On Thu, Mar 31, 2016 at 3:28 PM, Stefan Hajnoczi <stefa...@gmail.com> wrote:
> On Mon, Mar 28, 2016 at 06:04:11PM +0530, Pooja Dhannawat wrote:
> > nc_sendv_compat has a huge stack usage of 69680 bytes approx.
> > Moving large arrays to heap to reduce stack usage.
> >
> > Reviewed-by: Stefan Hajnoczi <stefa...@redhat.com>
> > Signed-off-by: Pooja Dhannawat <dhannawatpoo...@gmail.com>
> > ---
> > net/net.c | 13 +++++++++----
> > 1 file changed, 9 insertions(+), 4 deletions(-)
> >
> > diff --git a/net/net.c b/net/net.c
> > index b0c832e..663da13 100644
> > --- a/net/net.c
> > +++ b/net/net.c
> > @@ -709,23 +709,28 @@ ssize_t qemu_send_packet_raw(NetClientState *nc,
> const uint8_t *buf, int size)
> > static ssize_t nc_sendv_compat(NetClientState *nc, const struct iovec
> *iov,
> > int iovcnt, unsigned flags)
> > {
> > - uint8_t buf[NET_BUFSIZE];
> > + uint8_t *buf = NULL;
> > uint8_t *buffer;
> > size_t offset;
> > + ssize_t ret;
> >
> > if (iovcnt == 1) {
> > buffer = iov[0].iov_base;
> > offset = iov[0].iov_len;
> > } else {
> > + buf = g_new(uint8_t, NET_BUFSIZE);
> > buffer = buf;
> > - offset = iov_to_buf(iov, iovcnt, 0, buf, sizeof(buf));
> > + offset = iov_to_buf(iov, iovcnt, 0, buf, NET_BUFSIZE);
> > }
> >
> > if (flags & QEMU_NET_PACKET_FLAG_RAW && nc->info->receive_raw) {
> > - return nc->info->receive_raw(nc, buffer, offset);
> > + ret = nc->info->receive_raw(nc, buffer, offset);
> > } else {
> > - return nc->info->receive(nc, buffer, offset);
> > + ret = nc->info->receive(nc, buffer, offset);
> > }
> > +
> > + g_free(buf);
> > + return ret;
> > }
> >
> > ssize_t qemu_deliver_packet_iov(NetClientState *sender,
> > --
> > 2.5.0
>
> CCing Jason Wang, net subsystem maintainer. Please use
> scripts/get_maintainer.pl to find the right people to CC in future
> patches.
>
> Yes, I am sorry about that. Fam also pointed out this thing but I forgot
this time also.
I will keep that mind from onwards.
> We may also want to keep a smaller stack buffer so that reasonably-sized
> packets (e.g. up to 2 KB) can be send without the performance cost of
> g_malloc().
>
