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
pcmcia_cis.c
/*
* PCMCIA high-level CIS access functions
*
* The initial developer of the original code is David A. Hinds
* <dahinds@users.sourceforge.net>. Portions created by David A. Hinds
* are Copyright (C) 1999 David A. Hinds. All Rights Reserved.
*
* Copyright (C) 1999 David A. Hinds
* Copyright (C) 2004-2010 Dominik Brodowski
*
* 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.
*
*/
#include <linux/slab.h>
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/netdevice.h>
#include <pcmcia/cisreg.h>
#include <pcmcia/cistpl.h>
#include <pcmcia/ss.h>
#include <pcmcia/ds.h>
#include "cs_internal.h"
/**
* pccard_read_tuple() - internal CIS tuple access
* @s: the struct pcmcia_socket where the card is inserted
* @function: the device function we loop for
* @code: which CIS code shall we look for?
* @parse: buffer where the tuple shall be parsed (or NULL, if no parse)
*
* pccard_read_tuple() reads out one tuple and attempts to parse it
*/
int pccard_read_tuple(struct pcmcia_socket *s, unsigned int function,
cisdata_t code, void *parse)
{
tuple_t tuple;
cisdata_t *buf;
int ret;
buf = kmalloc(256, GFP_KERNEL);
if (buf == NULL) {
dev_warn(&s->dev, "no memory to read tuple\n");
return -ENOMEM;
}
tuple.DesiredTuple = code;
tuple.Attributes = 0;
if (function == BIND_FN_ALL)
tuple.Attributes = TUPLE_RETURN_COMMON;
ret = pccard_get_first_tuple(s, function, &tuple);
if (ret != 0)
goto done;
tuple.TupleData = buf;
tuple.TupleOffset = 0;
tuple.TupleDataMax = 255;
ret = pccard_get_tuple_data(s, &tuple);
if (ret != 0)
goto done;
ret = pcmcia_parse_tuple(&tuple, parse);
done:
kfree(buf);
return ret;
}
/**
* pccard_loop_tuple() - loop over tuples in the CIS
* @s: the struct pcmcia_socket where the card is inserted
* @function: the device function we loop for
* @code: which CIS code shall we look for?
* @parse: buffer where the tuple shall be parsed (or NULL, if no parse)
* @priv_data: private data to be passed to the loop_tuple function.
* @loop_tuple: function to call for each CIS entry of type @function. IT
* gets passed the raw tuple, the paresed tuple (if @parse is
* set) and @priv_data.
*
* pccard_loop_tuple() loops over all CIS entries of type @function, and
* calls the @loop_tuple function for each entry. If the call to @loop_tuple
* returns 0, the loop exits. Returns 0 on success or errorcode otherwise.
*/
int pccard_loop_tuple(struct pcmcia_socket *s, unsigned int function,
cisdata_t code, cisparse_t *parse, void *priv_data,
int (*loop_tuple) (tuple_t *tuple,
cisparse_t *parse,
void *priv_data))
{
tuple_t tuple;
cisdata_t *buf;
int ret;
buf = kzalloc(256, GFP_KERNEL);
if (buf == NULL) {
dev_warn(&s->dev, "no memory to read tuple\n");
return -ENOMEM;
}
tuple.TupleData = buf;
tuple.TupleDataMax = 255;
tuple.TupleOffset = 0;
tuple.DesiredTuple = code;
tuple.Attributes = 0;
ret = pccard_get_first_tuple(s, function, &tuple);
while (!ret) {
if (pccard_get_tuple_data(s, &tuple))
goto next_entry;
if (parse)
if (pcmcia_parse_tuple(&tuple, parse))
goto next_entry;
ret = loop_tuple(&tuple, parse, priv_data);
if (!ret)
break;
next_entry:
ret = pccard_get_next_tuple(s, function, &tuple);
}
kfree(buf);
return ret;
}
/**
* pcmcia_io_cfg_data_width() - convert cfgtable to data path width parameter
*/
static int pcmcia_io_cfg_data_width(unsigned int flags)
{
if (!(flags & CISTPL_IO_8BIT))
return IO_DATA_PATH_WIDTH_16;
if (!(flags & CISTPL_IO_16BIT))
return IO_DATA_PATH_WIDTH_8;
return IO_DATA_PATH_WIDTH_AUTO;
}
struct pcmcia_cfg_mem {
struct pcmcia_device *p_dev;
int (*conf_check) (struct pcmcia_device *p_dev, void *priv_data);
void *priv_data;
cisparse_t parse;
cistpl_cftable_entry_t dflt;
};
/**
* pcmcia_do_loop_config() - internal helper for pcmcia_loop_config()
*
* pcmcia_do_loop_config() is the internal callback for the call from
* pcmcia_loop_config() to pccard_loop_tuple(). Data is transferred
* by a struct pcmcia_cfg_mem.
*/
static int pcmcia_do_loop_config(tuple_t *tuple, cisparse_t *parse, void *priv)
{
struct pcmcia_cfg_mem *cfg_mem = priv;
struct pcmcia_device *p_dev = cfg_mem->p_dev;
cistpl_cftable_entry_t *cfg = &parse->cftable_entry;
cistpl_cftable_entry_t *dflt = &cfg_mem->dflt;
unsigned int flags = p_dev->config_flags;
unsigned int vcc = p_dev->socket->socket.Vcc;
dev_dbg(&p_dev->dev, "testing configuration %x, autoconf %x\n",
cfg->index, flags);
/* default values */
cfg_mem->p_dev->config_index = cfg->index;
if (cfg->flags & CISTPL_CFTABLE_DEFAULT)
cfg_mem->dflt = *cfg;
/* check for matching Vcc? */
if (flags & CONF_AUTO_CHECK_VCC) {
if (cfg->vcc.present & (1 << CISTPL_POWER_VNOM)) {
if (vcc != cfg->vcc.param[CISTPL_POWER_VNOM] / 10000)
return -ENODEV;
} else if (dflt->vcc.present & (1 << CISTPL_POWER_VNOM)) {
if (vcc != dflt->vcc.param[CISTPL_POWER_VNOM] / 10000)
return -ENODEV;
}
}
/* set Vpp? */
if (flags & CONF_AUTO_SET_VPP) {
if (cfg->vpp1.present & (1 << CISTPL_POWER_VNOM))
p_dev->vpp = cfg->vpp1.param[CISTPL_POWER_VNOM] / 10000;
else if (dflt->vpp1.present & (1 << CISTPL_POWER_VNOM))
p_dev->vpp =
dflt->vpp1.param[CISTPL_POWER_VNOM] / 10000;
}
/* enable audio? */
if ((flags & CONF_AUTO_AUDIO) && (cfg->flags & CISTPL_CFTABLE_AUDIO))
p_dev->config_flags |= CONF_ENABLE_SPKR;
/* IO window settings? */
if (flags & CONF_AUTO_SET_IO) {
cistpl_io_t *io = (cfg->io.nwin) ? &cfg->io : &dflt->io;
int i = 0;
p_dev->resource[0]->start = p_dev->resource[0]->end = 0;
p_dev->resource[1]->start = p_dev->resource[1]->end = 0;
if (io->nwin == 0)
return -ENODEV;
p_dev->resource[0]->flags &= ~IO_DATA_PATH_WIDTH;
p_dev->resource[0]->flags |=
pcmcia_io_cfg_data_width(io->flags);
if (io->nwin > 1) {
/* For multifunction cards, by convention, we
* configure the network function with window 0,
* and serial with window 1 */
i = (io->win[1].len > io->win[0].len);
p_dev->resource[1]->flags = p_dev->resource[0]->flags;
p_dev->resource[1]->start = io->win[1-i].base;
p_dev->resource[1]->end = io->win[1-i].len;
}
p_dev->resource[0]->start = io->win[i].base;
p_dev->resource[0]->end = io->win[i].len;
p_dev->io_lines = io->flags & CISTPL_IO_LINES_MASK;
}
/* MEM window settings? */
if (flags & CONF_AUTO_SET_IOMEM) {
/* so far, we only set one memory window */
cistpl_mem_t *mem = (cfg->mem.nwin) ? &cfg->mem : &dflt->mem;
p_dev->resource[2]->start = p_dev->resource[2]->end = 0;
if (mem->nwin == 0)
return -ENODEV;
p_dev->resource[2]->start = mem->win[0].host_addr;
p_dev->resource[2]->end = mem->win[0].len;
if (p_dev->resource[2]->end < 0x1000)
p_dev->resource[2]->end = 0x1000;
p_dev->card_addr = mem->win[0].card_addr;
}
dev_dbg(&p_dev->dev,
"checking configuration %x: %pr %pr %pr (%d lines)\n",
p_dev->config_index, p_dev->resource[0], p_dev->resource[1],
p_dev->resource[2], p_dev->io_lines);
return cfg_mem->conf_check(p_dev, cfg_mem->priv_data);
}
/**
* pcmcia_loop_config() - loop over configuration options
* @p_dev: the struct pcmcia_device which we need to loop for.
* @conf_check: function to call for each configuration option.
* It gets passed the struct pcmcia_device and private data
* being passed to pcmcia_loop_config()
* @priv_data: private data to be passed to the conf_check function.
*
* pcmcia_loop_config() loops over all configuration options, and calls
* the driver-specific conf_check() for each one, checking whether
* it is a valid one. Returns 0 on success or errorcode otherwise.
*/
int pcmcia_loop_config(struct pcmcia_device *p_dev,
int (*conf_check) (struct pcmcia_device *p_dev,
void *priv_data),
void *priv_data)
{
struct pcmcia_cfg_mem *cfg_mem;
int ret;
cfg_mem = kzalloc(sizeof(struct pcmcia_cfg_mem), GFP_KERNEL);
if (cfg_mem == NULL)
return -ENOMEM;
cfg_mem->p_dev = p_dev;
cfg_mem->conf_check = conf_check;
cfg_mem->priv_data = priv_data;
ret = pccard_loop_tuple(p_dev->socket, p_dev->func,
CISTPL_CFTABLE_ENTRY, &cfg_mem->parse,
cfg_mem, pcmcia_do_loop_config);
kfree(cfg_mem);
return ret;
}
EXPORT_SYMBOL(pcmcia_loop_config);
struct pcmcia_loop_mem {
struct pcmcia_device *p_dev;
void *priv_data;
int (*loop_tuple) (struct pcmcia_device *p_dev,
tuple_t *tuple,
void *priv_data);
};
/**
* pcmcia_do_loop_tuple() - internal helper for pcmcia_loop_config()
*
* pcmcia_do_loop_tuple() is the internal callback for the call from
* pcmcia_loop_tuple() to pccard_loop_tuple(). Data is transferred
* by a struct pcmcia_cfg_mem.
*/
static int pcmcia_do_loop_tuple(tuple_t *tuple, cisparse_t *parse, void *priv)
{
struct pcmcia_loop_mem *loop = priv;
return loop->loop_tuple(loop->p_dev, tuple, loop->priv_data);
};
/**
* pcmcia_loop_tuple() - loop over tuples in the CIS
* @p_dev: the struct pcmcia_device which we need to loop for.
* @code: which CIS code shall we look for?
* @priv_data: private data to be passed to the loop_tuple function.
* @loop_tuple: function to call for each CIS entry of type @function. IT
* gets passed the raw tuple and @priv_data.
*
* pcmcia_loop_tuple() loops over all CIS entries of type @function, and
* calls the @loop_tuple function for each entry. If the call to @loop_tuple
* returns 0, the loop exits. Returns 0 on success or errorcode otherwise.
*/
int pcmcia_loop_tuple(struct pcmcia_device *p_dev, cisdata_t code,
int (*loop_tuple) (struct pcmcia_device *p_dev,
tuple_t *tuple,
void *priv_data),
void *priv_data)
{
struct pcmcia_loop_mem loop = {
.p_dev = p_dev,
.loop_tuple = loop_tuple,
.priv_data = priv_data};
return pccard_loop_tuple(p_dev->socket, p_dev->func, code, NULL,
&loop, pcmcia_do_loop_tuple);
}
EXPORT_SYMBOL(pcmcia_loop_tuple);
struct pcmcia_loop_get {
size_t len;
cisdata_t **buf;
};
/**
* pcmcia_do_get_tuple() - internal helper for pcmcia_get_tuple()
*
* pcmcia_do_get_tuple() is the internal callback for the call from
* pcmcia_get_tuple() to pcmcia_loop_tuple(). As we're only interested in
* the first tuple, return 0 unconditionally. Create a memory buffer large
* enough to hold the content of the tuple, and fill it with the tuple data.
* The caller is responsible to free the buffer.
*/
static int pcmcia_do_get_tuple(struct pcmcia_device *p_dev, tuple_t *tuple,
void *priv)
{
struct pcmcia_loop_get *get = priv;
*get->buf = kzalloc(tuple->TupleDataLen, GFP_KERNEL);
if (*get->buf) {
get->len = tuple->TupleDataLen;
memcpy(*get->buf, tuple->TupleData, tuple->TupleDataLen);
} else
dev_dbg(&p_dev->dev, "do_get_tuple: out of memory\n");
return 0;
}
/**
* pcmcia_get_tuple() - get first tuple from CIS
* @p_dev: the struct pcmcia_device which we need to loop for.
* @code: which CIS code shall we look for?
* @buf: pointer to store the buffer to.
*
* pcmcia_get_tuple() gets the content of the first CIS entry of type @code.
* It returns the buffer length (or zero). The caller is responsible to free
* the buffer passed in @buf.
*/
size_t pcmcia_get_tuple(struct pcmcia_device *p_dev, cisdata_t code,
unsigned char **buf)
{
struct pcmcia_loop_get get = {
.len = 0,
.buf = buf,
};
*get.buf = NULL;
pcmcia_loop_tuple(p_dev, code, pcmcia_do_get_tuple, &get);
return get.len;
}
EXPORT_SYMBOL(pcmcia_get_tuple);
/**
* pcmcia_do_get_mac() - internal helper for pcmcia_get_mac_from_cis()
*
* pcmcia_do_get_mac() is the internal callback for the call from
* pcmcia_get_mac_from_cis() to pcmcia_loop_tuple(). We check whether the
* tuple contains a proper LAN_NODE_ID of length 6, and copy the data
* to struct net_device->dev_addr[i].
*/
static int pcmcia_do_get_mac(struct pcmcia_device *p_dev, tuple_t *tuple,
void *priv)
{
struct net_device *dev = priv;
int i;
if (tuple->TupleData[0] != CISTPL_FUNCE_LAN_NODE_ID)
return -EINVAL;
if (tuple->TupleDataLen < ETH_ALEN + 2) {
dev_warn(&p_dev->dev, "Invalid CIS tuple length for "
"LAN_NODE_ID\n");
return -EINVAL;
}
if (tuple->TupleData[1] != ETH_ALEN) {
dev_warn(&p_dev->dev, "Invalid header for LAN_NODE_ID\n");
return -EINVAL;
}
for (i = 0; i < 6; i++)
dev->dev_addr[i] = tuple->TupleData[i+2];
return 0;
}
/**
* pcmcia_get_mac_from_cis() - read out MAC address from CISTPL_FUNCE
* @p_dev: the struct pcmcia_device for which we want the address.
* @dev: a properly prepared struct net_device to store the info to.
*
* pcmcia_get_mac_from_cis() reads out the hardware MAC address from
* CISTPL_FUNCE and stores it into struct net_device *dev->dev_addr which
* must be set up properly by the driver (see examples!).
*/
int pcmcia_get_mac_from_cis(struct pcmcia_device *p_dev, struct net_device *dev)
{
return pcmcia_loop_tuple(p_dev, CISTPL_FUNCE, pcmcia_do_get_mac, dev);
}
EXPORT_SYMBOL(pcmcia_get_mac_from_cis);
![swh spinner](/static/img/swh-spinner.gif)
Computing file changes ...