diff mbox series

[34/33] netfs: Use in_interrupt() not in_softirq()

Message ID 1376938.1613429183@warthog.procyon.org.uk (mailing list archive)
State New
Headers show
Series Network fs helper library & fscache kiocb API [ver #3] | expand

Commit Message

David Howells Feb. 15, 2021, 10:46 p.m. UTC
The in_softirq() in netfs_rreq_terminated() works fine for the cache being
on a normal disk, as the completion handlers may get called in softirq
context, but for an NVMe drive, the completion handler may get called in
IRQ context.

Fix to use in_interrupt() instead of in_softirq() throughout the read
helpers, particularly when deciding whether to punt code that might sleep
off to a worker thread.

The symptom involves warnings like the following appearing and the kernel
hanging:

 WARNING: CPU: 0 PID: 0 at kernel/softirq.c:175 __local_bh_enable_ip+0x35/0x50
 ...
 RIP: 0010:__local_bh_enable_ip+0x35/0x50
 ...
 Call Trace:
  <IRQ>
  rxrpc_kernel_begin_call+0x7d/0x1b0 [rxrpc]
  ? afs_rx_new_call+0x40/0x40 [kafs]
  ? afs_alloc_call+0x28/0x120 [kafs]
  afs_make_call+0x120/0x510 [kafs]
  ? afs_rx_new_call+0x40/0x40 [kafs]
  ? afs_alloc_flat_call+0xba/0x100 [kafs]
  ? __kmalloc+0x167/0x2f0
  ? afs_alloc_flat_call+0x9b/0x100 [kafs]
  afs_wait_for_operation+0x2d/0x200 [kafs]
  afs_do_sync_operation+0x16/0x20 [kafs]
  afs_req_issue_op+0x8c/0xb0 [kafs]
  netfs_rreq_assess+0x125/0x7d0 [netfs]
  ? cachefiles_end_operation+0x40/0x40 [cachefiles]
  netfs_subreq_terminated+0x117/0x220 [netfs]
  cachefiles_read_complete+0x21/0x60 [cachefiles]
  iomap_dio_bio_end_io+0xdd/0x110
  blk_update_request+0x20a/0x380
  blk_mq_end_request+0x1c/0x120
  nvme_process_cq+0x159/0x1f0 [nvme]
  nvme_irq+0x10/0x20 [nvme]
  __handle_irq_event_percpu+0x37/0x150
  handle_irq_event+0x49/0xb0
  handle_edge_irq+0x7c/0x200
  asm_call_irq_on_stack+0xf/0x20
  </IRQ>
  common_interrupt+0xad/0x120
  asm_common_interrupt+0x1e/0x40
 ...

Reported-by: Marc Dionne <marc.dionne@auristor.com>
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Matthew Wilcox <willy@infradead.org>
cc: linux-mm@kvack.org
cc: linux-cachefs@redhat.com
cc: linux-afs@lists.infradead.org
cc: linux-nfs@vger.kernel.org
cc: linux-cifs@vger.kernel.org
cc: ceph-devel@vger.kernel.org
cc: v9fs-developer@lists.sourceforge.net
cc: linux-fsdevel@vger.kernel.org
---
 read_helper.c |    8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

Comments

Christoph Hellwig Feb. 16, 2021, 8:42 a.m. UTC | #1
On Mon, Feb 15, 2021 at 10:46:23PM +0000, David Howells wrote:
> The in_softirq() in netfs_rreq_terminated() works fine for the cache being
> on a normal disk, as the completion handlers may get called in softirq
> context, but for an NVMe drive, the completion handler may get called in
> IRQ context.
> 
> Fix to use in_interrupt() instead of in_softirq() throughout the read
> helpers, particularly when deciding whether to punt code that might sleep
> off to a worker thread.

We must not use either check, as they all are unreliable especially
for PREEMPT-RT.
Sebastian Andrzej Siewior Feb. 16, 2021, 9:06 a.m. UTC | #2
On 2021-02-16 09:42:30 [+0100], Christoph Hellwig wrote:
> On Mon, Feb 15, 2021 at 10:46:23PM +0000, David Howells wrote:
> > The in_softirq() in netfs_rreq_terminated() works fine for the cache being
> > on a normal disk, as the completion handlers may get called in softirq
> > context, but for an NVMe drive, the completion handler may get called in
> > IRQ context.
> > 
> > Fix to use in_interrupt() instead of in_softirq() throughout the read
> > helpers, particularly when deciding whether to punt code that might sleep
> > off to a worker thread.
> 
> We must not use either check, as they all are unreliable especially
> for PREEMPT-RT.

Yes, please. I try to cleanup the users one by one
    https://lore.kernel.org/r/20200914204209.256266093@linutronix.de/
    https://lore.kernel.org/amd-gfx/20210209124439.408140-1-bigeasy@linutronix.de/

Sebastian
David Howells Feb. 16, 2021, 9:29 a.m. UTC | #3
Christoph Hellwig <hch@lst.de> wrote:

> On Mon, Feb 15, 2021 at 10:46:23PM +0000, David Howells wrote:
> > The in_softirq() in netfs_rreq_terminated() works fine for the cache being
> > on a normal disk, as the completion handlers may get called in softirq
> > context, but for an NVMe drive, the completion handler may get called in
> > IRQ context.
> > 
> > Fix to use in_interrupt() instead of in_softirq() throughout the read
> > helpers, particularly when deciding whether to punt code that might sleep
> > off to a worker thread.
> 
> We must not use either check, as they all are unreliable especially
> for PREEMPT-RT.

Is there a better way to do it?  The intent is to process the assessment phase
in the calling thread's context if possible rather than bumping over to a
worker thread.  For synchronous I/O, for example, that's done in the caller's
thread.  Maybe that's the answer - if it's known to be asynchronous, I have to
punt, but otherwise don't have to.

David
Christoph Hellwig Feb. 16, 2021, 9:30 a.m. UTC | #4
On Tue, Feb 16, 2021 at 09:29:31AM +0000, David Howells wrote:
> Is there a better way to do it?  The intent is to process the assessment phase
> in the calling thread's context if possible rather than bumping over to a
> worker thread.  For synchronous I/O, for example, that's done in the caller's
> thread.  Maybe that's the answer - if it's known to be asynchronous, I have to
> punt, but otherwise don't have to.

Yes, i think you want an explicit flag instead.
diff mbox series

Patch

diff --git a/fs/netfs/read_helper.c b/fs/netfs/read_helper.c
index 9191a3617d91..db582008b4bd 100644
--- a/fs/netfs/read_helper.c
+++ b/fs/netfs/read_helper.c
@@ -96,7 +96,7 @@  static void netfs_free_read_request(struct work_struct *work)
 static void netfs_put_read_request(struct netfs_read_request *rreq)
 {
 	if (refcount_dec_and_test(&rreq->usage)) {
-		if (in_softirq()) {
+		if (in_interrupt()) {
 			rreq->work.func = netfs_free_read_request;
 			if (!queue_work(system_unbound_wq, &rreq->work))
 				BUG();
@@ -353,7 +353,7 @@  static void netfs_rreq_write_to_cache_work(struct work_struct *work)
 
 static void netfs_rreq_write_to_cache(struct netfs_read_request *rreq)
 {
-	if (in_softirq()) {
+	if (in_interrupt()) {
 		rreq->work.func = netfs_rreq_write_to_cache_work;
 		if (!queue_work(system_unbound_wq, &rreq->work))
 			BUG();
@@ -479,7 +479,7 @@  static bool netfs_rreq_perform_resubmissions(struct netfs_read_request *rreq)
 {
 	struct netfs_read_subrequest *subreq;
 
-	WARN_ON(in_softirq());
+	WARN_ON(in_interrupt());
 
 	trace_netfs_rreq(rreq, netfs_rreq_trace_resubmit);
 
@@ -577,7 +577,7 @@  static void netfs_rreq_work(struct work_struct *work)
 static void netfs_rreq_terminated(struct netfs_read_request *rreq)
 {
 	if (test_bit(NETFS_RREQ_INCOMPLETE_IO, &rreq->flags) &&
-	    in_softirq()) {
+	    in_interrupt()) {
 		if (!queue_work(system_unbound_wq, &rreq->work))
 			BUG();
 	} else {