Skip to contentSkip to conversation

io_uring: add more uring info to fdinfo for debug

PATCH In review Hao Xu 3 participants 6 messages Sep 13, 2021 – Sep 16, 2021

Subsystems (2)
lore
Share & tools
lore

Hao Xu

author
#0 lore raw

PATCH 1/1 io_uring: add more uring info to fdinfo for debug

Developers may need some uring info to help themselves debug and address issues, these info includes sqring/cqring head/tail and the detail sqe/cqe info, which is very useful when it stucks.

---
 fs/io_uring.c | 59 +++++++++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 55 insertions(+), 4 deletions(-)
fs/io_uring.c +55 -4
index f795ad281038..ac048592a3e8 100644
@@ -9950,8 +9950,48 @@ static int io_uring_show_cred(struct seq_file *m, unsigned int id,
 static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
 {
 	struct io_sq_data *sq = NULL;
+	struct io_overflow_cqe *ocqe;
+	struct io_rings *r = ctx->rings;
+	unsigned int sq_mask = ctx->sq_entries - 1, cq_mask = ctx->cq_entries - 1;
+	unsigned int cached_sq_head = ctx->cached_sq_head;
+	unsigned int cached_cq_tail = ctx->cached_cq_tail;
+	unsigned int sq_head = READ_ONCE(r->sq.head);
+	unsigned int sq_tail = READ_ONCE(r->sq.tail);
+	unsigned int cq_head = READ_ONCE(r->cq.head);
+	unsigned int cq_tail = READ_ONCE(r->cq.tail);
 	bool has_lock;
-	int i;
+	unsigned int i;
+
+	/*
+	 * we may get imprecise sqe and cqe info if uring is actively running
+	 * since we get cached_sq_head and cached_cq_tail without uring_lock
+	 * and sq_tail and cq_head are changed by userspace. But it's ok since
+	 * we usually use these info when it is stuck.
+	 */
+	seq_printf(m, "SqHead:\t%u\n", sq_head & sq_mask);
+	seq_printf(m, "SqTail:\t%u\n", sq_tail & sq_mask);
+	seq_printf(m, "CachedSqHead:\t%u\n", cached_sq_head & sq_mask);
+	seq_printf(m, "CqHead:\t%u\n", cq_head & cq_mask);
+	seq_printf(m, "CqTail:\t%u\n", cq_tail & cq_mask);
+	seq_printf(m, "CachedCqTail:\t%u\n", cached_cq_tail & cq_mask);
+	seq_printf(m, "SQEs:\t%u\n", sq_tail - cached_sq_head);
+	for (i = cached_sq_head; i < sq_tail; i++) {
+		unsigned int sq_idx = READ_ONCE(ctx->sq_array[i & sq_mask]);
+
+		if (likely(sq_idx <= sq_mask)) {
+			struct io_uring_sqe *sqe = &ctx->sq_sqes[sq_idx];
+
+			seq_printf(m, "%5u: opcode:%d, fd:%d, flags:%x, user_data:%llu\n",
+				   sq_idx, sqe->opcode, sqe->fd, sqe->flags, sqe->user_data);
+		}
+	}
+	seq_printf(m, "CQEs:\t%u\n", cached_cq_tail - cq_head);
+	for (i = cq_head; i < cached_cq_tail; i++) {
+		struct io_uring_cqe *cqe = &r->cqes[i & cq_mask];
+
+		seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x\n",
+			   i & cq_mask, cqe->user_data, cqe->res, cqe->flags);
+	}
 
 	/*
 	 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
@@ -9993,7 +10033,10 @@ static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
 		xa_for_each(&ctx->personalities, index, cred)
 			io_uring_show_cred(m, index, cred);
 	}
-	seq_printf(m, "PollList:\n");
+	if (has_lock)
+		mutex_unlock(&ctx->uring_lock);
+
+	seq_puts(m, "PollList:\n");
 	spin_lock(&ctx->completion_lock);
 	for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
 		struct hlist_head *list = &ctx->cancel_hash[i];
@@ -10003,9 +10046,17 @@ static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
 			seq_printf(m, "  op=%d, task_works=%d\n", req->opcode,
 					req->task->task_works != NULL);
 	}
+
+	seq_puts(m, "CqOverflowList:\n");
+	list_for_each_entry(ocqe, &ctx->cq_overflow_list, list) {
+		struct io_uring_cqe *cqe = &ocqe->cqe;
+
+		seq_printf(m, "  user_data=%llu, res=%d, flags=%x\n",
+			   cqe->user_data, cqe->res, cqe->flags);
+
+	}
+
 	spin_unlock(&ctx->completion_lock);
-	if (has_lock)
-		mutex_unlock(&ctx->uring_lock);
 }
 
 static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
Signature · 1 line
2.24.4
author
#1 lore raw

ping...

Quoted 96 lines · Hao Xu

Developers may need some uring info to help themselves debug and address issues, these info includes sqring/cqring head/tail and the detail sqe/cqe info, which is very useful when it stucks.

---
  fs/io_uring.c | 59 +++++++++++++++++++++++++++++++++++++++++++++++----
  1 file changed, 55 insertions(+), 4 deletions(-)
fs/io_uring.c +55 -4
index f795ad281038..ac048592a3e8 100644
@@ -9950,8 +9950,48 @@ static int io_uring_show_cred(struct seq_file *m, unsigned int id,
  static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
  {
  	struct io_sq_data *sq = NULL;
+	struct io_overflow_cqe *ocqe;
+	struct io_rings *r = ctx->rings;
+	unsigned int sq_mask = ctx->sq_entries - 1, cq_mask = ctx->cq_entries - 1;
+	unsigned int cached_sq_head = ctx->cached_sq_head;
+	unsigned int cached_cq_tail = ctx->cached_cq_tail;
+	unsigned int sq_head = READ_ONCE(r->sq.head);
+	unsigned int sq_tail = READ_ONCE(r->sq.tail);
+	unsigned int cq_head = READ_ONCE(r->cq.head);
+	unsigned int cq_tail = READ_ONCE(r->cq.tail);
  	bool has_lock;
-	int i;
+	unsigned int i;
+
+	/*
+	 * we may get imprecise sqe and cqe info if uring is actively running
+	 * since we get cached_sq_head and cached_cq_tail without uring_lock
+	 * and sq_tail and cq_head are changed by userspace. But it's ok since
+	 * we usually use these info when it is stuck.
+	 */
+	seq_printf(m, "SqHead:\t%u\n", sq_head & sq_mask);
+	seq_printf(m, "SqTail:\t%u\n", sq_tail & sq_mask);
+	seq_printf(m, "CachedSqHead:\t%u\n", cached_sq_head & sq_mask);
+	seq_printf(m, "CqHead:\t%u\n", cq_head & cq_mask);
+	seq_printf(m, "CqTail:\t%u\n", cq_tail & cq_mask);
+	seq_printf(m, "CachedCqTail:\t%u\n", cached_cq_tail & cq_mask);
+	seq_printf(m, "SQEs:\t%u\n", sq_tail - cached_sq_head);
+	for (i = cached_sq_head; i < sq_tail; i++) {
+		unsigned int sq_idx = READ_ONCE(ctx->sq_array[i & sq_mask]);
+
+		if (likely(sq_idx <= sq_mask)) {
+			struct io_uring_sqe *sqe = &ctx->sq_sqes[sq_idx];
+
+			seq_printf(m, "%5u: opcode:%d, fd:%d, flags:%x, user_data:%llu\n",
+				   sq_idx, sqe->opcode, sqe->fd, sqe->flags, sqe->user_data);
+		}
+	}
+	seq_printf(m, "CQEs:\t%u\n", cached_cq_tail - cq_head);
+	for (i = cq_head; i < cached_cq_tail; i++) {
+		struct io_uring_cqe *cqe = &r->cqes[i & cq_mask];
+
+		seq_printf(m, "%5u: user_data:%llu, res:%d, flag:%x\n",
+			   i & cq_mask, cqe->user_data, cqe->res, cqe->flags);
+	}
 
  	/*
  	 * Avoid ABBA deadlock between the seq lock and the io_uring mutex,
@@ -9993,7 +10033,10 @@ static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
  		xa_for_each(&ctx->personalities, index, cred)
  			io_uring_show_cred(m, index, cred);
  	}
-	seq_printf(m, "PollList:\n");
+	if (has_lock)
+		mutex_unlock(&ctx->uring_lock);
+
+	seq_puts(m, "PollList:\n");
  	spin_lock(&ctx->completion_lock);
  	for (i = 0; i < (1U << ctx->cancel_hash_bits); i++) {
  		struct hlist_head *list = &ctx->cancel_hash[i];
@@ -10003,9 +10046,17 @@ static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
  			seq_printf(m, "  op=%d, task_works=%d\n", req->opcode,
  					req->task->task_works != NULL);
  	}
+
+	seq_puts(m, "CqOverflowList:\n");
+	list_for_each_entry(ocqe, &ctx->cq_overflow_list, list) {
+		struct io_uring_cqe *cqe = &ocqe->cqe;
+
+		seq_printf(m, "  user_data=%llu, res=%d, flags=%x\n",
+			   cqe->user_data, cqe->res, cqe->flags);
+
+	}
+
  	spin_unlock(&ctx->completion_lock);
-	if (has_lock)
-		mutex_unlock(&ctx->uring_lock);
  }
 
  static void io_uring_show_fdinfo(struct seq_file *m, struct file *f)
maintainer
#2 lore raw
In reply to Hao Xu
Quoted 3 lines · Hao Xu

Developers may need some uring info to help themselves debug and address issues, these info includes sqring/cqring head/tail and the detail sqe/cqe info, which is very useful when it stucks.

I think this is a good addition, more info to help you debug a stuck case is always good. I'll queue this up for 5.16.

Signature · 1 line
Jens Axboe
#3 lore raw
Quoted 7 lines · Jens Axboe
Quoted 3 lines · Hao Xu

Developers may need some uring info to help themselves debug and address issues, these info includes sqring/cqring head/tail and the detail sqe/cqe info, which is very useful when it stucks.

I think this is a good addition, more info to help you debug a stuck case is always good. I'll queue this up for 5.16.

Are there limits how much we can print? I remember people were couldn't even show a list of CPUs (was it proc?). The overflow list may be huge.

Signature · 1 line
Pavel Begunkov
maintainer
#4 lore raw
Quoted 11 lines · Pavel Begunkov
Quoted 7 lines · Jens Axboe
Quoted 3 lines · Hao Xu

Developers may need some uring info to help themselves debug and address issues, these info includes sqring/cqring head/tail and the detail sqe/cqe info, which is very useful when it stucks.

I think this is a good addition, more info to help you debug a stuck case is always good. I'll queue this up for 5.16.

Are there limits how much we can print? I remember people were couldn't even show a list of CPUs (was it proc?). The overflow list may be huge.

It's using seq_file, so I _think_ we should be fine here. Not sure when/if it truncates.

Signature · 1 line
Jens Axboe
author
#5 lore raw
Quoted 7 lines · Jens Axboe
Quoted 6 lines · Pavel Begunkov
Quoted 5 lines · Jens Axboe
Quoted 3 lines · Hao Xu

Developers may need some uring info to help themselves debug and address issues, these info includes sqring/cqring head/tail and the detail sqe/cqe info, which is very useful when it stucks.

Quoted 9 lines
Quoted 5 lines
Quoted 2 lines

I think this is a good addition, more info to help you debug a stuck case is always good. I'll queue this up for 5.16.

Are there limits how much we can print? I remember people were couldn't even show a list of CPUs (was it proc?). The overflow list may be huge.

It's using seq_file, so I _think_ we should be fine here. Not sure when/if it truncates.

It seems the default buf size of seq_file is PAGE_SIZE.