Kevin Wolf wrote:
> [Crossposting to xen-devel]
>
> Ian, we need something like this for qemu-xen (or ioemu-remote or
> whatever it's called now). Currently you must attach to the console of a
> domain, otherwise it won't boot up and keep hanging in a blocking write
> because the buffer is full.
>
> The old ioemu had a hack in unix_write (doing a select before the write)
> which you didn't merge into qemu-xen. In fact, I noticed that you even
> removed that function entirely and I'm wondering why.
For completeness: You also need the attached patch for unix_write,
otherwise you'll end up with qemu burning cpu cycles. If you can't
write to a non-blocking file handle the write will instantly return with
-EAGAIN. Calling it again of course doesn't change the result, so
better don't do that ...
--
http://kraxel.fedorapeople.org/xenner/
>From d9454802fa2105bc399518eebfa1e1415ff07143 Mon Sep 17 00:00:00 2001
From: Gerd Hoffmann <kraxel@xxxxxxxxxx>
Date: Wed, 9 Jul 2008 09:41:31 +0200
Subject: [PATCH 02/12] unix_write: don't block on non-blocking file handles.
Signed-off-by: Gerd Hoffmann <kraxel@xxxxxxxxxx>
---
vl.c | 12 +++++++++++-
1 files changed, 11 insertions(+), 1 deletions(-)
diff --git a/vl.c b/vl.c
index c8db579..2e2d883 100644
--- a/vl.c
+++ b/vl.c
@@ -2116,14 +2116,24 @@ void socket_set_nonblock(int fd)
static int unix_write(int fd, const uint8_t *buf, int len1)
{
+ int nonblock = fcntl(fd, F_GETFL) & O_NONBLOCK;
int ret, len;
len = len1;
while (len > 0) {
ret = write(fd, buf, len);
if (ret < 0) {
- if (errno != EINTR && errno != EAGAIN)
+ if (errno == EINTR) {
+ continue;
+ } else if (errno == EAGAIN) {
+ if (!nonblock)
+ continue;
+ if (len1 != len)
+ break; /* partial write, return written bytes */
+ return -1;
+ } else {
return -1;
+ }
} else if (ret == 0) {
break;
} else {
--
1.5.4.1
_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel
|
Home