OpenCloudOS-Kernel/drivers/xen/xenbus/xenbus_comms.c

493 lines
11 KiB
C
Raw Normal View History

/******************************************************************************
* xenbus_comms.c
*
* Low level code to talks to Xen Store: ringbuffer and event channel.
*
* Copyright (C) 2005 Rusty Russell, IBM Corporation
*
* 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; or, when distributed
* separately from the Linux kernel or incorporated into other
* software packages, subject to the following license:
*
* Permission is hereby granted, free of charge, to any person obtaining a copy
* of this source file (the "Software"), to deal in the Software without
* restriction, including without limitation the rights to use, copy, modify,
* merge, publish, distribute, sublicense, and/or sell copies of the Software,
* and to permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice shall be included in
* all copies or substantial portions of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
* AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
* IN THE SOFTWARE.
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/wait.h>
#include <linux/interrupt.h>
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
#include <linux/kthread.h>
#include <linux/sched.h>
#include <linux/err.h>
#include <xen/xenbus.h>
#include <asm/xen/hypervisor.h>
#include <xen/events.h>
#include <xen/page.h>
#include "xenbus.h"
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
/* A list of replies. Currently only one will ever be outstanding. */
LIST_HEAD(xs_reply_list);
/* A list of write requests. */
LIST_HEAD(xb_write_list);
DECLARE_WAIT_QUEUE_HEAD(xb_waitq);
DEFINE_MUTEX(xb_write_mutex);
/* Protect xenbus reader thread against save/restore. */
DEFINE_MUTEX(xs_response_mutex);
static int xenbus_irq;
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
static struct task_struct *xenbus_task;
static DECLARE_WORK(probe_work, xenbus_probe);
static irqreturn_t wake_waiting(int irq, void *unused)
{
if (unlikely(xenstored_ready == 0)) {
xenstored_ready = 1;
schedule_work(&probe_work);
}
wake_up(&xb_waitq);
return IRQ_HANDLED;
}
static int check_indexes(XENSTORE_RING_IDX cons, XENSTORE_RING_IDX prod)
{
return ((prod - cons) <= XENSTORE_RING_SIZE);
}
static void *get_output_chunk(XENSTORE_RING_IDX cons,
XENSTORE_RING_IDX prod,
char *buf, uint32_t *len)
{
*len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(prod);
if ((XENSTORE_RING_SIZE - (prod - cons)) < *len)
*len = XENSTORE_RING_SIZE - (prod - cons);
return buf + MASK_XENSTORE_IDX(prod);
}
static const void *get_input_chunk(XENSTORE_RING_IDX cons,
XENSTORE_RING_IDX prod,
const char *buf, uint32_t *len)
{
*len = XENSTORE_RING_SIZE - MASK_XENSTORE_IDX(cons);
if ((prod - cons) < *len)
*len = prod - cons;
return buf + MASK_XENSTORE_IDX(cons);
}
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
static int xb_data_to_write(void)
{
struct xenstore_domain_interface *intf = xen_store_interface;
return (intf->req_prod - intf->req_cons) != XENSTORE_RING_SIZE &&
!list_empty(&xb_write_list);
}
/**
* xb_write - low level write
* @data: buffer to send
* @len: length of buffer
*
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
* Returns number of bytes written or -err.
*/
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
static int xb_write(const void *data, unsigned int len)
{
struct xenstore_domain_interface *intf = xen_store_interface;
XENSTORE_RING_IDX cons, prod;
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
unsigned int bytes = 0;
while (len != 0) {
void *dst;
unsigned int avail;
/* Read indexes, then verify. */
cons = intf->req_cons;
prod = intf->req_prod;
if (!check_indexes(cons, prod)) {
intf->req_cons = intf->req_prod = 0;
return -EIO;
}
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
if (!xb_data_to_write())
return bytes;
/* Must write data /after/ reading the consumer index. */
virt_mb();
dst = get_output_chunk(cons, prod, intf->req, &avail);
if (avail == 0)
continue;
if (avail > len)
avail = len;
memcpy(dst, data, avail);
data += avail;
len -= avail;
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
bytes += avail;
/* Other side must not see new producer until data is there. */
virt_wmb();
intf->req_prod += avail;
/* Implies mb(): other side will see the updated producer. */
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
if (prod <= intf->req_cons)
notify_remote_via_evtchn(xen_store_evtchn);
}
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
return bytes;
}
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
static int xb_data_to_read(void)
{
struct xenstore_domain_interface *intf = xen_store_interface;
return (intf->rsp_cons != intf->rsp_prod);
}
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
static int xb_read(void *data, unsigned int len)
{
struct xenstore_domain_interface *intf = xen_store_interface;
XENSTORE_RING_IDX cons, prod;
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
unsigned int bytes = 0;
while (len != 0) {
unsigned int avail;
const char *src;
/* Read indexes, then verify. */
cons = intf->rsp_cons;
prod = intf->rsp_prod;
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
if (cons == prod)
return bytes;
if (!check_indexes(cons, prod)) {
intf->rsp_cons = intf->rsp_prod = 0;
return -EIO;
}
src = get_input_chunk(cons, prod, intf->rsp, &avail);
if (avail == 0)
continue;
if (avail > len)
avail = len;
/* Must read data /after/ reading the producer index. */
virt_rmb();
memcpy(data, src, avail);
data += avail;
len -= avail;
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
bytes += avail;
/* Other side must not see free space until we've copied out */
virt_mb();
intf->rsp_cons += avail;
/* Implies mb(): other side will see the updated consumer. */
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
if (intf->rsp_prod - cons >= XENSTORE_RING_SIZE)
notify_remote_via_evtchn(xen_store_evtchn);
}
return bytes;
}
static int process_msg(void)
{
static struct {
struct xsd_sockmsg msg;
char *body;
union {
void *alloc;
struct xs_watch_event *watch;
};
bool in_msg;
bool in_hdr;
unsigned int read;
} state;
struct xb_req_data *req;
int err;
unsigned int len;
if (!state.in_msg) {
state.in_msg = true;
state.in_hdr = true;
state.read = 0;
/*
* We must disallow save/restore while reading a message.
* A partial read across s/r leaves us out of sync with
* xenstored.
* xs_response_mutex is locked as long as we are processing one
* message. state.in_msg will be true as long as we are holding
* the lock here.
*/
mutex_lock(&xs_response_mutex);
if (!xb_data_to_read()) {
/* We raced with save/restore: pending data 'gone'. */
mutex_unlock(&xs_response_mutex);
state.in_msg = false;
return 0;
}
}
if (state.in_hdr) {
if (state.read != sizeof(state.msg)) {
err = xb_read((void *)&state.msg + state.read,
sizeof(state.msg) - state.read);
if (err < 0)
goto out;
state.read += err;
if (state.read != sizeof(state.msg))
return 0;
if (state.msg.len > XENSTORE_PAYLOAD_MAX) {
err = -EINVAL;
goto out;
}
}
len = state.msg.len + 1;
if (state.msg.type == XS_WATCH_EVENT)
len += sizeof(*state.watch);
state.alloc = kmalloc(len, GFP_NOIO | __GFP_HIGH);
if (!state.alloc)
return -ENOMEM;
if (state.msg.type == XS_WATCH_EVENT)
state.body = state.watch->body;
else
state.body = state.alloc;
state.in_hdr = false;
state.read = 0;
}
err = xb_read(state.body + state.read, state.msg.len - state.read);
if (err < 0)
goto out;
state.read += err;
if (state.read != state.msg.len)
return 0;
state.body[state.msg.len] = '\0';
if (state.msg.type == XS_WATCH_EVENT) {
state.watch->len = state.msg.len;
err = xs_watch_msg(state.watch);
} else {
err = -ENOENT;
mutex_lock(&xb_write_mutex);
list_for_each_entry(req, &xs_reply_list, list) {
if (req->msg.req_id == state.msg.req_id) {
xen: avoid deadlock in xenbus driver There has been a report about a deadlock in the xenbus driver: [ 247.979498] ====================================================== [ 247.985688] WARNING: possible circular locking dependency detected [ 247.991882] 4.12.0-rc4-00022-gc4b25c0 #575 Not tainted [ 247.997040] ------------------------------------------------------ [ 248.003232] xenbus/91 is trying to acquire lock: [ 248.007875] (&u->msgbuffer_mutex){+.+.+.}, at: [<ffff00000863e904>] xenbus_dev_queue_reply+0x3c/0x230 [ 248.017163] [ 248.017163] but task is already holding lock: [ 248.023096] (xb_write_mutex){+.+...}, at: [<ffff00000863a940>] xenbus_thread+0x5f0/0x798 [ 248.031267] [ 248.031267] which lock already depends on the new lock. [ 248.031267] [ 248.039615] [ 248.039615] the existing dependency chain (in reverse order) is: [ 248.047176] [ 248.047176] -> #1 (xb_write_mutex){+.+...}: [ 248.052943] __lock_acquire+0x1728/0x1778 [ 248.057498] lock_acquire+0xc4/0x288 [ 248.061630] __mutex_lock+0x84/0x868 [ 248.065755] mutex_lock_nested+0x3c/0x50 [ 248.070227] xs_send+0x164/0x1f8 [ 248.074015] xenbus_dev_request_and_reply+0x6c/0x88 [ 248.079427] xenbus_file_write+0x260/0x420 [ 248.084073] __vfs_write+0x48/0x138 [ 248.088113] vfs_write+0xa8/0x1b8 [ 248.091983] SyS_write+0x54/0xb0 [ 248.095768] el0_svc_naked+0x24/0x28 [ 248.099897] [ 248.099897] -> #0 (&u->msgbuffer_mutex){+.+.+.}: [ 248.106088] print_circular_bug+0x80/0x2e0 [ 248.110730] __lock_acquire+0x1768/0x1778 [ 248.115288] lock_acquire+0xc4/0x288 [ 248.119417] __mutex_lock+0x84/0x868 [ 248.123545] mutex_lock_nested+0x3c/0x50 [ 248.128016] xenbus_dev_queue_reply+0x3c/0x230 [ 248.133005] xenbus_thread+0x788/0x798 [ 248.137306] kthread+0x110/0x140 [ 248.141087] ret_from_fork+0x10/0x40 It is rather easy to avoid by dropping xb_write_mutex before calling xenbus_dev_queue_reply(). Fixes: fd8aa9095a95c02dcc35540a263267c29b8fda9d ("xen: optimize xenbus driver for multiple concurrent xenstore accesses"). Cc: <stable@vger.kernel.org> # 4.11 Reported-by: Andre Przywara <andre.przywara@arm.com> Signed-off-by: Juergen Gross <jgross@suse.com> Tested-by: Andre Przywara <andre.przywara@arm.com> Signed-off-by: Juergen Gross <jgross@suse.com>
2017-06-08 22:03:42 +08:00
list_del(&req->list);
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
err = 0;
break;
}
}
mutex_unlock(&xb_write_mutex);
if (err)
goto out;
xen: avoid deadlock in xenbus driver There has been a report about a deadlock in the xenbus driver: [ 247.979498] ====================================================== [ 247.985688] WARNING: possible circular locking dependency detected [ 247.991882] 4.12.0-rc4-00022-gc4b25c0 #575 Not tainted [ 247.997040] ------------------------------------------------------ [ 248.003232] xenbus/91 is trying to acquire lock: [ 248.007875] (&u->msgbuffer_mutex){+.+.+.}, at: [<ffff00000863e904>] xenbus_dev_queue_reply+0x3c/0x230 [ 248.017163] [ 248.017163] but task is already holding lock: [ 248.023096] (xb_write_mutex){+.+...}, at: [<ffff00000863a940>] xenbus_thread+0x5f0/0x798 [ 248.031267] [ 248.031267] which lock already depends on the new lock. [ 248.031267] [ 248.039615] [ 248.039615] the existing dependency chain (in reverse order) is: [ 248.047176] [ 248.047176] -> #1 (xb_write_mutex){+.+...}: [ 248.052943] __lock_acquire+0x1728/0x1778 [ 248.057498] lock_acquire+0xc4/0x288 [ 248.061630] __mutex_lock+0x84/0x868 [ 248.065755] mutex_lock_nested+0x3c/0x50 [ 248.070227] xs_send+0x164/0x1f8 [ 248.074015] xenbus_dev_request_and_reply+0x6c/0x88 [ 248.079427] xenbus_file_write+0x260/0x420 [ 248.084073] __vfs_write+0x48/0x138 [ 248.088113] vfs_write+0xa8/0x1b8 [ 248.091983] SyS_write+0x54/0xb0 [ 248.095768] el0_svc_naked+0x24/0x28 [ 248.099897] [ 248.099897] -> #0 (&u->msgbuffer_mutex){+.+.+.}: [ 248.106088] print_circular_bug+0x80/0x2e0 [ 248.110730] __lock_acquire+0x1768/0x1778 [ 248.115288] lock_acquire+0xc4/0x288 [ 248.119417] __mutex_lock+0x84/0x868 [ 248.123545] mutex_lock_nested+0x3c/0x50 [ 248.128016] xenbus_dev_queue_reply+0x3c/0x230 [ 248.133005] xenbus_thread+0x788/0x798 [ 248.137306] kthread+0x110/0x140 [ 248.141087] ret_from_fork+0x10/0x40 It is rather easy to avoid by dropping xb_write_mutex before calling xenbus_dev_queue_reply(). Fixes: fd8aa9095a95c02dcc35540a263267c29b8fda9d ("xen: optimize xenbus driver for multiple concurrent xenstore accesses"). Cc: <stable@vger.kernel.org> # 4.11 Reported-by: Andre Przywara <andre.przywara@arm.com> Signed-off-by: Juergen Gross <jgross@suse.com> Tested-by: Andre Przywara <andre.przywara@arm.com> Signed-off-by: Juergen Gross <jgross@suse.com>
2017-06-08 22:03:42 +08:00
if (req->state == xb_req_state_wait_reply) {
xenbus: track caller request id Commit fd8aa9095a95 ("xen: optimize xenbus driver for multiple concurrent xenstore accesses") optimized xenbus concurrent accesses but in doing so broke UABI of /dev/xen/xenbus. Through /dev/xen/xenbus applications are in charge of xenbus message exchange with the correct header and body. Now, after the mentioned commit the replies received by application will no longer have the header req_id echoed back as it was on request (see specification below for reference), because that particular field is being overwritten by kernel. struct xsd_sockmsg { uint32_t type; /* XS_??? */ uint32_t req_id;/* Request identifier, echoed in daemon's response. */ uint32_t tx_id; /* Transaction id (0 if not related to a transaction). */ uint32_t len; /* Length of data following this. */ /* Generally followed by nul-terminated string(s). */ }; Before there was only one request at a time so req_id could simply be forwarded back and forth. To allow simultaneous requests we need a different req_id for each message thus kernel keeps a monotonic increasing counter for this field and is written on every request irrespective of userspace value. Forwarding again the req_id on userspace requests is not a solution because we would open the possibility of userspace-generated req_id colliding with kernel ones. So this patch instead takes another route which is to artificially keep user req_id while keeping the xenbus logic as is. We do that by saving the original req_id before xs_send(), use the private kernel counter as req_id and then once reply comes and was validated, we restore back the original req_id. Cc: <stable@vger.kernel.org> # 4.11 Fixes: fd8aa9095a ("xen: optimize xenbus driver for multiple concurrent xenstore accesses") Reported-by: Bhavesh Davda <bhavesh.davda@oracle.com> Signed-off-by: Joao Martins <joao.m.martins@oracle.com> Reviewed-by: Juergen Gross <jgross@suse.com> Signed-off-by: Juergen Gross <jgross@suse.com>
2018-02-03 01:42:33 +08:00
req->msg.req_id = req->caller_req_id;
xen: avoid deadlock in xenbus driver There has been a report about a deadlock in the xenbus driver: [ 247.979498] ====================================================== [ 247.985688] WARNING: possible circular locking dependency detected [ 247.991882] 4.12.0-rc4-00022-gc4b25c0 #575 Not tainted [ 247.997040] ------------------------------------------------------ [ 248.003232] xenbus/91 is trying to acquire lock: [ 248.007875] (&u->msgbuffer_mutex){+.+.+.}, at: [<ffff00000863e904>] xenbus_dev_queue_reply+0x3c/0x230 [ 248.017163] [ 248.017163] but task is already holding lock: [ 248.023096] (xb_write_mutex){+.+...}, at: [<ffff00000863a940>] xenbus_thread+0x5f0/0x798 [ 248.031267] [ 248.031267] which lock already depends on the new lock. [ 248.031267] [ 248.039615] [ 248.039615] the existing dependency chain (in reverse order) is: [ 248.047176] [ 248.047176] -> #1 (xb_write_mutex){+.+...}: [ 248.052943] __lock_acquire+0x1728/0x1778 [ 248.057498] lock_acquire+0xc4/0x288 [ 248.061630] __mutex_lock+0x84/0x868 [ 248.065755] mutex_lock_nested+0x3c/0x50 [ 248.070227] xs_send+0x164/0x1f8 [ 248.074015] xenbus_dev_request_and_reply+0x6c/0x88 [ 248.079427] xenbus_file_write+0x260/0x420 [ 248.084073] __vfs_write+0x48/0x138 [ 248.088113] vfs_write+0xa8/0x1b8 [ 248.091983] SyS_write+0x54/0xb0 [ 248.095768] el0_svc_naked+0x24/0x28 [ 248.099897] [ 248.099897] -> #0 (&u->msgbuffer_mutex){+.+.+.}: [ 248.106088] print_circular_bug+0x80/0x2e0 [ 248.110730] __lock_acquire+0x1768/0x1778 [ 248.115288] lock_acquire+0xc4/0x288 [ 248.119417] __mutex_lock+0x84/0x868 [ 248.123545] mutex_lock_nested+0x3c/0x50 [ 248.128016] xenbus_dev_queue_reply+0x3c/0x230 [ 248.133005] xenbus_thread+0x788/0x798 [ 248.137306] kthread+0x110/0x140 [ 248.141087] ret_from_fork+0x10/0x40 It is rather easy to avoid by dropping xb_write_mutex before calling xenbus_dev_queue_reply(). Fixes: fd8aa9095a95c02dcc35540a263267c29b8fda9d ("xen: optimize xenbus driver for multiple concurrent xenstore accesses"). Cc: <stable@vger.kernel.org> # 4.11 Reported-by: Andre Przywara <andre.przywara@arm.com> Signed-off-by: Juergen Gross <jgross@suse.com> Tested-by: Andre Przywara <andre.przywara@arm.com> Signed-off-by: Juergen Gross <jgross@suse.com>
2017-06-08 22:03:42 +08:00
req->msg.type = state.msg.type;
req->msg.len = state.msg.len;
req->body = state.body;
xenbus: req->body should be updated before req->state The req->body should be updated before req->state is updated and the order should be guaranteed by a barrier. Otherwise, read_reply() might return req->body = NULL. Below is sample callstack when the issue is reproduced on purpose by reordering the updates of req->body and req->state and adding delay in code between updates of req->state and req->body. [ 22.356105] general protection fault: 0000 [#1] SMP PTI [ 22.361185] CPU: 2 PID: 52 Comm: xenwatch Not tainted 5.5.0xen+ #6 [ 22.366727] Hardware name: Xen HVM domU, BIOS ... [ 22.372245] RIP: 0010:_parse_integer_fixup_radix+0x6/0x60 ... ... [ 22.392163] RSP: 0018:ffffb2d64023fdf0 EFLAGS: 00010246 [ 22.395933] RAX: 0000000000000000 RBX: 75746e7562755f6d RCX: 0000000000000000 [ 22.400871] RDX: 0000000000000000 RSI: ffffb2d64023fdfc RDI: 75746e7562755f6d [ 22.405874] RBP: 0000000000000000 R08: 00000000000001e8 R09: 0000000000cdcdcd [ 22.410945] R10: ffffb2d6402ffe00 R11: ffff9d95395eaeb0 R12: ffff9d9535935000 [ 22.417613] R13: ffff9d9526d4a000 R14: ffff9d9526f4f340 R15: ffff9d9537654000 [ 22.423726] FS: 0000000000000000(0000) GS:ffff9d953bc80000(0000) knlGS:0000000000000000 [ 22.429898] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033 [ 22.434342] CR2: 000000c4206a9000 CR3: 00000001ea3fc002 CR4: 00000000001606e0 [ 22.439645] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000 [ 22.444941] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400 [ 22.450342] Call Trace: [ 22.452509] simple_strtoull+0x27/0x70 [ 22.455572] xenbus_transaction_start+0x31/0x50 [ 22.459104] netback_changed+0x76c/0xcc1 [xen_netfront] [ 22.463279] ? find_watch+0x40/0x40 [ 22.466156] xenwatch_thread+0xb4/0x150 [ 22.469309] ? wait_woken+0x80/0x80 [ 22.472198] kthread+0x10e/0x130 [ 22.474925] ? kthread_park+0x80/0x80 [ 22.477946] ret_from_fork+0x35/0x40 [ 22.480968] Modules linked in: xen_kbdfront xen_fbfront(+) xen_netfront xen_blkfront [ 22.486783] ---[ end trace a9222030a747c3f7 ]--- [ 22.490424] RIP: 0010:_parse_integer_fixup_radix+0x6/0x60 The virt_rmb() is added in the 'true' path of test_reply(). The "while" is changed to "do while" so that test_reply() is used as a read memory barrier. Signed-off-by: Dongli Zhang <dongli.zhang@oracle.com> Link: https://lore.kernel.org/r/20200303221423.21962-1-dongli.zhang@oracle.com Reviewed-by: Julien Grall <jgrall@amazon.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2020-03-04 06:14:22 +08:00
/* write body, then update state */
virt_wmb();
xen: avoid deadlock in xenbus driver There has been a report about a deadlock in the xenbus driver: [ 247.979498] ====================================================== [ 247.985688] WARNING: possible circular locking dependency detected [ 247.991882] 4.12.0-rc4-00022-gc4b25c0 #575 Not tainted [ 247.997040] ------------------------------------------------------ [ 248.003232] xenbus/91 is trying to acquire lock: [ 248.007875] (&u->msgbuffer_mutex){+.+.+.}, at: [<ffff00000863e904>] xenbus_dev_queue_reply+0x3c/0x230 [ 248.017163] [ 248.017163] but task is already holding lock: [ 248.023096] (xb_write_mutex){+.+...}, at: [<ffff00000863a940>] xenbus_thread+0x5f0/0x798 [ 248.031267] [ 248.031267] which lock already depends on the new lock. [ 248.031267] [ 248.039615] [ 248.039615] the existing dependency chain (in reverse order) is: [ 248.047176] [ 248.047176] -> #1 (xb_write_mutex){+.+...}: [ 248.052943] __lock_acquire+0x1728/0x1778 [ 248.057498] lock_acquire+0xc4/0x288 [ 248.061630] __mutex_lock+0x84/0x868 [ 248.065755] mutex_lock_nested+0x3c/0x50 [ 248.070227] xs_send+0x164/0x1f8 [ 248.074015] xenbus_dev_request_and_reply+0x6c/0x88 [ 248.079427] xenbus_file_write+0x260/0x420 [ 248.084073] __vfs_write+0x48/0x138 [ 248.088113] vfs_write+0xa8/0x1b8 [ 248.091983] SyS_write+0x54/0xb0 [ 248.095768] el0_svc_naked+0x24/0x28 [ 248.099897] [ 248.099897] -> #0 (&u->msgbuffer_mutex){+.+.+.}: [ 248.106088] print_circular_bug+0x80/0x2e0 [ 248.110730] __lock_acquire+0x1768/0x1778 [ 248.115288] lock_acquire+0xc4/0x288 [ 248.119417] __mutex_lock+0x84/0x868 [ 248.123545] mutex_lock_nested+0x3c/0x50 [ 248.128016] xenbus_dev_queue_reply+0x3c/0x230 [ 248.133005] xenbus_thread+0x788/0x798 [ 248.137306] kthread+0x110/0x140 [ 248.141087] ret_from_fork+0x10/0x40 It is rather easy to avoid by dropping xb_write_mutex before calling xenbus_dev_queue_reply(). Fixes: fd8aa9095a95c02dcc35540a263267c29b8fda9d ("xen: optimize xenbus driver for multiple concurrent xenstore accesses"). Cc: <stable@vger.kernel.org> # 4.11 Reported-by: Andre Przywara <andre.przywara@arm.com> Signed-off-by: Juergen Gross <jgross@suse.com> Tested-by: Andre Przywara <andre.przywara@arm.com> Signed-off-by: Juergen Gross <jgross@suse.com>
2017-06-08 22:03:42 +08:00
req->state = xb_req_state_got_reply;
req->cb(req);
} else
kfree(req);
}
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
mutex_unlock(&xs_response_mutex);
state.in_msg = false;
state.alloc = NULL;
return err;
out:
mutex_unlock(&xs_response_mutex);
state.in_msg = false;
kfree(state.alloc);
state.alloc = NULL;
return err;
}
static int process_writes(void)
{
static struct {
struct xb_req_data *req;
int idx;
unsigned int written;
} state;
void *base;
unsigned int len;
int err = 0;
if (!xb_data_to_write())
return 0;
mutex_lock(&xb_write_mutex);
if (!state.req) {
state.req = list_first_entry(&xb_write_list,
struct xb_req_data, list);
state.idx = -1;
state.written = 0;
}
if (state.req->state == xb_req_state_aborted)
goto out_err;
while (state.idx < state.req->num_vecs) {
if (state.idx < 0) {
base = &state.req->msg;
len = sizeof(state.req->msg);
} else {
base = state.req->vec[state.idx].iov_base;
len = state.req->vec[state.idx].iov_len;
}
err = xb_write(base + state.written, len - state.written);
if (err < 0)
goto out_err;
state.written += err;
if (state.written != len)
goto out;
state.idx++;
state.written = 0;
}
list_del(&state.req->list);
state.req->state = xb_req_state_wait_reply;
list_add_tail(&state.req->list, &xs_reply_list);
state.req = NULL;
out:
mutex_unlock(&xb_write_mutex);
return 0;
out_err:
state.req->msg.type = XS_ERROR;
state.req->err = err;
list_del(&state.req->list);
if (state.req->state == xb_req_state_aborted)
kfree(state.req);
else {
/* write err, then update state */
virt_wmb();
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
state.req->state = xb_req_state_got_reply;
wake_up(&state.req->wq);
}
mutex_unlock(&xb_write_mutex);
state.req = NULL;
return err;
}
static int xb_thread_work(void)
{
return xb_data_to_read() || xb_data_to_write();
}
static int xenbus_thread(void *unused)
{
int err;
while (!kthread_should_stop()) {
if (wait_event_interruptible(xb_waitq, xb_thread_work()))
continue;
err = process_msg();
if (err == -ENOMEM)
schedule();
else if (err)
pr_warn_ratelimited("error %d while reading message\n",
err);
err = process_writes();
if (err)
pr_warn_ratelimited("error %d while writing message\n",
err);
}
xenbus_task = NULL;
return 0;
}
/**
* xb_init_comms - Set up interrupt handler off store event channel.
*/
int xb_init_comms(void)
{
struct xenstore_domain_interface *intf = xen_store_interface;
if (intf->req_prod != intf->req_cons)
pr_err("request ring is not quiescent (%08x:%08x)!\n",
intf->req_cons, intf->req_prod);
if (intf->rsp_prod != intf->rsp_cons) {
pr_warn("response ring is not quiescent (%08x:%08x): fixing up\n",
intf->rsp_cons, intf->rsp_prod);
/* breaks kdump */
if (!reset_devices)
intf->rsp_cons = intf->rsp_prod;
}
if (xenbus_irq) {
/* Already have an irq; assume we're resuming */
rebind_evtchn_irq(xen_store_evtchn, xenbus_irq);
} else {
int err;
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
err = bind_evtchn_to_irqhandler(xen_store_evtchn, wake_waiting,
0, "xenbus", &xb_waitq);
if (err < 0) {
pr_err("request irq failed %i\n", err);
return err;
}
xenbus_irq = err;
xen: optimize xenbus driver for multiple concurrent xenstore accesses Handling of multiple concurrent Xenstore accesses through xenbus driver either from the kernel or user land is rather lame today: xenbus is capable to have one access active only at one point of time. Rewrite xenbus to handle multiple requests concurrently by making use of the request id of the Xenstore protocol. This requires to: - Instead of blocking inside xb_read() when trying to read data from the xenstore ring buffer do so only in the main loop of xenbus_thread(). - Instead of doing writes to the xenstore ring buffer in the context of the caller just queue the request and do the write in the dedicated xenbus thread. - Instead of just forwarding the request id specified by the caller of xenbus to xenstore use a xenbus internal unique request id. This will allow multiple outstanding requests. - Modify the locking scheme in order to allow multiple requests being active in parallel. - Instead of waiting for the reply of a user's xenstore request after writing the request to the xenstore ring buffer return directly to the caller and do the waiting in the read path. Additionally signal handling was optimized by avoiding waking up the xenbus thread or sending an event to Xenstore in case the addressed entity is known to be running already. As a result communication with Xenstore is sped up by a factor of up to 5: depending on the request type (read or write) and the amount of data transferred the gain was at least 20% (small reads) and went up to a factor of 5 for large writes. In the end some more rough edges of xenbus have been smoothed: - Handling of memory shortage when reading from xenstore ring buffer in the xenbus driver was not optimal: it was busy looping and issuing a warning in each loop. - In case of xenstore not running in dom0 but in a stubdom we end up with two xenbus threads running as the initialization of xenbus in dom0 expecting a local xenstored will be redone later when connecting to the xenstore domain. Up to now this was no problem as locking would prevent the two xenbus threads interfering with each other, but this was just a waste of kernel resources. - An out of memory situation while writing to or reading from the xenstore ring buffer no longer will lead to a possible loss of synchronization with xenstore. - The user read and write part are now interruptible by signals. Signed-off-by: Juergen Gross <jgross@suse.com> Signed-off-by: Boris Ostrovsky <boris.ostrovsky@oracle.com>
2017-02-09 21:39:58 +08:00
if (!xenbus_task) {
xenbus_task = kthread_run(xenbus_thread, NULL,
"xenbus");
if (IS_ERR(xenbus_task))
return PTR_ERR(xenbus_task);
}
}
return 0;
}
void xb_deinit_comms(void)
{
unbind_from_irqhandler(xenbus_irq, &xb_waitq);
xenbus_irq = 0;
}