Re: [Qemu-devel] [PATCH v3] net: Allocating Large sized arrays to heap

[Stefan Hajnoczi]

On Tue, Mar 15, 2016 at 11:55:13PM +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.
> 
> Signed-off-by: Pooja Dhannawat 
> ---
>  net/net.c | 14 ++
>  1 file changed, 10 insertions(+), 4 deletions(-)
> 
> diff --git a/net/net.c b/net/net.c
> index b0c832e..f03c571 100644
> --- a/net/net.c
> +++ b/net/net.c
> @@ -709,23 +709,29 @@ 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;
>  uint8_t *buffer;
>  size_t offset;
> +ssize_t ret;
> +
> +buf = g_new(uint8_t, NET_BUFSIZE);

The linear buffer is only needed when iovcnt > 1.  I suggest the
following instead:

uint8_t *buf = NULL;

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, NET_BUFSIZE);
}

This way the allocation is only made when we actually need to linearize
the buffer.


signature.asc
Description: PGP signature

Re: [Qemu-devel] [PATCH v3] net: Allocating Large sized arrays to heap

[Pooja Dhannawat]

On Thu, Mar 17, 2016 at 8:23 PM, Stefan Hajnoczi  wrote:

> On Tue, Mar 15, 2016 at 11:55:13PM +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.
> >
> > Signed-off-by: Pooja Dhannawat 
> > ---
> >  net/net.c | 14 ++
> >  1 file changed, 10 insertions(+), 4 deletions(-)
> >
> > diff --git a/net/net.c b/net/net.c
> > index b0c832e..f03c571 100644
> > --- a/net/net.c
> > +++ b/net/net.c
> > @@ -709,23 +709,29 @@ 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;
> >  uint8_t *buffer;
> >  size_t offset;
> > +ssize_t ret;
> > +
> > +buf = g_new(uint8_t, NET_BUFSIZE);
>
> The linear buffer is only needed when iovcnt > 1.  I suggest the
> following instead:
>
> uint8_t *buf = NULL;
>
> 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, NET_BUFSIZE);
> }
>
> This way the allocation is only made when we actually need to linearize
> the buffer.
>

Thanks Stefan for pointing out this one :)  I will make desired changes and
push patch.

[Qemu-devel] [PATCH v3] net: Allocating Large sized arrays to heap

[Pooja Dhannawat]

nc_sendv_compat has a huge stack usage of 69680 bytes approx.
Moving large arrays to heap to reduce stack usage.

Signed-off-by: Pooja Dhannawat 
---
 net/net.c | 14 ++
 1 file changed, 10 insertions(+), 4 deletions(-)

diff --git a/net/net.c b/net/net.c
index b0c832e..f03c571 100644
--- a/net/net.c
+++ b/net/net.c
@@ -709,23 +709,29 @@ 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;
 uint8_t *buffer;
 size_t offset;
+ssize_t ret;
+
+buf = g_new(uint8_t, NET_BUFSIZE);
 
 if (iovcnt == 1) {
 buffer = iov[0].iov_base;
 offset = iov[0].iov_len;
 } else {
 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