printk: Wait for all reserved records with pr_flush()

Currently pr_flush() will only wait for records that were
available to readers at the time of the call (using
prb_next_seq()). But there may be more records (non-finalized)
that have following finalized records. pr_flush() should wait
for these to print as well. Particularly because any trailing
finalized records may be the messages that the calling context
wants to ensure are printed.

Add a new ringbuffer function prb_next_reserve_seq() to return
the sequence number following the most recently reserved record.
This guarantees that pr_flush() will wait until all current
printk() messages (completed or in progress) have been printed.

Fixes: 3b604ca812 ("printk: add pr_flush()")
Signed-off-by: John Ogness <john.ogness@linutronix.de>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Link: https://lore.kernel.org/r/20240207134103.1357162-10-john.ogness@linutronix.de
Signed-off-by: Petr Mladek <pmladek@suse.com>

kernel/printk/printk.c

@@ -3755,7 +3755,7 @@ static bool __pr_flush(struct console *con, int timeout_ms, bool reset_on_progre
 
 	might_sleep();
 
-	seq = prb_next_seq(prb);
+	seq = prb_next_reserve_seq(prb);
 
 	/* Flush the consoles so that records up to @seq are printed. */
 	console_lock();

kernel/printk/printk_ringbuffer.c

@@ -1986,6 +1986,111 @@ u64 prb_first_seq(struct printk_ringbuffer *rb)
 	return seq;
 }
 
+/**
+ * prb_next_reserve_seq() - Get the sequence number after the most recently
+ *                  reserved record.
+ *
+ * @rb:  The ringbuffer to get the sequence number from.
+ *
+ * This is the public function available to readers to see what sequence
+ * number will be assigned to the next reserved record.
+ *
+ * Note that depending on the situation, this value can be equal to or
+ * higher than the sequence number returned by prb_next_seq().
+ *
+ * Context: Any context.
+ * Return: The sequence number that will be assigned to the next record
+ *         reserved.
+ */
+u64 prb_next_reserve_seq(struct printk_ringbuffer *rb)
+{
+	struct prb_desc_ring *desc_ring = &rb->desc_ring;
+	unsigned long last_finalized_id;
+	atomic_long_t *state_var;
+	u64 last_finalized_seq;
+	unsigned long head_id;
+	struct prb_desc desc;
+	unsigned long diff;
+	struct prb_desc *d;
+	int err;
+
+	/*
+	 * It may not be possible to read a sequence number for @head_id.
+	 * So the ID of @last_finailzed_seq is used to calculate what the
+	 * sequence number of @head_id will be.
+	 */
+
+try_again:
+	last_finalized_seq = desc_last_finalized_seq(rb);
+
+	/*
+	 * @head_id is loaded after @last_finalized_seq to ensure that
+	 * it points to the record with @last_finalized_seq or newer.
+	 *
+	 * Memory barrier involvement:
+	 *
+	 * If desc_last_finalized_seq:A reads from
+	 * desc_update_last_finalized:A, then
+	 * prb_next_reserve_seq:A reads from desc_reserve:D.
+	 *
+	 * Relies on:
+	 *
+	 * RELEASE from desc_reserve:D to desc_update_last_finalized:A
+	 *    matching
+	 * ACQUIRE from desc_last_finalized_seq:A to prb_next_reserve_seq:A
+	 *
+	 * Note: desc_reserve:D and desc_update_last_finalized:A can be
+	 *       different CPUs. However, the desc_update_last_finalized:A CPU
+	 *       (which performs the release) must have previously seen
+	 *       desc_read:C, which implies desc_reserve:D can be seen.
+	 */
+	head_id = atomic_long_read(&desc_ring->head_id); /* LMM(prb_next_reserve_seq:A) */
+
+	d = to_desc(desc_ring, last_finalized_seq);
+	state_var = &d->state_var;
+
+	/* Extract the ID, used to specify the descriptor to read. */
+	last_finalized_id = DESC_ID(atomic_long_read(state_var));
+
+	/* Ensure @last_finalized_id is correct. */
+	err = desc_read_finalized_seq(desc_ring, last_finalized_id, last_finalized_seq, &desc);
+
+	if (err == -EINVAL) {
+		if (last_finalized_seq == 0) {
+			/*
+			 * No record has been finalized or even reserved yet.
+			 *
+			 * The @head_id is initialized such that the first
+			 * increment will yield the first record (seq=0).
+			 * Handle it separately to avoid a negative @diff
+			 * below.
+			 */
+			if (head_id == DESC0_ID(desc_ring->count_bits))
+				return 0;
+
+			/*
+			 * One or more descriptors are already reserved. Use
+			 * the descriptor ID of the first one (@seq=0) for
+			 * the @diff below.
+			 */
+			last_finalized_id = DESC0_ID(desc_ring->count_bits) + 1;
+		} else {
+			/* Record must have been overwritten. Try again. */
+			goto try_again;
+		}
+	}
+
+	/* Diff of known descriptor IDs to compute related sequence numbers. */
+	diff = head_id - last_finalized_id;
+
+	/*
+	 * @head_id points to the most recently reserved record, but this
+	 * function returns the sequence number that will be assigned to the
+	 * next (not yet reserved) record. Thus +1 is needed.
+	 */
+	return (last_finalized_seq + diff + 1);
+}
+
 /*
  * Non-blocking read of a record.
  *

kernel/printk/printk_ringbuffer.h

@@ -395,6 +395,7 @@ bool prb_read_valid_info(struct printk_ringbuffer *rb, u64 seq,
 u64 prb_first_seq(struct printk_ringbuffer *rb);
 u64 prb_first_valid_seq(struct printk_ringbuffer *rb);
 u64 prb_next_seq(struct printk_ringbuffer *rb);
+u64 prb_next_reserve_seq(struct printk_ringbuffer *rb);
 
 #ifdef CONFIG_64BIT