Hi Max, When I committed another patch which named as "hw/net/virtio-net: Allocating Large sized arrays to heap" .
Christian Borntraeger said that 16k is usually perfectly fine for a userspace stack and doing allocations in a hot path might actually hurt performance. Although the size is 65536 bytes here, I think open_eth_start_xmit is in a hot path. So, it is OK, if you think that this patch should not be applied. Sincerely, Zhou Jie On 2016/4/27 10:07, Zhou Jie wrote:
open_eth_start_xmit has a huge stack usage of 65536 bytes approx. Moving large arrays to heap to reduce stack usage. Signed-off-by: Zhou Jie <zhoujie2...@cn.fujitsu.com> --- hw/net/opencores_eth.c | 11 ++++++++++- 1 file changed, 10 insertions(+), 1 deletion(-) diff --git a/hw/net/opencores_eth.c b/hw/net/opencores_eth.c index c6094fb..fa0a4e7 100644 --- a/hw/net/opencores_eth.c +++ b/hw/net/opencores_eth.c @@ -483,7 +483,8 @@ static NetClientInfo net_open_eth_info = { static void open_eth_start_xmit(OpenEthState *s, desc *tx) { - uint8_t buf[65536]; + uint8_t *buf = NULL; + uint8_t buffer[0x600]; unsigned len = GET_FIELD(tx->len_flags, TXD_LEN); unsigned tx_len = len; @@ -498,6 +499,11 @@ static void open_eth_start_xmit(OpenEthState *s, desc *tx) trace_open_eth_start_xmit(tx->buf_ptr, len, tx_len); + if (tx_len > 0x600) { + buf = g_new(uint8_t, tx_len); + } else { + buf = buffer; + } if (len > tx_len) { len = tx_len; } @@ -506,6 +512,9 @@ static void open_eth_start_xmit(OpenEthState *s, desc *tx) memset(buf + len, 0, tx_len - len); } qemu_send_packet(qemu_get_queue(s->nic), buf, tx_len); + if (tx_len > 0x600) { + g_free(buf); + } if (tx->len_flags & TXD_WR) { s->tx_desc = 0;
-- ------------------------------------------------ 周潔 Dept 1 No. 6 Wenzhu Road, Nanjing, 210012, China TEL:+86+25-86630566-8557 FUJITSU INTERNAL:7998-8557 E-Mail:zhoujie2...@cn.fujitsu.com ------------------------------------------------