Comments
Patch
@@ -36,14 +36,13 @@ static int vmc_write(SpiceCharDeviceInstance *sin, const uint8_t *buf, int len)
while (len > 0) {
last_out = MIN(len, VMC_MAX_HOST_WRITE);
- qemu_chr_read(scd->chr, p, last_out);
- if (last_out > 0) {
- out += last_out;
- len -= last_out;
- p += last_out;
- } else {
+ if (qemu_chr_can_read(scd->chr) < last_out) {
break;
}
+ qemu_chr_read(scd->chr, p, last_out);
+ out += last_out;
+ len -= last_out;
+ p += last_out;
}
dprintf(scd, 3, "%s: %lu/%zd\n", __func__, out, len + out);
In the old spice-vmc device we used to have: last_out = virtio_serial_write(&svc->port, p, MIN(len, VMC_MAX_HOST_WRITE)); if (last_out > 0) ... Now in the chardev backend we have: last_out = MIN(len, VMC_MAX_HOST_WRITE); qemu_chr_read(scd->chr, p, last_out); if (last_out > 0) { ... Which causes us to no longer detect if the virtio port is not ready to receive data from us. chardev actually has a mechanism to detect this, but it requires a separate call to qemu_chr_can_read, before calling qemu_chr_read (which return void). This patch uses qemu_chr_can_read to fix the flow control from client to guest. Signed-off-by: Hans de Goede <hdegoede@redhat.com> --- spice-qemu-char.c | 11 +++++------ 1 files changed, 5 insertions(+), 6 deletions(-)