data
structure then locking can be avoided.

Create a struct bpf_net_context which contains struct bpf_redirect_info.
Define the variable on stack, use bpf_net_ctx_set() to save a pointer to
it, bpf_net_ctx_clear() removes it again.
The bpf_net_ctx_set() may nest. For instance a function can be used from
within NET_RX_SOFTIRQ/ net_rx_action which uses bpf_net_ctx_set() and
NET_TX_SOFTIRQ which does not. Therefore only the first invocations
updates the pointer.
Use bpf_net_ctx_get_ri() as a wrapper to retrieve the current struct
bpf_redirect_info. The returned data structure is zero initialized to
ensure nothing is leaked from stack. This is done on first usage of the
struct. bpf_net_ctx_set() sets bpf_redirect_info::kern_flags to 0 to
note that initialisation is required. First invocation of
bpf_net_ctx_get_ri() will memset() the data structure and update
bpf_redirect_info::kern_flags.
bpf_redirect_info::nh is excluded from memset because it is only used
once BPF_F_NEIGH is set which also sets the nh member. The kern_flags is
moved past nh to exclude it from memset.

The pointer to bpf_net_context is saved task's task_struct. Using
always the bpf_net_context approach has the advantage that there is
almost zero differences between PREEMPT_RT and non-PREEMPT_RT builds.

Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andrii Nakryiko <andrii@kernel.org>
Cc: Eduard Zingerman <eddyz87@gmail.com>
Cc: Hao Luo <haoluo@google.com>
Cc: Jesper Dangaard Brouer <hawk@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: KP Singh <kpsingh@kernel.org>
Cc: Martin KaFai Lau <martin.lau@linux.dev>
Cc: Song Liu <song@kernel.org>
Cc: Stanislav Fomichev <sdf@google.com>
Cc: Toke Høiland-Jørgensen <toke@redhat.com>
Cc: Yonghong Song <yonghong.song@linux.dev>
Cc: bpf@vger.kernel.org
Acked-by: Alexei Starovoitov <ast@kernel.org>
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 include/linux/filter.h | 56 ++++++++++++++++++++++++++++++++++--------
 include/linux/sched.h  |  3 +++
 kernel/bpf/cpumap.c    |  3 +++
 kernel/bpf/devmap.c    |  9 ++++++-
 kernel/fork.c          |  1 +
 net/bpf/test_run.c     | 11 ++++++++-
 net/core/dev.c         | 26 +++++++++++++++++++-
 net/core/filter.c      | 44 +++++++++------------------------
 net/core/lwt_bpf.c     |  3 +++
 9 files changed, 111 insertions(+), 45 deletions(-)

diff --git a/include/linux/filter.h b/include/linux/filter.h
index b02aea291b7e8..0a7f6e4a00b60 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -733,21 +733,59 @@ struct bpf_nh_params {
 	};
 };
 
+/* flags for bpf_redirect_info kern_flags */
+#define BPF_RI_F_RF_NO_DIRECT	BIT(0)	/* no napi_direct on return_frame */
+#define BPF_RI_F_RI_INIT	BIT(1)
+
 struct bpf_redirect_info {
 	u64 tgt_index;
 	void *tgt_value;
 	struct bpf_map *map;
 	u32 flags;
-	u32 kern_flags;
 	u32 map_id;
 	enum bpf_map_type map_type;
 	struct bpf_nh_params nh;
+	u32 kern_flags;
 };
 
-DECLARE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
+struct bpf_net_context {
+	struct bpf_redirect_info ri;
+};
 
-/* flags for bpf_redirect_info kern_flags */
-#define BPF_RI_F_RF_NO_DIRECT	BIT(0)	/* no napi_direct on return_frame */
+static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx)
+{
+	struct task_struct *tsk = current;
+
+	if (tsk->bpf_net_context != NULL)
+		return NULL;
+	bpf_net_ctx->ri.kern_flags = 0;
+
+	tsk->bpf_net_context = bpf_net_ctx;
+	return bpf_net_ctx;
+}
+
+static inline void bpf_net_ctx_clear(struct bpf_net_context *bpf_net_ctx)
+{
+	if (bpf_net_ctx)
+		current->bpf_net_context = NULL;
+}
+
+static inline struct bpf_net_context *bpf_net_ctx_get(void)
+{
+	return current->bpf_net_context;
+}
+
+static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void)
+{
+	struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
+
+	if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_RI_INIT)) {
+		memset(&bpf_net_ctx->ri, 0, offsetof(struct bpf_net_context, ri.nh));
+		bpf_net_ctx->ri.kern_flags |= BPF_RI_F_RI_INIT;
+	}
+
+	return &bpf_net_ctx->ri;
+}
 
 /* Compute the linear packet data range [data, data_end) which
  * will be accessed by various program types (cls_bpf, act_bpf,
@@ -1018,25 +1056,23 @@ struct bpf_prog *bpf_patch_insn_single(struct bpf_prog *prog, u32 off,
 				       const struct bpf_insn *patch, u32 len);
 int bpf_remove_insns(struct bpf_prog *prog, u32 off, u32 cnt);
 
-void bpf_clear_redirect_map(struct bpf_map *map);
-
 static inline bool xdp_return_frame_no_direct(void)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 
 	return ri->kern_flags & BPF_RI_F_RF_NO_DIRECT;
 }
 
 static inline void xdp_set_return_frame_no_direct(void)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 
 	ri->kern_flags |= BPF_RI_F_RF_NO_DIRECT;
 }
 
 static inline void xdp_clear_return_frame_no_direct(void)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 
 	ri->kern_flags &= ~BPF_RI_F_RF_NO_DIRECT;
 }
@@ -1592,7 +1628,7 @@ static __always_inline long __bpf_xdp_redirect_map(struct bpf_map *map, u64 inde
 						   u64 flags, const u64 flag_mask,
 						   void *lookup_elem(struct bpf_map *map, u32 key))
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 	const u64 action_mask = XDP_ABORTED | XDP_DROP | XDP_PASS | XDP_TX;
 
 	/* Lower bits of the flags are used as return code on lookup failure */
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 5187486c25222..5ff5e65a46277 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -54,6 +54,7 @@ struct bio_list;
 struct blk_plug;
 struct bpf_local_storage;
 struct bpf_run_ctx;
+struct bpf_net_context;
 struct capture_control;
 struct cfs_rq;
 struct fs_struct;
@@ -1509,6 +1510,8 @@ struct task_struct {
 	/* Used for BPF run context */
 	struct bpf_run_ctx		*bpf_ctx;
 #endif
+	/* Used by BPF for per-TASK xdp storage */
+	struct bpf_net_context		*bpf_net_context;
 
 #ifdef CONFIG_GCC_PLUGIN_STACKLEAK
 	unsigned long			lowest_stack;
diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c
index a8e34416e960f..66974bd027109 100644
--- a/kernel/bpf/cpumap.c
+++ b/kernel/bpf/cpumap.c
@@ -240,12 +240,14 @@ static int cpu_map_bpf_prog_run(struct bpf_cpu_map_entry *rcpu, void **frames,
 				int xdp_n, struct xdp_cpumap_stats *stats,
 				struct list_head *list)
 {
+	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
 	int nframes;
 
 	if (!rcpu->prog)
 		return xdp_n;
 
 	rcu_read_lock_bh();
+	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
 
 	nframes = cpu_map_bpf_prog_run_xdp(rcpu, frames, xdp_n, stats);
 
@@ -255,6 +257,7 @@ static int cpu_map_bpf_prog_run(struct bpf_cpu_map_entry *rcpu, void **frames,
 	if (unlikely(!list_empty(list)))
 		cpu_map_bpf_prog_run_skb(rcpu, list, stats);
 
+	bpf_net_ctx_clear(bpf_net_ctx);
 	rcu_read_unlock_bh(); /* resched point, may call do_softirq() */
 
 	return nframes;
diff --git a/kernel/bpf/devmap.c b/kernel/bpf/devmap.c
index 7f3b34452243c..fbfdfb60db8d7 100644
--- a/kernel/bpf/devmap.c
+++ b/kernel/bpf/devmap.c
@@ -196,7 +196,14 @@ static void dev_map_free(struct bpf_map *map)
 	list_del_rcu(&dtab->list);
 	spin_unlock(&dev_map_lock);
 
-	bpf_clear_redirect_map(map);
+	/* bpf_redirect_info->map is assigned in __bpf_xdp_redirect_map()
+	 * during NAPI callback and cleared after the XDP redirect. There is no
+	 * explicit RCU read section which protects bpf_redirect_info->map but
+	 * local_bh_disable() also marks the beginning an RCU section. This
+	 * makes the complete softirq callback RCU protected. Thus after
+	 * following synchronize_rcu() there no bpf_redirect_info->map == map
+	 * assignment.
+	 */
 	synchronize_rcu();
 
 	/* Make sure prior __dev_map_entry_free() have completed. */
diff --git a/kernel/fork.c b/kernel/fork.c
index 99076dbe27d83..f314bdd7e6108 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -2355,6 +2355,7 @@ __latent_entropy struct task_struct *copy_process(
 	RCU_INIT_POINTER(p->bpf_storage, NULL);
 	p->bpf_ctx = NULL;
 #endif
+	p->bpf_net_context =  NULL;
 
 	/* Perform scheduler related setup. Assign this task to a CPU. */
 	retval = sched_fork(clone_flags, p);
diff --git a/net/bpf/test_run.c b/net/bpf/test_run.c
index 36ae54f57bf57..a6d7f790cdda8 100644
--- a/net/bpf/test_run.c
+++ b/net/bpf/test_run.c
@@ -283,9 +283,10 @@ static int xdp_recv_frames(struct xdp_frame **frames, int nframes,
 static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
 			      u32 repeat)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
 	int err = 0, act, ret, i, nframes = 0, batch_sz;
 	struct xdp_frame **frames = xdp->frames;
+	struct bpf_redirect_info *ri;
 	struct xdp_page_head *head;
 	struct xdp_frame *frm;
 	bool redirect = false;
@@ -295,6 +296,8 @@ static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
 	batch_sz = min_t(u32, repeat, xdp->batch_size);
 
 	local_bh_disable();
+	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
+	ri = bpf_net_ctx_get_ri();
 	xdp_set_return_frame_no_direct();
 
 	for (i = 0; i < batch_sz; i++) {
@@ -359,6 +362,7 @@ static int xdp_test_run_batch(struct xdp_test_data *xdp, struct bpf_prog *prog,
 	}
 
 	xdp_clear_return_frame_no_direct();
+	bpf_net_ctx_clear(bpf_net_ctx);
 	local_bh_enable();
 	return err;
 }
@@ -394,6 +398,7 @@ static int bpf_test_run_xdp_live(struct bpf_prog *prog, struct xdp_buff *ctx,
 static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
 			u32 *retval, u32 *time, bool xdp)
 {
+	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
 	struct bpf_prog_array_item item = {.prog = prog};
 	struct bpf_run_ctx *old_ctx;
 	struct bpf_cg_run_ctx run_ctx;
@@ -419,10 +424,14 @@ static int bpf_test_run(struct bpf_prog *prog, void *ctx, u32 repeat,
 	do {
 		run_ctx.prog_item = &item;
 		local_bh_disable();
+		bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
+
 		if (xdp)
 			*retval = bpf_prog_run_xdp(prog, ctx);
 		else
 			*retval = bpf_prog_run(prog, ctx);
+
+		bpf_net_ctx_clear(bpf_net_ctx);
 		local_bh_enable();
 	} while (bpf_test_timer_continue(&t, 1, repeat, &ret, time));
 	bpf_reset_run_ctx(old_ctx);
diff --git a/net/core/dev.c b/net/core/dev.c
index 42ac76524e84e..1e0ce68c78b48 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4045,10 +4045,13 @@ sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
 {
 	struct bpf_mprog_entry *entry = rcu_dereference_bh(skb->dev->tcx_ingress);
 	enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_INGRESS;
+	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
 	int sch_ret;
 
 	if (!entry)
 		return skb;
+
+	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
 	if (*pt_prev) {
 		*ret = deliver_skb(skb, *pt_prev, orig_dev);
 		*pt_prev = NULL;
@@ -4077,10 +4080,12 @@ sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
 			break;
 		}
 		*ret = NET_RX_SUCCESS;
+		bpf_net_ctx_clear(bpf_net_ctx);
 		return NULL;
 	case TC_ACT_SHOT:
 		kfree_skb_reason(skb, drop_reason);
 		*ret = NET_RX_DROP;
+		bpf_net_ctx_clear(bpf_net_ctx);
 		return NULL;
 	/* used by tc_run */
 	case TC_ACT_STOLEN:
@@ -4090,8 +4095,10 @@ sch_handle_ingress(struct sk_buff *skb, struct packet_type **pt_prev, int *ret,
 		fallthrough;
 	case TC_ACT_CONSUMED:
 		*ret = NET_RX_SUCCESS;
+		bpf_net_ctx_clear(bpf_net_ctx);
 		return NULL;
 	}
+	bpf_net_ctx_clear(bpf_net_ctx);
 
 	return skb;
 }
@@ -4101,11 +4108,14 @@ sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
 {
 	struct bpf_mprog_entry *entry = rcu_dereference_bh(dev->tcx_egress);
 	enum skb_drop_reason drop_reason = SKB_DROP_REASON_TC_EGRESS;
+	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
 	int sch_ret;
 
 	if (!entry)
 		return skb;
 
+	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
+
 	/* qdisc_skb_cb(skb)->pkt_len & tcx_set_ingress() was
 	 * already set by the caller.
 	 */
@@ -4121,10 +4131,12 @@ sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
 		/* No need to push/pop skb's mac_header here on egress! */
 		skb_do_redirect(skb);
 		*ret = NET_XMIT_SUCCESS;
+		bpf_net_ctx_clear(bpf_net_ctx);
 		return NULL;
 	case TC_ACT_SHOT:
 		kfree_skb_reason(skb, drop_reason);
 		*ret = NET_XMIT_DROP;
+		bpf_net_ctx_clear(bpf_net_ctx);
 		return NULL;
 	/* used by tc_run */
 	case TC_ACT_STOLEN:
@@ -4134,8 +4146,10 @@ sch_handle_egress(struct sk_buff *skb, int *ret, struct net_device *dev)
 		fallthrough;
 	case TC_ACT_CONSUMED:
 		*ret = NET_XMIT_SUCCESS;
+		bpf_net_ctx_clear(bpf_net_ctx);
 		return NULL;
 	}
+	bpf_net_ctx_clear(bpf_net_ctx);
 
 	return skb;
 }
@@ -6373,6 +6387,7 @@ static void __napi_busy_loop(unsigned int napi_id,
 {
 	unsigned long start_time = loop_end ? busy_loop_current_time() : 0;
 	int (*napi_poll)(struct napi_struct *napi, int budget);
+	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
 	void *have_poll_lock = NULL;
 	struct napi_struct *napi;
 
@@ -6391,6 +6406,7 @@ static void __napi_busy_loop(unsigned int napi_id,
 		int work = 0;
 
 		local_bh_disable();
+		bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
 		if (!napi_poll) {
 			unsigned long val = READ_ONCE(napi->state);
 
@@ -6421,6 +6437,7 @@ static void __napi_busy_loop(unsigned int napi_id,
 			__NET_ADD_STATS(dev_net(napi->dev),
 					LINUX_MIB_BUSYPOLLRXPACKETS, work);
 		skb_defer_free_flush(this_cpu_ptr(&softnet_data));
+		bpf_net_ctx_clear(bpf_net_ctx);
 		local_bh_enable();
 
 		if (!loop_end || loop_end(loop_end_arg, start_time))
@@ -6848,6 +6865,7 @@ static int napi_thread_wait(struct napi_struct *napi)
 
 static void napi_threaded_poll_loop(struct napi_struct *napi)
 {
+	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
 	struct softnet_data *sd;
 	unsigned long last_qs = jiffies;
 
@@ -6856,6 +6874,8 @@ static void napi_threaded_poll_loop(struct napi_struct *napi)
 		void *have;
 
 		local_bh_disable();
+		bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
+
 		sd = this_cpu_ptr(&softnet_data);
 		sd->in_napi_threaded_poll = true;
 
@@ -6871,6 +6891,7 @@ static void napi_threaded_poll_loop(struct napi_struct *napi)
 			net_rps_action_and_irq_enable(sd);
 		}
 		skb_defer_free_flush(sd);
+		bpf_net_ctx_clear(bpf_net_ctx);
 		local_bh_enable();
 
 		if (!repoll)
@@ -6896,10 +6917,12 @@ static __latent_entropy void net_rx_action(struct softirq_action *h)
 	struct softnet_data *sd = this_cpu_ptr(&softnet_data);
 	unsigned long time_limit = jiffies +
 		usecs_to_jiffies(READ_ONCE(net_hotdata.netdev_budget_usecs));
+	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
 	int budget = READ_ONCE(net_hotdata.netdev_budget);
 	LIST_HEAD(list);
 	LIST_HEAD(repoll);
 
+	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
 start:
 	sd->in_net_rx_action = true;
 	local_irq_disable();
@@ -6952,7 +6975,8 @@ static __latent_entropy void net_rx_action(struct softirq_action *h)
 		sd->in_net_rx_action = false;
 
 	net_rps_action_and_irq_enable(sd);
-end:;
+end:
+	bpf_net_ctx_clear(bpf_net_ctx);
 }
 
 struct netdev_adjacent {
diff --git a/net/core/filter.c b/net/core/filter.c
index fbcfd563dccfd..f40b8393dd58f 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -2478,9 +2478,6 @@ static const struct bpf_func_proto bpf_clone_redirect_proto = {
 	.arg3_type      = ARG_ANYTHING,
 };
 
-DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info);
-EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info);
-
 static struct net_device *skb_get_peer_dev(struct net_device *dev)
 {
 	const struct net_device_ops *ops = dev->netdev_ops;
@@ -2493,7 +2490,7 @@ static struct net_device *skb_get_peer_dev(struct net_device *dev)
 
 int skb_do_redirect(struct sk_buff *skb)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 	struct net *net = dev_net(skb->dev);
 	struct net_device *dev;
 	u32 flags = ri->flags;
@@ -2526,7 +2523,7 @@ int skb_do_redirect(struct sk_buff *skb)
 
 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 
 	if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL)))
 		return TC_ACT_SHOT;
@@ -2547,7 +2544,7 @@ static const struct bpf_func_proto bpf_redirect_proto = {
 
 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 
 	if (unlikely(flags))
 		return TC_ACT_SHOT;
@@ -2569,7 +2566,7 @@ static const struct bpf_func_proto bpf_redirect_peer_proto = {
 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params,
 	   int, plen, u64, flags)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 
 	if (unlikely((plen && plen < sizeof(*params)) || flags))
 		return TC_ACT_SHOT;
@@ -4295,30 +4292,13 @@ void xdp_do_check_flushed(struct napi_struct *napi)
 }
 #endif
 
-void bpf_clear_redirect_map(struct bpf_map *map)
-{
-	struct bpf_redirect_info *ri;
-	int cpu;
-
-	for_each_possible_cpu(cpu) {
-		ri = per_cpu_ptr(&bpf_redirect_info, cpu);
-		/* Avoid polluting remote cacheline due to writes if
-		 * not needed. Once we pass this test, we need the
-		 * cmpxchg() to make sure it hasn't been changed in
-		 * the meantime by remote CPU.
-		 */
-		if (unlikely(READ_ONCE(ri->map) == map))
-			cmpxchg(&ri->map, map, NULL);
-	}
-}
-
 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key);
 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key);
 
 u32 xdp_master_redirect(struct xdp_buff *xdp)
 {
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 	struct net_device *master, *slave;
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
 
 	master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev);
 	slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp);
@@ -4390,7 +4370,7 @@ static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
 			map = READ_ONCE(ri->map);
 
 			/* The map pointer is cleared when the map is being torn
-			 * down by bpf_clear_redirect_map()
+			 * down by dev_map_free()
 			 */
 			if (unlikely(!map)) {
 				err = -ENOENT;
@@ -4435,7 +4415,7 @@ static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri,
 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp,
 		    struct bpf_prog *xdp_prog)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 	enum bpf_map_type map_type = ri->map_type;
 
 	if (map_type == BPF_MAP_TYPE_XSKMAP)
@@ -4449,7 +4429,7 @@ EXPORT_SYMBOL_GPL(xdp_do_redirect);
 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp,
 			  struct xdp_frame *xdpf, struct bpf_prog *xdp_prog)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 	enum bpf_map_type map_type = ri->map_type;
 
 	if (map_type == BPF_MAP_TYPE_XSKMAP)
@@ -4466,7 +4446,7 @@ static int xdp_do_generic_redirect_map(struct net_device *dev,
 				       enum bpf_map_type map_type, u32 map_id,
 				       u32 flags)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 	struct bpf_map *map;
 	int err;
 
@@ -4478,7 +4458,7 @@ static int xdp_do_generic_redirect_map(struct net_device *dev,
 			map = READ_ONCE(ri->map);
 
 			/* The map pointer is cleared when the map is being torn
-			 * down by bpf_clear_redirect_map()
+			 * down by dev_map_free()
 			 */
 			if (unlikely(!map)) {
 				err = -ENOENT;
@@ -4520,7 +4500,7 @@ static int xdp_do_generic_redirect_map(struct net_device *dev,
 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
 			    struct xdp_buff *xdp, struct bpf_prog *xdp_prog)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 	enum bpf_map_type map_type = ri->map_type;
 	void *fwd = ri->tgt_value;
 	u32 map_id = ri->map_id;
@@ -4556,7 +4536,7 @@ int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb,
 
 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags)
 {
-	struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info);
+	struct bpf_redirect_info *ri = bpf_net_ctx_get_ri();
 
 	if (unlikely(flags))
 		return XDP_ABORTED;
diff --git a/net/core/lwt_bpf.c b/net/core/lwt_bpf.c
index a94943681e5aa..afb05f58b64c5 100644
--- a/net/core/lwt_bpf.c
+++ b/net/core/lwt_bpf.c
@@ -38,12 +38,14 @@ static inline struct bpf_lwt *bpf_lwt_lwtunnel(struct lwtunnel_state *lwt)
 static int run_lwt_bpf(struct sk_buff *skb, struct bpf_lwt_prog *lwt,
 		       struct dst_entry *dst, bool can_redirect)
 {
+	struct bpf_net_context __bpf_net_ctx, *bpf_net_ctx;
 	int ret;
 
 	/* Disabling BH is needed to protect per-CPU bpf_redirect_info between
 	 * BPF prog and skb_do_redirect().
 	 */
 	local_bh_disable();
+	bpf_net_ctx = bpf_net_ctx_set(&__bpf_net_ctx);
 	bpf_compute_data_pointers(skb);
 	ret = bpf_prog_run_save_cb(lwt->prog, skb);
 
@@ -76,6 +78,7 @@ static int run_lwt_bpf(struct sk_buff *skb, struct bpf_lwt_prog *lwt,
 		break;
 	}
 
+	bpf_net_ctx_clear(bpf_net_ctx);
 	local_bh_enable();
 
 	return ret;
-- 
2.45.2

Re: [PATCH v7 net-next 14/15] net: Reference bpf_redirect_info via task_struct on PREEMPT_RT.

[Jesper Dangaard Brouer]

On 18/06/2024 09.13, Sebastian Andrzej Siewior wrote:
> The XDP redirect process is two staged:
> - bpf_prog_run_xdp() is invoked to run a eBPF program which inspects the
>    packet and makes decisions. While doing that, the per-CPU variable
>    bpf_redirect_info is used.
> 
> - Afterwards xdp_do_redirect() is invoked and accesses bpf_redirect_info
>    and it may also access other per-CPU variables like xskmap_flush_list.
> 
> At the very end of the NAPI callback, xdp_do_flush() is invoked which
> does not access bpf_redirect_info but will touch the individual per-CPU
> lists.
> 
> The per-CPU variables are only used in the NAPI callback hence disabling
> bottom halves is the only protection mechanism. Users from preemptible
> context (like cpu_map_kthread_run()) explicitly disable bottom halves
> for protections reasons.
> Without locking in local_bh_disable() on PREEMPT_RT this data structure
> requires explicit locking.
> 
> PREEMPT_RT has forced-threaded interrupts enabled and every
> NAPI-callback runs in a thread. If each thread has its own data
> structure then locking can be avoided.
> 
> Create a struct bpf_net_context which contains struct bpf_redirect_info.
> Define the variable on stack, use bpf_net_ctx_set() to save a pointer to
> it, bpf_net_ctx_clear() removes it again.
> The bpf_net_ctx_set() may nest. For instance a function can be used from
> within NET_RX_SOFTIRQ/ net_rx_action which uses bpf_net_ctx_set() and
> NET_TX_SOFTIRQ which does not. Therefore only the first invocations
> updates the pointer.
> Use bpf_net_ctx_get_ri() as a wrapper to retrieve the current struct
> bpf_redirect_info. The returned data structure is zero initialized to
> ensure nothing is leaked from stack. This is done on first usage of the
> struct. bpf_net_ctx_set() sets bpf_redirect_info::kern_flags  to 0 to
> note that initialisation is required. First invocation of
> bpf_net_ctx_get_ri() will memset() the data structure and update
> bpf_redirect_info::kern_flags.
> bpf_redirect_info::nh  is excluded from memset because it is only used
> once BPF_F_NEIGH is set which also sets the nh member. The kern_flags is
> moved past nh to exclude it from memset.
> 
> The pointer to bpf_net_context is saved task's task_struct. Using
> always the bpf_net_context approach has the advantage that there is
> almost zero differences between PREEMPT_RT and non-PREEMPT_RT builds.
> 
> Cc: Alexei Starovoitov<ast@kernel.org>
> Cc: Andrii Nakryiko<andrii@kernel.org>
> Cc: Eduard Zingerman<eddyz87@gmail.com>
> Cc: Hao Luo<haoluo@google.com>
> Cc: Jesper Dangaard Brouer<hawk@kernel.org>
> Cc: Jiri Olsa<jolsa@kernel.org>
> Cc: John Fastabend<john.fastabend@gmail.com>
> Cc: KP Singh<kpsingh@kernel.org>
> Cc: Martin KaFai Lau<martin.lau@linux.dev>
> Cc: Song Liu<song@kernel.org>
> Cc: Stanislav Fomichev<sdf@google.com>
> Cc: Toke Høiland-Jørgensen<toke@redhat.com>
> Cc: Yonghong Song<yonghong.song@linux.dev>
> Cc:bpf@vger.kernel.org
> Acked-by: Alexei Starovoitov<ast@kernel.org>
> Reviewed-by: Toke Høiland-Jørgensen<toke@redhat.com>
> Signed-off-by: Sebastian Andrzej Siewior<bigeasy@linutronix.de>

Acked-by: Jesper Dangaard Brouer <hawk@kernel.org>

> ---
>   include/linux/filter.h | 56 ++++++++++++++++++++++++++++++++++--------
>   include/linux/sched.h  |  3 +++
>   kernel/bpf/cpumap.c    |  3 +++
>   kernel/bpf/devmap.c    |  9 ++++++-
>   kernel/fork.c          |  1 +
>   net/bpf/test_run.c     | 11 ++++++++-
>   net/core/dev.c         | 26 +++++++++++++++++++-
>   net/core/filter.c      | 44 +++++++++------------------------
>   net/core/lwt_bpf.c     |  3 +++
>   9 files changed, 111 insertions(+), 45 deletions(-)

[PATCH v7 net-next 15/15] net: Move per-CPU flush-lists to bpf_net_context on PREEMPT_RT.

[Sebastian Andrzej Siewior]

The per-CPU flush lists, which are accessed from within the NAPI callback
(xdp_do_flush() for instance), are per-CPU. There are subject to the
same problem as struct bpf_redirect_info.

Add the per-CPU lists cpu_map_flush_list, dev_map_flush_list and
xskmap_map_flush_list to struct bpf_net_context. Add wrappers for the
access. The lists initialized on first usage (similar to
bpf_net_ctx_get_ri()).

Cc: "Björn Töpel" <bjorn@kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Andrii Nakryiko <andrii@kernel.org>
Cc: Eduard Zingerman <eddyz87@gmail.com>
Cc: Hao Luo <haoluo@google.com>
Cc: Jesper Dangaard Brouer <hawk@kernel.org>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jonathan Lemon <jonathan.lemon@gmail.com>
Cc: KP Singh <kpsingh@kernel.org>
Cc: Maciej Fijalkowski <maciej.fijalkowski@intel.com>
Cc: Magnus Karlsson <magnus.karlsson@intel.com>
Cc: Martin KaFai Lau <martin.lau@linux.dev>
Cc: Song Liu <song@kernel.org>
Cc: Stanislav Fomichev <sdf@google.com>
Cc: Toke Høiland-Jørgensen <toke@redhat.com>
Cc: Yonghong Song <yonghong.song@linux.dev>
Cc: bpf@vger.kernel.org
Reviewed-by: Toke Høiland-Jørgensen <toke@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 include/linux/filter.h | 42 ++++++++++++++++++++++++++++++++++++++++++
 kernel/bpf/cpumap.c    | 19 +++----------------
 kernel/bpf/devmap.c    | 11 +++--------
 net/xdp/xsk.c          | 12 ++++--------
 4 files changed, 52 insertions(+), 32 deletions(-)

diff --git a/include/linux/filter.h b/include/linux/filter.h
index 0a7f6e4a00b60..c0349522de8fb 100644
--- a/include/linux/filter.h
+++ b/include/linux/filter.h
@@ -736,6 +736,9 @@ struct bpf_nh_params {
 /* flags for bpf_redirect_info kern_flags */
 #define BPF_RI_F_RF_NO_DIRECT	BIT(0)	/* no napi_direct on return_frame */
 #define BPF_RI_F_RI_INIT	BIT(1)
+#define BPF_RI_F_CPU_MAP_INIT	BIT(2)
+#define BPF_RI_F_DEV_MAP_INIT	BIT(3)
+#define BPF_RI_F_XSK_MAP_INIT	BIT(4)
 
 struct bpf_redirect_info {
 	u64 tgt_index;
@@ -750,6 +753,9 @@ struct bpf_redirect_info {
 
 struct bpf_net_context {
 	struct bpf_redirect_info ri;
+	struct list_head cpu_map_flush_list;
+	struct list_head dev_map_flush_list;
+	struct list_head xskmap_map_flush_list;
 };
 
 static inline struct bpf_net_context *bpf_net_ctx_set(struct bpf_net_context *bpf_net_ctx)
@@ -787,6 +793,42 @@ static inline struct bpf_redirect_info *bpf_net_ctx_get_ri(void)
 	return &bpf_net_ctx->ri;
 }
 
+static inline struct list_head *bpf_net_ctx_get_cpu_map_flush_list(void)
+{
+	struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
+
+	if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_CPU_MAP_INIT)) {
+		INIT_LIST_HEAD(&bpf_net_ctx->cpu_map_flush_list);
+		bpf_net_ctx->ri.kern_flags |= BPF_RI_F_CPU_MAP_INIT;
+	}
+
+	return &bpf_net_ctx->cpu_map_flush_list;
+}
+
+static inline struct list_head *bpf_net_ctx_get_dev_flush_list(void)
+{
+	struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
+
+	if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_DEV_MAP_INIT)) {
+		INIT_LIST_HEAD(&bpf_net_ctx->dev_map_flush_list);
+		bpf_net_ctx->ri.kern_flags |= BPF_RI_F_DEV_MAP_INIT;
+	}
+
+	return &bpf_net_ctx->dev_map_flush_list;
+}
+
+static inline struct list_head *bpf_net_ctx_get_xskmap_flush_list(void)
+{
+	struct bpf_net_context *bpf_net_ctx = bpf_net_ctx_get();
+
+	if (!(bpf_net_ctx->ri.kern_flags & BPF_RI_F_XSK_MAP_INIT)) {
+		INIT_LIST_HEAD(&bpf_net_ctx->xskmap_map_flush_list);
+		bpf_net_ctx->ri.kern_flags |= BPF_RI_F_XSK_MAP_INIT;
+	}
+
+	return &bpf_net_ctx->xskmap_map_flush_list;
+}
+
 /* Compute the linear packet data range [data, data_end) which
  * will be accessed by various program types (cls_bpf, act_bpf,
  * lwt, ...). Subsystems allowing direct data access must (!)
diff --git a/kernel/bpf/cpumap.c b/kernel/bpf/cpumap.c
index 66974bd027109..068e994ed781a 100644
--- a/kernel/bpf/cpumap.c
+++ b/kernel/bpf/cpumap.c
@@ -79,8 +79,6 @@ struct bpf_cpu_map {
 	struct bpf_cpu_map_entry __rcu **cpu_map;
 };
 
-static DEFINE_PER_CPU(struct list_head, cpu_map_flush_list);
-
 static struct bpf_map *cpu_map_alloc(union bpf_attr *attr)
 {
 	u32 value_size = attr->value_size;
@@ -709,7 +707,7 @@ static void bq_flush_to_queue(struct xdp_bulk_queue *bq)
  */
 static void bq_enqueue(struct bpf_cpu_map_entry *rcpu, struct xdp_frame *xdpf)
 {
-	struct list_head *flush_list = this_cpu_ptr(&cpu_map_flush_list);
+	struct list_head *flush_list = bpf_net_ctx_get_cpu_map_flush_list();
 	struct xdp_bulk_queue *bq = this_cpu_ptr(rcpu->bulkq);
 
 	if (unlikely(bq->count == CPU_MAP_BULK_SIZE))
@@ -761,7 +759,7 @@ int cpu_map_generic_redirect(struct bpf_cpu_map_entry *rcpu,
 
 void __cpu_map_flush(void)
 {
-	struct list_head *flush_list = this_cpu_ptr(&cpu_map_flush_list);
+	struct list_head *flush_list = bpf_net_ctx_get_cpu_map_flush_list();
 	struct xdp_bulk_queue *bq, *tmp;
 
 	list_for_each_entry_safe(bq, tmp, flush_list, flush_node) {
@@ -775,20 +773,9 @@ void __cpu_map_flush(void)
 #ifdef CONFIG_DEBUG_NET
 bool cpu_map_check_flush(void)
 {
-	if (list_empty(this_cpu_ptr(&cpu_map_flush_list)))
+	if (list_empty(bpf_net_ctx_get_cpu_map_flush_list()))
 		return false;
 	__cpu_map_flush();
 	return true;
 }
 #endif
-
-static int __init cpu_map_init(void)
-{
-	int cpu;
-
-	for_each_possible_cpu(cpu)
-		INIT_LIST_HEAD(&per_cpu(cpu_map_flush_list, cpu));
-	return 0;
-}
-
-subsys_initcall(cpu_map_init);
diff --git a/kernel/bpf/devmap.c b/kernel/bpf/devmap.c
index fbfdfb60db8d7..317ac2d66ebd1 100644
--- a/kernel/bpf/devmap.c
+++ b/kernel/bpf/devmap.c
@@ -83,7 +83,6 @@ struct bpf_dtab {
 	u32 n_buckets;
 };
 
-static DEFINE_PER_CPU(struct list_head, dev_flush_list);
 static DEFINE_SPINLOCK(dev_map_lock);
 static LIST_HEAD(dev_map_list);
 
@@ -415,7 +414,7 @@ static void bq_xmit_all(struct xdp_dev_bulk_queue *bq, u32 flags)
  */
 void __dev_flush(void)
 {
-	struct list_head *flush_list = this_cpu_ptr(&dev_flush_list);
+	struct list_head *flush_list = bpf_net_ctx_get_dev_flush_list();
 	struct xdp_dev_bulk_queue *bq, *tmp;
 
 	list_for_each_entry_safe(bq, tmp, flush_list, flush_node) {
@@ -429,7 +428,7 @@ void __dev_flush(void)
 #ifdef CONFIG_DEBUG_NET
 bool dev_check_flush(void)
 {
-	if (list_empty(this_cpu_ptr(&dev_flush_list)))
+	if (list_empty(bpf_net_ctx_get_dev_flush_list()))
 		return false;
 	__dev_flush();
 	return true;
@@ -460,7 +459,7 @@ static void *__dev_map_lookup_elem(struct bpf_map *map, u32 key)
 static void bq_enqueue(struct net_device *dev, struct xdp_frame *xdpf,
 		       struct net_device *dev_rx, struct bpf_prog *xdp_prog)
 {
-	struct list_head *flush_list = this_cpu_ptr(&dev_flush_list);
+	struct list_head *flush_list = bpf_net_ctx_get_dev_flush_list();
 	struct xdp_dev_bulk_queue *bq = this_cpu_ptr(dev->xdp_bulkq);
 
 	if (unlikely(bq->count == DEV_MAP_BULK_SIZE))
@@ -1160,15 +1159,11 @@ static struct notifier_block dev_map_notifier = {
 
 static int __init dev_map_init(void)
 {
-	int cpu;
-
 	/* Assure tracepoint shadow struct _bpf_dtab_netdev is in sync */
 	BUILD_BUG_ON(offsetof(struct bpf_dtab_netdev, dev) !=
 		     offsetof(struct _bpf_dtab_netdev, dev));
 	register_netdevice_notifier(&dev_map_notifier);
 
-	for_each_possible_cpu(cpu)
-		INIT_LIST_HEAD(&per_cpu(dev_flush_list, cpu));
 	return 0;
 }
 
diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index 7d1c0986f9bb3..ed062e0383896 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -35,8 +35,6 @@
 #define TX_BATCH_SIZE 32
 #define MAX_PER_SOCKET_BUDGET (TX_BATCH_SIZE)
 
-static DEFINE_PER_CPU(struct list_head, xskmap_flush_list);
-
 void xsk_set_rx_need_wakeup(struct xsk_buff_pool *pool)
 {
 	if (pool->cached_need_wakeup & XDP_WAKEUP_RX)
@@ -372,7 +370,7 @@ static int xsk_rcv(struct xdp_sock *xs, struct xdp_buff *xdp)
 
 int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp)
 {
-	struct list_head *flush_list = this_cpu_ptr(&xskmap_flush_list);
+	struct list_head *flush_list = bpf_net_ctx_get_xskmap_flush_list();
 	int err;
 
 	err = xsk_rcv(xs, xdp);
@@ -387,7 +385,7 @@ int __xsk_map_redirect(struct xdp_sock *xs, struct xdp_buff *xdp)
 
 void __xsk_map_flush(void)
 {
-	struct list_head *flush_list = this_cpu_ptr(&xskmap_flush_list);
+	struct list_head *flush_list = bpf_net_ctx_get_xskmap_flush_list();
 	struct xdp_sock *xs, *tmp;
 
 	list_for_each_entry_safe(xs, tmp, flush_list, flush_node) {
@@ -399,7 +397,7 @@ void __xsk_map_flush(void)
 #ifdef CONFIG_DEBUG_NET
 bool xsk_map_check_flush(void)
 {
-	if (list_empty(this_cpu_ptr(&xskmap_flush_list)))
+	if (list_empty(bpf_net_ctx_get_xskmap_flush_list()))
 		return false;
 	__xsk_map_flush();
 	return true;
@@ -1772,7 +1770,7 @@ static struct pernet_operations xsk_net_ops = {
 
 static int __init xsk_init(void)
 {
-	int err, cpu;
+	int err;
 
 	err = proto_register(&xsk_proto, 0 /* no slab */);
 	if (err)
@@ -1790,8 +1788,6 @@ static int __init xsk_init(void)
 	if (err)
 		goto out_pernet;
 
-	for_each_possible_cpu(cpu)
-		INIT_LIST_HEAD(&per_cpu(xskmap_flush_list, cpu));
 	return 0;
 
 out_pernet:
-- 
2.45.2

Re: [PATCH v7 net-next 15/15] net: Move per-CPU flush-lists to bpf_net_context on PREEMPT_RT.

[Jesper Dangaard Brouer]

On 18/06/2024 09.13, Sebastian Andrzej Siewior wrote:
> The per-CPU flush lists, which are accessed from within the NAPI callback
> (xdp_do_flush() for instance), are per-CPU. There are subject to the
> same problem as struct bpf_redirect_info.
> 
> Add the per-CPU lists cpu_map_flush_list, dev_map_flush_list and
> xskmap_map_flush_list to struct bpf_net_context. Add wrappers for the
> access. The lists initialized on first usage (similar to
> bpf_net_ctx_get_ri()).
> 
> Cc: "Björn Töpel"<bjorn@kernel.org>
> Cc: Alexei Starovoitov<ast@kernel.org>
> Cc: Andrii Nakryiko<andrii@kernel.org>
> Cc: Eduard Zingerman<eddyz87@gmail.com>
> Cc: Hao Luo<haoluo@google.com>
> Cc: Jesper Dangaard Brouer<hawk@kernel.org>
> Cc: Jiri Olsa<jolsa@kernel.org>
> Cc: John Fastabend<john.fastabend@gmail.com>
> Cc: Jonathan Lemon<jonathan.lemon@gmail.com>
> Cc: KP Singh<kpsingh@kernel.org>
> Cc: Maciej Fijalkowski<maciej.fijalkowski@intel.com>
> Cc: Magnus Karlsson<magnus.karlsson@intel.com>
> Cc: Martin KaFai Lau<martin.lau@linux.dev>
> Cc: Song Liu<song@kernel.org>
> Cc: Stanislav Fomichev<sdf@google.com>
> Cc: Toke Høiland-Jørgensen<toke@redhat.com>
> Cc: Yonghong Song<yonghong.song@linux.dev>
> Cc:bpf@vger.kernel.org
> Reviewed-by: Toke Høiland-Jørgensen<toke@redhat.com>
> Signed-off-by: Sebastian Andrzej Siewior<bigeasy@linutronix.de>

Acked-by: Jesper Dangaard Brouer <hawk@kernel.org>

> ---
>   include/linux/filter.h | 42 ++++++++++++++++++++++++++++++++++++++++++
>   kernel/bpf/cpumap.c    | 19 +++----------------
>   kernel/bpf/devmap.c    | 11 +++--------
>   net/xdp/xsk.c          | 12 ++++--------
>   4 files changed, 52 insertions(+), 32 deletions(-)

Re: [PATCH v7 net-next 00/15] locking: Introduce nested-BH locking.

[Jakub Kicinski]

On Tue, 18 Jun 2024 09:13:16 +0200 Sebastian Andrzej Siewior wrote:
> Disabling bottoms halves acts as per-CPU BKL. On PREEMPT_RT code within
> local_bh_disable() section remains preemtible. As a result high prior
> tasks (or threaded interrupts) will be blocked by lower-prio task (or
> threaded interrupts) which are long running which includes softirq
> sections.

Trivial conflict on patch 8 but enough to make the build bots give up..
Please rebase?