uml: fix console writing bugs

The previous console cleanup patch switched generic_read and generic_write
from calling os_{read,write}_file to calling read and write directly.  Because
the calling convention is different, they now need to get any error from errno
rather than the return value.  I did this for generic_read, but forgot about
generic_write.

While chasing some output corruption, I noticed that line_write was
unnecessarily calling flush_buffer, and deleted it.  I don't understand why,
but the corruption disappeared.  This is unneeded because there already is a
perfectly good mechanism for finding out when the host output device has some
room to write data - there is an interrupt that comes in when writes can
happen again.  line_write calling flush_buffer seemed to just be an attempt to
opportunistically get some data out to the host.

I also made write_chan short-circuit calling into the host-level code for
zero-length writes.  Calling libc write with a length of zero conflated write
not being able to write anything with asking it not to write anything.  Better
to just cut it off as soon as possible.

Signed-off-by: Jeff Dike <jdike@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

1 files changed, 10 insertions, 1 deletions

diff --git a/arch/um/drivers/chan_user.c b/arch/um/drivers/chan_user.c
index 77557e233f58..d29e56d8ced4 100644
--- a/arch/um/drivers/chan_user.c
+++ b/arch/um/drivers/chan_user.c
@@ -38,7 +38,16 @@ int generic_read(int fd, char *c_out, void *unused)
 
 int generic_write(int fd, const char *buf, int n, void *unused)
 {
-        return write(fd, buf, n);
+        int err;
+
+        err = write(fd, buf, n);
+        if (err > 0)
+                return err;
+        else if (errno == EAGAIN)
+                return 0;
+        else if (err == 0)
+                return -EIO;
+        return -errno;
 }
 
 int generic_window_size(int fd, void *unused, unsigned short *rows_out,