numam-dpdk/examples/vhost_scsi/vhost_scsi.c
Stefan Hajnoczi d56f632075 examples/vhost_scsi: drop unimplemented event index feature
The vhost_scsi example application negotiates the
VIRTIO_RING_F_EVENT_IDX feature bit but does not honor it when accessing
vrings.

In particular, commit e37ff95440 ("vhost:
support virtqueue interrupt/notification suppression") broke vring call
because vq->last_used_idx is never updated by vhost_scsi.  The
vq->last_used_idx field is not even available via the librte_vhost
public API, so VIRTIO_RING_F_EVENT_IDX is currently only usable by the
built-in virtio_net.c driver in librte_vhost.

This patch drops VIRTIO_RING_F_EVENT_IDX from vhost_scsi so that vring
call works again.

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Changpeng Liu <changpeng.liu@intel.com>
Acked-by: Yuanhan Liu <yliu@fridaylinux.org>
2018-02-05 15:18:56 +01:00

445 lines
11 KiB
C

/* SPDX-License-Identifier: BSD-3-Clause
* Copyright(c) 2010-2017 Intel Corporation
*/
#include <stdint.h>
#include <unistd.h>
#include <stdbool.h>
#include <signal.h>
#include <assert.h>
#include <semaphore.h>
#include <linux/virtio_scsi.h>
#include <linux/virtio_ring.h>
#include <rte_atomic.h>
#include <rte_cycles.h>
#include <rte_log.h>
#include <rte_malloc.h>
#include <rte_vhost.h>
#include "vhost_scsi.h"
#include "scsi_spec.h"
#define VIRTIO_SCSI_FEATURES ((1 << VIRTIO_F_NOTIFY_ON_EMPTY) |\
(1 << VIRTIO_SCSI_F_INOUT) |\
(1 << VIRTIO_SCSI_F_CHANGE))
/* Path to folder where character device will be created. Can be set by user. */
static char dev_pathname[PATH_MAX] = "";
static struct vhost_scsi_ctrlr *g_vhost_ctrlr;
static int g_should_stop;
static sem_t exit_sem;
static struct vhost_scsi_ctrlr *
vhost_scsi_ctrlr_find(__rte_unused const char *ctrlr_name)
{
/* currently we only support 1 socket file fd */
return g_vhost_ctrlr;
}
static uint64_t gpa_to_vva(int vid, uint64_t gpa)
{
char path[PATH_MAX];
struct vhost_scsi_ctrlr *ctrlr;
int ret = 0;
ret = rte_vhost_get_ifname(vid, path, PATH_MAX);
if (ret) {
fprintf(stderr, "Cannot get socket name\n");
assert(ret != 0);
}
ctrlr = vhost_scsi_ctrlr_find(path);
if (!ctrlr) {
fprintf(stderr, "Controller is not ready\n");
assert(ctrlr != NULL);
}
assert(ctrlr->mem != NULL);
return rte_vhost_gpa_to_vva(ctrlr->mem, gpa);
}
static struct vring_desc *
descriptor_get_next(struct vring_desc *vq_desc, struct vring_desc *cur_desc)
{
return &vq_desc[cur_desc->next];
}
static bool
descriptor_has_next(struct vring_desc *cur_desc)
{
return !!(cur_desc->flags & VRING_DESC_F_NEXT);
}
static bool
descriptor_is_wr(struct vring_desc *cur_desc)
{
return !!(cur_desc->flags & VRING_DESC_F_WRITE);
}
static void
submit_completion(struct vhost_scsi_task *task, uint32_t q_idx)
{
struct rte_vhost_vring *vq;
struct vring_used *used;
vq = task->vq;
used = vq->used;
/* Fill out the next entry in the "used" ring. id = the
* index of the descriptor that contained the SCSI request.
* len = the total amount of data transferred for the SCSI
* request. We must report the correct len, for variable
* length SCSI CDBs, where we may return less data than
* allocated by the guest VM.
*/
used->ring[used->idx & (vq->size - 1)].id = task->req_idx;
used->ring[used->idx & (vq->size - 1)].len = task->data_len;
used->idx++;
/* Send an interrupt back to the guest VM so that it knows
* a completion is ready to be processed.
*/
rte_vhost_vring_call(task->bdev->vid, q_idx);
}
static void
vhost_process_read_payload_chain(struct vhost_scsi_task *task)
{
void *data;
task->iovs_cnt = 0;
task->resp = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
task->desc->addr);
while (descriptor_has_next(task->desc)) {
task->desc = descriptor_get_next(task->vq->desc, task->desc);
data = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
task->desc->addr);
task->iovs[task->iovs_cnt].iov_base = data;
task->iovs[task->iovs_cnt].iov_len = task->desc->len;
task->data_len += task->desc->len;
task->iovs_cnt++;
}
}
static void
vhost_process_write_payload_chain(struct vhost_scsi_task *task)
{
void *data;
task->iovs_cnt = 0;
do {
data = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
task->desc->addr);
task->iovs[task->iovs_cnt].iov_base = data;
task->iovs[task->iovs_cnt].iov_len = task->desc->len;
task->data_len += task->desc->len;
task->iovs_cnt++;
task->desc = descriptor_get_next(task->vq->desc, task->desc);
} while (descriptor_has_next(task->desc));
task->resp = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
task->desc->addr);
}
static struct vhost_block_dev *
vhost_scsi_bdev_construct(const char *bdev_name, const char *bdev_serial,
uint32_t blk_size, uint64_t blk_cnt,
bool wce_enable)
{
struct vhost_block_dev *bdev;
bdev = rte_zmalloc(NULL, sizeof(*bdev), RTE_CACHE_LINE_SIZE);
if (!bdev)
return NULL;
strncpy(bdev->name, bdev_name, sizeof(bdev->name));
strncpy(bdev->product_name, bdev_serial, sizeof(bdev->product_name));
bdev->blocklen = blk_size;
bdev->blockcnt = blk_cnt;
bdev->write_cache = wce_enable;
/* use memory as disk storage space */
bdev->data = rte_zmalloc(NULL, blk_cnt * blk_size, 0);
if (!bdev->data) {
fprintf(stderr, "no enough reseverd huge memory for disk\n");
return NULL;
}
return bdev;
}
static void
process_requestq(struct vhost_scsi_ctrlr *ctrlr, uint32_t q_idx)
{
int ret;
struct vhost_scsi_queue *scsi_vq;
struct rte_vhost_vring *vq;
scsi_vq = &ctrlr->bdev->queues[q_idx];
vq = &scsi_vq->vq;
ret = rte_vhost_get_vhost_vring(ctrlr->bdev->vid, q_idx, vq);
assert(ret == 0);
while (vq->avail->idx != scsi_vq->last_used_idx) {
int req_idx;
uint16_t last_idx;
struct vhost_scsi_task *task;
last_idx = scsi_vq->last_used_idx & (vq->size - 1);
req_idx = vq->avail->ring[last_idx];
task = rte_zmalloc(NULL, sizeof(*task), 0);
assert(task != NULL);
task->ctrlr = ctrlr;
task->bdev = ctrlr->bdev;
task->vq = vq;
task->req_idx = req_idx;
task->desc = &task->vq->desc[task->req_idx];
/* does not support indirect descriptors */
assert((task->desc->flags & VRING_DESC_F_INDIRECT) == 0);
scsi_vq->last_used_idx++;
task->req = (void *)(uintptr_t)gpa_to_vva(task->bdev->vid,
task->desc->addr);
task->desc = descriptor_get_next(task->vq->desc, task->desc);
if (!descriptor_has_next(task->desc)) {
task->dxfer_dir = SCSI_DIR_NONE;
task->resp = (void *)(uintptr_t)
gpa_to_vva(task->bdev->vid,
task->desc->addr);
} else if (!descriptor_is_wr(task->desc)) {
task->dxfer_dir = SCSI_DIR_TO_DEV;
vhost_process_write_payload_chain(task);
} else {
task->dxfer_dir = SCSI_DIR_FROM_DEV;
vhost_process_read_payload_chain(task);
}
ret = vhost_bdev_process_scsi_commands(ctrlr->bdev, task);
if (ret) {
/* invalid response */
task->resp->response = VIRTIO_SCSI_S_BAD_TARGET;
} else {
/* successfully */
task->resp->response = VIRTIO_SCSI_S_OK;
task->resp->status = 0;
task->resp->resid = 0;
}
submit_completion(task, q_idx);
rte_free(task);
}
}
/* Main framework for processing IOs */
static void *
ctrlr_worker(void *arg)
{
uint32_t idx, num;
struct vhost_scsi_ctrlr *ctrlr = (struct vhost_scsi_ctrlr *)arg;
cpu_set_t cpuset;
pthread_t thread;
thread = pthread_self();
CPU_ZERO(&cpuset);
CPU_SET(0, &cpuset);
pthread_setaffinity_np(thread, sizeof(cpu_set_t), &cpuset);
num = rte_vhost_get_vring_num(ctrlr->bdev->vid);
fprintf(stdout, "Ctrlr Worker Thread Started with %u Vring\n", num);
if (num != NUM_OF_SCSI_QUEUES) {
fprintf(stderr, "Only 1 IO queue are supported\n");
exit(0);
}
while (!g_should_stop && ctrlr->bdev != NULL) {
/* At least 3 vrings, currently only can support 1 IO queue
* Queue 2 for IO queue, does not support TMF and hotplug
* for the example application now
*/
for (idx = 2; idx < num; idx++)
process_requestq(ctrlr, idx);
}
fprintf(stdout, "Ctrlr Worker Thread Exiting\n");
sem_post(&exit_sem);
return NULL;
}
static int
new_device(int vid)
{
char path[PATH_MAX];
struct vhost_scsi_ctrlr *ctrlr;
struct vhost_scsi_queue *scsi_vq;
struct rte_vhost_vring *vq;
pthread_t tid;
int i, ret;
ret = rte_vhost_get_ifname(vid, path, PATH_MAX);
if (ret) {
fprintf(stderr, "Cannot get socket name\n");
return -1;
}
ctrlr = vhost_scsi_ctrlr_find(path);
if (!ctrlr) {
fprintf(stderr, "Controller is not ready\n");
return -1;
}
ret = rte_vhost_get_mem_table(vid, &ctrlr->mem);
if (ret) {
fprintf(stderr, "Get Controller memory region failed\n");
return -1;
}
assert(ctrlr->mem != NULL);
/* hardcoded block device information with 128MiB */
ctrlr->bdev = vhost_scsi_bdev_construct("malloc0", "vhost_scsi_malloc0",
4096, 32768, 0);
if (!ctrlr->bdev)
return -1;
ctrlr->bdev->vid = vid;
/* Disable Notifications */
for (i = 0; i < NUM_OF_SCSI_QUEUES; i++) {
rte_vhost_enable_guest_notification(vid, i, 0);
/* restore used index */
scsi_vq = &ctrlr->bdev->queues[i];
vq = &scsi_vq->vq;
ret = rte_vhost_get_vhost_vring(ctrlr->bdev->vid, i, vq);
assert(ret == 0);
scsi_vq->last_used_idx = vq->used->idx;
scsi_vq->last_avail_idx = vq->used->idx;
}
g_should_stop = 0;
fprintf(stdout, "New Device %s, Device ID %d\n", path, vid);
if (pthread_create(&tid, NULL, &ctrlr_worker, ctrlr) < 0) {
fprintf(stderr, "Worker Thread Started Failed\n");
return -1;
}
pthread_detach(tid);
return 0;
}
static void
destroy_device(int vid)
{
char path[PATH_MAX];
struct vhost_scsi_ctrlr *ctrlr;
rte_vhost_get_ifname(vid, path, PATH_MAX);
fprintf(stdout, "Destroy %s Device ID %d\n", path, vid);
ctrlr = vhost_scsi_ctrlr_find(path);
if (!ctrlr) {
fprintf(stderr, "Destroy Ctrlr Failed\n");
return;
}
ctrlr->bdev = NULL;
g_should_stop = 1;
sem_wait(&exit_sem);
}
static const struct vhost_device_ops vhost_scsi_device_ops = {
.new_device = new_device,
.destroy_device = destroy_device,
};
static struct vhost_scsi_ctrlr *
vhost_scsi_ctrlr_construct(const char *ctrlr_name)
{
int ret;
struct vhost_scsi_ctrlr *ctrlr;
char *path;
char cwd[PATH_MAX];
/* always use current directory */
path = getcwd(cwd, PATH_MAX);
if (!path) {
fprintf(stderr, "Cannot get current working directory\n");
return NULL;
}
snprintf(dev_pathname, sizeof(dev_pathname), "%s/%s", path, ctrlr_name);
if (access(dev_pathname, F_OK) != -1) {
if (unlink(dev_pathname) != 0)
rte_exit(EXIT_FAILURE, "Cannot remove %s.\n",
dev_pathname);
}
if (rte_vhost_driver_register(dev_pathname, 0) != 0) {
fprintf(stderr, "socket %s already exists\n", dev_pathname);
return NULL;
}
fprintf(stdout, "socket file: %s created\n", dev_pathname);
ret = rte_vhost_driver_set_features(dev_pathname, VIRTIO_SCSI_FEATURES);
if (ret != 0) {
fprintf(stderr, "Set vhost driver features failed\n");
return NULL;
}
ctrlr = rte_zmalloc(NULL, sizeof(*ctrlr), RTE_CACHE_LINE_SIZE);
if (!ctrlr)
return NULL;
rte_vhost_driver_callback_register(dev_pathname,
&vhost_scsi_device_ops);
return ctrlr;
}
static void
signal_handler(__rte_unused int signum)
{
if (access(dev_pathname, F_OK) == 0)
unlink(dev_pathname);
exit(0);
}
int main(int argc, char *argv[])
{
int ret;
signal(SIGINT, signal_handler);
/* init EAL */
ret = rte_eal_init(argc, argv);
if (ret < 0)
rte_exit(EXIT_FAILURE, "Error with EAL initialization\n");
g_vhost_ctrlr = vhost_scsi_ctrlr_construct("vhost.socket");
if (g_vhost_ctrlr == NULL) {
fprintf(stderr, "Construct vhost scsi controller failed\n");
return 0;
}
if (sem_init(&exit_sem, 0, 0) < 0) {
fprintf(stderr, "Error init exit_sem\n");
return -1;
}
rte_vhost_driver_start(dev_pathname);
/* loop for exit the application */
while (1)
sleep(1);
return 0;
}