Revision 6e474083f3daf3a3546737f5d7d502ad12eb257c authored by Wei Xu on 01 December 2017, 10:10:36 UTC, committed by David S. Miller on 03 December 2017, 02:31:03 UTC
Matthew found a roughly 40% tcp throughput regression with commit c67df11f(vhost_net: try batch dequing from skb array) as discussed in the following thread: https://www.mail-archive.com/netdev@vger.kernel.org/msg187936.html Eventually we figured out that it was a skb leak in handle_rx() when sending packets to the VM. This usually happens when a guest can not drain out vq as fast as vhost fills in, afterwards it sets off the traffic jam and leaks skb(s) which occurs as no headcount to send on the vq from vhost side. This can be avoided by making sure we have got enough headcount before actually consuming a skb from the batched rx array while transmitting, which is simply done by moving checking the zero headcount a bit ahead. Signed-off-by: Wei Xu <wexu@redhat.com> Reported-by: Matthew Rosato <mjrosato@linux.vnet.ibm.com> Acked-by: Michael S. Tsirkin <mst@redhat.com> Signed-off-by: David S. Miller <davem@davemloft.net>
1 parent fa935ca
sa1111_lubbock.c
/*
* linux/drivers/pcmcia/pxa2xx_lubbock.c
*
* Author: George Davis
* Created: Jan 10, 2002
* Copyright: MontaVista Software Inc.
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License version 2 as
* published by the Free Software Foundation.
*
* Originally based upon linux/drivers/pcmcia/sa1100_neponset.c
*
* Lubbock PCMCIA specific routines.
*
*/
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/device.h>
#include <linux/errno.h>
#include <linux/init.h>
#include <linux/delay.h>
#include <mach/hardware.h>
#include <asm/hardware/sa1111.h>
#include <asm/mach-types.h>
#include <mach/lubbock.h>
#include "sa1111_generic.h"
static int
lubbock_pcmcia_configure_socket(struct soc_pcmcia_socket *skt,
const socket_state_t *state)
{
struct sa1111_pcmcia_socket *s = to_skt(skt);
unsigned int pa_dwr_mask, pa_dwr_set, misc_mask, misc_set;
int ret = 0;
pa_dwr_mask = pa_dwr_set = misc_mask = misc_set = 0;
/* Lubbock uses the Maxim MAX1602, with the following connections:
*
* Socket 0 (PCMCIA):
* MAX1602 Lubbock Register
* Pin Signal
* ----- ------- ----------------------
* A0VPP S0_PWR0 SA-1111 GPIO A<0>
* A1VPP S0_PWR1 SA-1111 GPIO A<1>
* A0VCC S0_PWR2 SA-1111 GPIO A<2>
* A1VCC S0_PWR3 SA-1111 GPIO A<3>
* VX VCC
* VY +3.3V
* 12IN +12V
* CODE +3.3V Cirrus Code, CODE = High (VY)
*
* Socket 1 (CF):
* MAX1602 Lubbock Register
* Pin Signal
* ----- ------- ----------------------
* A0VPP GND VPP is not connected
* A1VPP GND VPP is not connected
* A0VCC S1_PWR0 MISC_WR<14>
* A1VCC S1_PWR1 MISC_WR<15>
* VX VCC
* VY +3.3V
* 12IN GND VPP is not connected
* CODE +3.3V Cirrus Code, CODE = High (VY)
*
*/
again:
switch (skt->nr) {
case 0:
pa_dwr_mask = GPIO_A0 | GPIO_A1 | GPIO_A2 | GPIO_A3;
switch (state->Vcc) {
case 0: /* Hi-Z */
break;
case 33: /* VY */
pa_dwr_set |= GPIO_A3;
break;
case 50: /* VX */
pa_dwr_set |= GPIO_A2;
break;
default:
printk(KERN_ERR "%s(): unrecognized Vcc %u\n",
__func__, state->Vcc);
ret = -1;
}
switch (state->Vpp) {
case 0: /* Hi-Z */
break;
case 120: /* 12IN */
pa_dwr_set |= GPIO_A1;
break;
default: /* VCC */
if (state->Vpp == state->Vcc)
pa_dwr_set |= GPIO_A0;
else {
printk(KERN_ERR "%s(): unrecognized Vpp %u\n",
__func__, state->Vpp);
ret = -1;
break;
}
}
break;
case 1:
misc_mask = (1 << 15) | (1 << 14);
switch (state->Vcc) {
case 0: /* Hi-Z */
break;
case 33: /* VY */
misc_set |= 1 << 15;
break;
case 50: /* VX */
misc_set |= 1 << 14;
break;
default:
printk(KERN_ERR "%s(): unrecognized Vcc %u\n",
__func__, state->Vcc);
ret = -1;
break;
}
if (state->Vpp != state->Vcc && state->Vpp != 0) {
printk(KERN_ERR "%s(): CF slot cannot support Vpp %u\n",
__func__, state->Vpp);
ret = -1;
break;
}
break;
default:
ret = -1;
}
if (ret == 0)
ret = sa1111_pcmcia_configure_socket(skt, state);
if (ret == 0) {
lubbock_set_misc_wr(misc_mask, misc_set);
sa1111_set_io(s->dev, pa_dwr_mask, pa_dwr_set);
}
#if 1
if (ret == 0 && state->Vcc == 33) {
struct pcmcia_state new_state;
/*
* HACK ALERT:
* We can't sense the voltage properly on Lubbock before
* actually applying some power to the socket (catch 22).
* Resense the socket Voltage Sense pins after applying
* socket power.
*
* Note: It takes about 2.5ms for the MAX1602 VCC output
* to rise.
*/
mdelay(3);
sa1111_pcmcia_socket_state(skt, &new_state);
if (!new_state.vs_3v && !new_state.vs_Xv) {
/*
* Switch to 5V, Configure socket with 5V voltage
*/
lubbock_set_misc_wr(misc_mask, 0);
sa1111_set_io(s->dev, pa_dwr_mask, 0);
/*
* It takes about 100ms to turn off Vcc.
*/
mdelay(100);
/*
* We need to hack around the const qualifier as
* well to keep this ugly workaround localized and
* not force it to the rest of the code. Barf bags
* available in the seat pocket in front of you!
*/
((socket_state_t *)state)->Vcc = 50;
((socket_state_t *)state)->Vpp = 50;
goto again;
}
}
#endif
return ret;
}
static struct pcmcia_low_level lubbock_pcmcia_ops = {
.owner = THIS_MODULE,
.configure_socket = lubbock_pcmcia_configure_socket,
.first = 0,
.nr = 2,
};
#include "pxa2xx_base.h"
int pcmcia_lubbock_init(struct sa1111_dev *sadev)
{
/*
* Set GPIO_A<3:0> to be outputs for the MAX1600,
* and switch to standby mode.
*/
sa1111_set_io_dir(sadev, GPIO_A0|GPIO_A1|GPIO_A2|GPIO_A3, 0, 0);
sa1111_set_io(sadev, GPIO_A0|GPIO_A1|GPIO_A2|GPIO_A3, 0);
sa1111_set_sleep_io(sadev, GPIO_A0|GPIO_A1|GPIO_A2|GPIO_A3, 0);
/* Set CF Socket 1 power to standby mode. */
lubbock_set_misc_wr((1 << 15) | (1 << 14), 0);
pxa2xx_drv_pcmcia_ops(&lubbock_pcmcia_ops);
pxa2xx_configure_sockets(&sadev->dev, &lubbock_pcmcia_ops);
return sa1111_pcmcia_add(sadev, &lubbock_pcmcia_ops,
pxa2xx_drv_pcmcia_add_one);
}
MODULE_LICENSE("GPL");
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...