towards fixing intermingled dmesg

[Andriy Gapon]

Here is a simple extension of msgbuf_addchar that is supposed to allow to put N
characters consecutively into msgbuf.
The idea is very simple: atomically forward msg_wseq by N and then copy
characters into the reserved space.

Could you please check if this is implemented correctly and if it would actually
work the way I expect?

There is still a harder issue of actually plugging this routine into subr_prf.c :-)

--- a/sys/kern/subr_msgbuf.c
+++ b/sys/kern/subr_msgbuf.c
@@ -131,6 +131,26 @@ msgbuf_addchar(struct msgbuf *mbp, int c)
 	mbp->msg_ptr[pos] = c;
 }

+void
+msgbuf_addchars(struct msgbuf *mbp, const char *s, const int n)
+{
+	u_int new_seq, pos, seq;
+	int c, i;
+
+	do {
+		seq = mbp->msg_wseq;
+		new_seq = MSGBUF_SEQNORM(mbp, seq + n);
+	} while (atomic_cmpset_rel_int(&mbp->msg_wseq, seq, new_seq) == 0);
+
+	for (i = 0; i < n; i++) {
+		pos = MSGBUF_SEQ_TO_POS(mbp, seq + i);
+		c = s[i];
+		atomic_add_int(&mbp->msg_cksum, (u_int)(u_char)c -
+		    (u_int)(u_char)mbp->msg_ptr[pos]);
+		mbp->msg_ptr[pos] = c;
+	}
+}
+
 /*
  * Read and mark as read a character from a message buffer.
  * Returns the character, or -1 if no characters are available.

-- 
Andriy Gapon