#define RENDEZVOUS_WAIT_SECS 60
#define RENDEZVOUS_RETRY_SECS 1
+#define DEEP_DEBUG 0
+#define SHALLOW_DEBUG 0
+
+#if DEEP_DEBUG
+#define DEEP_DEBUG_PRINT(fmt, args...) printk((fmt), ##args)
+#else
+#define DEEP_DEBUG_PRINT(fmt, args...)
+#endif
+
+#if SHALLOW_DEBUG
+#define SHALLOW_DEBUG_PRINT(fmt, args...) printk((fmt), ##args)
+#else
+#define SHALLOW_DEBUG_PRINT(fmt, args...)
+#endif
+
+
+#define ERROR(fmt, args...) printk((fmt), ##args)
+#define INFO(fmt, args...) printk((fmt), ##args)
struct palacios_host_device_user {
spinlock_t lock;
kfree(dev);
}
+
+//
+// Is this structure big enough for the data_size we will use?
+//
+// THIS FUNCTION CAN BE CALLED WHILE INTERRUPTS ARE OFF
+//
+static int palacios_bigenough_reqresp(struct palacios_host_dev_host_request_response *r, uint64_t data_size)
+{
+ if (!r) {
+ return 0;
+ } else {
+ if (((r->len)-sizeof(struct palacios_host_dev_host_request_response)) < data_size) {
+ return 0;
+ } else {
+ return 1;
+ }
+ }
+}
+
+//
+// Resize a request/response structure so that it will fit data_size bytes
+//
+// At the end of this, *r->len >= sizeof(struct)+data_size
+//
+// THIS FUNCTION MAY SLEEP AS IT CALLS KMALLOC
+// DO NOT CALL IT WHILE HOLDING A SPIN LOCK WITH INTERRUPTS OFF
+//
static int palacios_resize_reqresp(struct palacios_host_dev_host_request_response **r, uint64_t data_size, int copy)
{
- if (!*r) {
+
+ DEEP_DEBUG_PRINT("palacios: hostdev: resize 0x%p to %llu\n",*r,data_size);
+
+ if ((*r)==0) {
// allocate it
+ DEEP_DEBUG_PRINT("palacios: hostdev: attempt alloc\n");
*r = kmalloc(sizeof(struct palacios_host_dev_host_request_response)+data_size,GFP_KERNEL);
- if (!*r) {
+ DEEP_DEBUG_PRINT("palacios: hostdev: kmalloc done\n");
+ if ((*r)==0) {
+ ERROR("palacios: hostdev: failed to allocate\n");
return -1;
} else {
(*r)->len=sizeof(struct palacios_host_dev_host_request_response)+data_size;
+ DEEP_DEBUG_PRINT("palacios: hostdev: allocated\n");
return 0;
}
} else {
if (data_size<=cur_len) {
// do nothing
+ DEEP_DEBUG_PRINT("palacios: hostdev: size ok\n");
return 0;
} else {
struct palacios_host_dev_host_request_response *new;
}
new = kmalloc(sizeof(struct palacios_host_dev_host_request_response)+data_size,GFP_KERNEL);
if (!new) {
+ ERROR("palacios: hostdev: failed to reallocate\n");
return -1;
} else {
new->len=sizeof(struct palacios_host_dev_host_request_response)+data_size;
kfree(*r);
}
*r=new;
+ DEEP_DEBUG_PRINT("palacios: hostdev: reallocated\n");
return 0;
}
}
static void cycle_request_response(struct palacios_host_device_user *dev)
{
+ DEEP_DEBUG_PRINT("palacios: hostdev: cycle request to response\n");
// wake up user side so that polls fall through
wake_up_interruptible(&(dev->user_wait_queue));
// put us to sleep until the user side wakes us up
- wait_event_interruptible((dev->host_wait_queue), (dev->waiting==0));
+ while (wait_event_interruptible((dev->host_wait_queue), (dev->waiting==0)) != 0) {}
+
+ DEEP_DEBUG_PRINT("palacios: hostdev: cycle request to response - done\n");
}
static void cycle_response_request(struct palacios_host_device_user *dev)
{
+ DEEP_DEBUG_PRINT("palacios: hostdev: cycle response to request\n");
// wake up host side
wake_up_interruptible(&(dev->host_wait_queue));
+ DEEP_DEBUG_PRINT("palacios: hostdev: cycle response to request - done\n");
}
struct palacios_host_device_user * dev = filp->private_data;
unsigned long f;
+ SHALLOW_DEBUG_PRINT("palacios: hostdev: poll\n");
+
if (!dev->connected) {
+ ERROR("palcios: hostdev: poll on unconnected device\n");
return -EFAULT;
}
if (dev->waiting) {
// Yes, we have a request if you want it!
spin_unlock_irqrestore(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: hostdev: poll done immediate\n");
return POLLIN | POLLRDNORM;
}
spin_unlock_irqrestore(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: hostdev: poll delayed\n");
// We will get called again when that queue is woken up
return 0;
struct palacios_host_device_user *dev = filp->private_data;
unsigned long f;
- printk("palacios: user side is closing host device \"%s\"\n",dev->url);
+ INFO("palacios: user side is closing host device \"%s\"\n",dev->url);
spin_lock_irqsave(&(dev->lock), f);
dev->connected = 0;
struct palacios_host_device_user *dev = fp->private_data;
+ DEEP_DEBUG_PRINT("palacios: hostdev: ioctl %u\n",val);
+
if (!dev->connected) {
+ ERROR("palacios: hostdev: ioctl on unconnected device\n");
return -EFAULT;
}
struct palacios_host_dev_user_op op;
if (copy_from_user(&op,argp,sizeof(struct palacios_host_dev_user_op))) {
- printk("palacios: unable to copy from user for host device \"%s\"\n",dev->url);
+ ERROR("palacios: unable to copy from user for host device \"%s\"\n",dev->url);
return -EFAULT;
}
+
+ DEEP_DEBUG_PRINT("palacios: hostdev: user request push, type %d\n",op.type);
switch (op.type) {
case PALACIOS_HOST_DEV_USER_REQUEST_READ_GUEST: {
void *temp = kmalloc(op.len,GFP_KERNEL);
+ DEEP_DEBUG_PRINT("palacios: hostdev: read guest\n");
+
if (!temp) {
- printk("palacios: unable to allocate enough for read guest request for host device \"%s\"\n",dev->url);
+ ERROR("palacios: unable to allocate enough for read guest request for host device \"%s\"\n",dev->url);
return -EFAULT;
}
op.gpa,
temp,
op.len) != op.len) {
- printk("palacios: unable to read enough from guest for host device \"%s\"\n",dev->url);
+ ERROR("palacios: unable to read enough from guest for host device \"%s\"\n",dev->url);
kfree(temp);
return -EFAULT;
}
if (copy_to_user(op.data,temp,op.len)) {
- printk("palacios: unable to copy to user for host device \"%s\"\n",dev->url);
+ ERROR("palacios: unable to copy to user for host device \"%s\"\n",dev->url);
kfree(temp);
return -EFAULT;
}
case PALACIOS_HOST_DEV_USER_REQUEST_WRITE_GUEST: {
+ void *temp;
+
+ DEEP_DEBUG_PRINT("palacios: hostdev: write guest\n");
- void *temp = kmalloc(op.len,GFP_KERNEL);
+ temp = kmalloc(op.len,GFP_KERNEL);
if (!temp) {
- printk("palacios: unable to allocate enough for write guest request for host device \"%s\"\n",dev->url);
+ ERROR("palacios: unable to allocate enough for write guest request for host device \"%s\"\n",dev->url);
return -EFAULT;
}
if (copy_from_user(temp,op.data,op.len)) {
- printk("palacios: unable to copy from user for host device \"%s\"\n",dev->url);
+ ERROR("palacios: unable to copy from user for host device \"%s\"\n",dev->url);
kfree(temp);
return -EFAULT;
}
op.gpa,
temp,
op.len) != op.len) {
- printk("palacios: unable to write enough to guest for host device \"%s\"\n",dev->url);
+ ERROR("palacios: unable to write enough to guest for host device \"%s\"\n",dev->url);
kfree(temp);
return -EFAULT;
}
case PALACIOS_HOST_DEV_USER_REQUEST_IRQ_GUEST: {
+ DEEP_DEBUG_PRINT("palacios: hostdev: irq guest\n");
+
return v3_host_dev_raise_irq(dev->guestdev, dev, op.irq);
}
break;
default:
- printk("palacios: unknown user request to host device \"%s\"\n",dev->url);
+ ERROR("palacios: unknown user request to host device \"%s\"\n",dev->url);
return -EFAULT;
break;
}
case V3_HOST_DEV_HOST_REQUEST_SIZE_IOCTL: {
unsigned long f;
+
+
+ DEEP_DEBUG_PRINT("palacios: hostdev: request size of request\n");
spin_lock_irqsave(&(dev->lock),f);
if (!(dev->waiting)) {
spin_unlock_irqrestore(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: hostdev: no request available\n");
+ schedule(); // avoid livelock for polling user space process SUSPICOUS
return 0; // no request available now
}
if (copy_to_user(argp,&(dev->req->data_len),sizeof(uint64_t))) {
- printk("palacios: unable to copy to user for host device \"%s\"\n",dev->url);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: unable to copy to user for host device \"%s\"\n",dev->url);
return -EFAULT; // failed to copy!
+
}
spin_unlock_irqrestore(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: hostdev: have request\n");
+
return 1; // have request for you
}
spin_lock_irqsave(&(dev->lock),f);
- if (!(dev->waiting)) {
+ DEEP_DEBUG_PRINT("palacios: hostdev: pull request\n");
+
+ if (!(dev->waiting) || !(dev->req)) {
spin_unlock_irqrestore(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: hostdev: no request to pull\n");
return 0; // no request available now
}
+
if (copy_to_user(argp,dev->req,dev->req->data_len)) {
- printk("palacios: unable to copy to user for host device \"%s\"\n",dev->url);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: unable to copy to user for host device \"%s\"\n",dev->url);
return -EFAULT; // failed to copy!
}
-
+
spin_unlock_irqrestore(&(dev->lock),f);
+
+ DEEP_DEBUG_PRINT("palacios: hostdev: request pulled\n");
return 1; // copied request for you
}
unsigned long f;
uint64_t user_datalen;
+ uint64_t old_len;
spin_lock_irqsave(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: hostdev: push response\n");
+
if (!(dev->waiting)) {
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: hostdev: no matching request for pushed response\n");
return 0; // no request outstanding, so we do not need a response!
}
if (copy_from_user(&user_datalen,argp,sizeof(uint64_t))) {
- printk("palacios: unable to copy from user for host device \"%s\"\n",dev->url);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: unable to copy from user for host device \"%s\"\n",dev->url);
return -EFAULT; // failed to copy!
}
-
- if (palacios_resize_reqresp(&(dev->resp),user_datalen,0)) {
- printk("palacios: unable to resize to accept request of size %llu from user for host device \"%s\"\n",user_datalen,dev->url);
+
+ if (user_datalen<sizeof(struct palacios_host_dev_host_request_response)) {
+ // bad user
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: user has response that is too small on host device \"%s\"\n",dev->url);
return -EFAULT;
- }
-
+ }
+
+ if (!palacios_bigenough_reqresp(dev->resp,user_datalen-sizeof(struct palacios_host_dev_host_request_response))) {
+ // not enough room.
+ // we drop the lock, turn on interrupts, resize, and then retry
+ DEEP_DEBUG_PRINT("palacios: response not big enough, dropping lock to resize on device \"%s\"\n",dev->url);
+
+ spin_unlock_irqrestore(&(dev->lock),f);
+
+ if (palacios_resize_reqresp(&(dev->resp),user_datalen-sizeof(struct palacios_host_dev_host_request_response),0)) {
+ ERROR("palacios: unable to resize to accept response of size %llu from user for host device \"%s\"\n",user_datalen,dev->url);
+ return -EFAULT;
+ } else {
+ // reacquire the lock
+ // There shouldn't be a race here, since there should
+ // be exactly one user space thread giving us a response for this device
+ // and it is blocked waiting for us to finish
+ spin_lock_irqsave(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: reacuired lock on device \"%s\"\n",dev->url);
+ }
+ }
+
+ //We only copy data_len bytes from user, but we will
+ //overwrite the len field, so we preserve and then restore
+ old_len = dev->resp->len;
if (copy_from_user(dev->resp, argp, user_datalen)) {
- printk("palacios: unable to copy from user for host device \"%s\"\n",dev->url);
+ dev->resp->len=old_len;
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: unable to copy from user for host device \"%s\"\n",dev->url);
return -EFAULT; // failed to copy!
}
+ dev->resp->len=old_len;
+ DEEP_DEBUG_PRINT("palacios: hostdev: valid response pushed\n");
// now have valid response!
dev->waiting=0;
+ spin_unlock_irqrestore(&(dev->lock),f);
+
// wake the palacios side up so that it sees it
cycle_response_request(dev);
- spin_unlock_irqrestore(&(dev->lock),f);
-
return 1; // done
}
break;
default:
- printk("palacios: unknown ioctl for host device \"%s\"\n",dev->url);
+ ERROR("palacios: unknown ioctl for host device \"%s\"\n",dev->url);
return -EFAULT;
break;
}
// currently only support user: types:
if (strncasecmp(url,"user:",5)) {
- printk("palacios: do not currently support host devices of type in \"%s\"\n",url);
+ ERROR("palacios: do not currently support host devices of type in \"%s\"\n",url);
return -1;
}
- printk("palacios: attempting to rendezvous with palacios side of host device \"%s\"\n",url);
+ INFO("palacios: attempting to rendezvous with palacios side of host device \"%s\"\n",url);
// We will scan the list looking for the relevant
// URL. If we don't find it after a while, we give up
// found it
spin_lock_irqsave(&(dev->lock),f2);
if (dev->connected) {
- printk("palacios: device for \"%s\" is already connected!\n",url);
+ ERROR("palacios: device for \"%s\" is already connected!\n",url);
spin_unlock_irqrestore(&(dev->lock),f2);
spin_unlock_irqrestore(&(guest->hostdev.lock),f1);
return -1;
} else {
dev->fd = anon_inode_getfd("v3-hostdev", &host_dev_fops, dev, 0);
if (dev->fd<0) {
- printk("palacios: cannot create fd for device \"%s\"\n",url);
+ ERROR("palacios: cannot create fd for device \"%s\"\n",url);
spin_unlock_irqrestore(&(dev->lock),f2);
spin_unlock_irqrestore(&(guest->hostdev.lock),f1);
return -1;
kfree(dev->resp);
dev->resp=0;
}
- printk("palacios: connected fd for device \"%s\"\n",url);
+ INFO("palacios: connected fd for device \"%s\"\n",url);
spin_unlock_irqrestore(&(dev->lock),f2);
spin_unlock_irqrestore(&(guest->hostdev.lock),f1);
return dev->fd;
ssleep(RENDEZVOUS_RETRY_SECS);
}
- printk("palacios: timeout waiting for connection for device \"%s\"",url);
+ ERROR("palacios: timeout waiting for connection for device \"%s\"",url);
return -1;
if (strncasecmp(url,"user:",5)) {
- printk("palacios: do not currently support devices of type in \"%s\"\n",url);
+ ERROR("palacios: do not currently support devices of type in \"%s\"\n",url);
return NULL;
}
if (!strncasecmp(url,dev->url,MAX_URL)) {
// found it
spin_unlock_irqrestore(&(guest->hostdev.lock),f1);
- printk("palacios: a host device with url \"%s\" already exists in the guest!\n",url);
+ ERROR("palacios: a host device with url \"%s\" already exists in the guest!\n",url);
return NULL;
}
}
spin_unlock_irqrestore(&(guest->hostdev.lock),f1);
- printk("palacios: creating host device \"%s\"\n",url);
+ INFO("palacios: creating host device \"%s\"\n",url);
dev = kmalloc(sizeof(struct palacios_host_device_user),GFP_KERNEL);
if (!dev) {
- printk("palacios: cannot allocate for host device \"%s\"\n",url);
+ ERROR("palacios: cannot allocate for host device \"%s\"\n",url);
return NULL;
}
init_waitqueue_head(&(dev->user_wait_queue));
init_waitqueue_head(&(dev->host_wait_queue));
- printk("palacios: attempting to rendezvous with user side of host device \"%s\"\n",url);
+ INFO("palacios: attempting to rendezvous with user side of host device \"%s\"\n",url);
// Insert ourselves into the list
spin_lock_irqsave(&(guest->hostdev.lock),f1);
for (i=0;i<RENDEZVOUS_WAIT_SECS/RENDEZVOUS_RETRY_SECS;i++) {
spin_lock_irqsave(&(dev->lock),f2);
if (dev->connected){
- printk("palacios: connection with user side established for host device \"%s\" fd=%d\n",dev->url,dev->fd);
+ INFO("palacios: connection with user side established for host device \"%s\" fd=%d\n",dev->url,dev->fd);
spin_unlock_irqrestore(&(dev->lock),f2);
return dev;
}
ssleep(RENDEZVOUS_RETRY_SECS);
}
- printk("palacios: timeout waiting for user side to connect to host device \"%s\"",url);
+ ERROR("palacios: timeout waiting for user side to connect to host device \"%s\"",url);
// get us out of the list
spin_lock_irqsave(&(guest->hostdev.lock),f1);
struct palacios_host_device_user *dev = (struct palacios_host_device_user *) hostdev;
- printk("palacios: closing host device \"%s\"\n",dev->url);
+ INFO("palacios: closing host device \"%s\"\n",dev->url);
spin_lock_irqsave(&(dev->guest->hostdev.lock),f1);
unsigned long f;
uint64_t op_len;
+ DEEP_DEBUG_PRINT("palacios: hostdev: read io port 0x%x\n",port);
+
+
spin_lock_irqsave(&(dev->lock),f);
if (dev->waiting) {
- printk("palacios: guest issued i/o read request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: guest issued i/o read request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
return 0;
}
if (!dev->connected) {
- printk("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
return 0;
}
- // resize request and response in case they will need it
- palacios_resize_reqresp(&(dev->req),0,0); // request doesn't carry data
+
+ // resize request (no data)
+ if (!palacios_bigenough_reqresp(dev->req,0)) {
+ // not enough room.
+ // we drop the lock, turn on interrupts, resize, and then retry
+ DEEP_DEBUG_PRINT("palacios: request not big enough, dropping lock to resize on device \"%s\"\n",dev->url);
+
+ spin_unlock_irqrestore(&(dev->lock),f);
+
+ if (palacios_resize_reqresp(&(dev->req),0,0)) {
+ ERROR("palacios: cannot resize for request on device \"%s\"\n",dev->url);
+ return 0;
+ } else {
+ // reacquire the lock
+ // There shouldn't be a race here since there should not be another
+ // request from palacios until this one finishes
+ spin_lock_irqsave(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: reacquired lock on device \"%s\"\n",dev->url);
+ }
+ }
+
dev->req->type=PALACIOS_HOST_DEV_HOST_REQUEST_READ_IO;
dev->req->port=port;
unsigned long f;
uint64_t op_len;
+ DEEP_DEBUG_PRINT("palacios: hostdev: read mem 0x%p\n",gpa);
+
spin_lock_irqsave(&(dev->lock),f);
if (dev->waiting) {
- printk("palacios: guest issued memory read request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: guest issued memory read request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
return 0;
}
if (!dev->connected) {
- printk("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
return 0;
}
- // resize request and response in case they will need it
- palacios_resize_reqresp(&(dev->req),0,0); // request doesn't carry data
+ // resize request (no data)
+ if (!palacios_bigenough_reqresp(dev->req,0)) {
+ // not enough room.
+ // we drop the lock, turn on interrupts, resize, and then retry
+ DEEP_DEBUG_PRINT("palacios: request not big enough, dropping lock to resize on device \"%s\"\n",dev->url);
+
+ spin_unlock_irqrestore(&(dev->lock),f);
+
+ if (palacios_resize_reqresp(&(dev->req),0,0)) {
+ ERROR("palacios: cannot resize for request on device \"%s\"\n",dev->url);
+ return 0;
+ } else {
+ // reacquire the lock
+ // There shouldn't be a race here since there should not be another
+ // request from palacios until this one finishes
+ spin_lock_irqsave(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: reacquired lock on device \"%s\"\n",dev->url);
+ }
+ }
dev->req->type=PALACIOS_HOST_DEV_HOST_REQUEST_READ_MEM;
dev->req->port=0;
unsigned long f;
uint64_t op_len;
+ DEEP_DEBUG_PRINT("palacios: hostdev: read conf 0x%p\n",(void*)offset);
+
spin_lock_irqsave(&(dev->lock),f);
if (dev->waiting) {
- printk("palacios: guest issued config read request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: guest issued config read request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
return 0;
}
if (!dev->connected) {
- printk("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
return 0;
}
- // resize request and response in case they will need it
- palacios_resize_reqresp(&(dev->req),0,0); // request doesn't carry data
+ // resize request (no data)
+ if (!palacios_bigenough_reqresp(dev->req,0)) {
+ // not enough room.
+ // we drop the lock, turn on interrupts, resize, and then retry
+ DEEP_DEBUG_PRINT("palacios: request not big enough, dropping lock to resize on device \"%s\"\n",dev->url);
+
+ spin_unlock_irqrestore(&(dev->lock),f);
+
+ if (palacios_resize_reqresp(&(dev->req),0,0)) {
+ ERROR("palacios: cannot resize for request on device \"%s\"\n",dev->url);
+ return 0;
+ } else {
+ // reacquire the lock
+ // There shouldn't be a race here since there should not be another
+ // request from palacios until this one finishes
+ spin_lock_irqsave(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: reacquired lock on device \"%s\"\n",dev->url);
+ }
+ }
dev->req->type=PALACIOS_HOST_DEV_HOST_REQUEST_READ_CONF;
dev->req->port=0;
unsigned long f;
uint64_t op_len;
+ DEEP_DEBUG_PRINT("palacios: hostdev: write io port 0x%x \n",port);
+
spin_lock_irqsave(&(dev->lock),f);
if (dev->waiting) {
- printk("palacios: guest issued i/o write request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: guest issued i/o write request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
return 0;
}
if (!dev->connected) {
- printk("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
return 0;
}
- // resize request and response in case they will need it
- palacios_resize_reqresp(&(dev->req),len,0); // make room for data
+ // resize request
+ if (!palacios_bigenough_reqresp(dev->req,len)) {
+ // not enough room.
+ // we drop the lock, turn on interrupts, resize, and then retry
+ DEEP_DEBUG_PRINT("palacios: request not big enough, dropping lock to resize on device \"%s\"\n",dev->url);
+
+ spin_unlock_irqrestore(&(dev->lock),f);
+
+ if (palacios_resize_reqresp(&(dev->req),len,0)) {
+ ERROR("palacios: cannot resize for request on device \"%s\"\n",dev->url);
+ return 0;
+ } else {
+ // reacquire the lock
+ // There shouldn't be a race here since there should not be another
+ // request from palacios until this one finishes
+ spin_lock_irqsave(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: reacquired lock on device \"%s\"\n",dev->url);
+ }
+ }
dev->req->type=PALACIOS_HOST_DEV_HOST_REQUEST_WRITE_IO;
dev->req->port=port;
memcpy(dev->req->data,src,len);
dev->waiting=1;
-
- spin_unlock_irqrestore(&(dev->lock),f);
+ spin_unlock_irqrestore(&(dev->lock),f);
+
// hand over to the user space and wait for it to respond
cycle_request_response(dev);
unsigned long f;
uint64_t op_len;
+ DEEP_DEBUG_PRINT("palacios: hostdev: write mem 0x%p\n",gpa);
+
spin_lock_irqsave(&(dev->lock),f);
if (dev->waiting) {
- printk("palacios: guest issued memory write request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: guest issued memory write request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
return 0;
}
if (!dev->connected) {
- printk("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
return 0;
}
- // resize request and response in case they will need it
- palacios_resize_reqresp(&(dev->req),len,0); // make room for data
+ // resize request
+ if (!palacios_bigenough_reqresp(dev->req,len)) {
+ // not enough room.
+ // we drop the lock, turn on interrupts, resize, and then retry
+ DEEP_DEBUG_PRINT("palacios: request not big enough, dropping lock to resize on device \"%s\"\n",dev->url);
+
+ spin_unlock_irqrestore(&(dev->lock),f);
+
+ if (palacios_resize_reqresp(&(dev->req),len,0)) {
+ ERROR("palacios: cannot resize for request on device \"%s\"\n",dev->url);
+ return 0;
+ } else {
+ // reacquire the lock
+ // There shouldn't be a race here since there should not be another
+ // request from palacios until this one finishes
+ spin_lock_irqsave(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: reacquired lock on device \"%s\"\n",dev->url);
+ }
+ }
dev->req->type=PALACIOS_HOST_DEV_HOST_REQUEST_WRITE_MEM;
dev->req->port=0;
}
-static int palacios_host_dev_ack_irq(v3_host_dev_t hostdev, uint8_t irq)
-{
- // we don't care
- return 0;
-}
-
static uint64_t palacios_host_dev_write_conf(v3_host_dev_t hostdev,
unsigned long f;
uint64_t op_len;
+ DEEP_DEBUG_PRINT("palacios: hostdev: write conf 0x%p\n",(void*)offset);
+
spin_lock_irqsave(&(dev->lock),f);
if (dev->waiting) {
- printk("palacios: guest issued config write request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: guest issued config write request with host device \"%s\" in wrong state (waiting=%d, connected=%d)\n",dev->url,dev->waiting,dev->connected);
return 0;
}
if (!dev->connected) {
- printk("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
spin_unlock_irqrestore(&(dev->lock),f);
+ ERROR("palacios: ignoring request as user side is not connected for host device \"%s\"\n",dev->url);
return 0;
}
- // resize request and response in case they will need it
- palacios_resize_reqresp(&(dev->req),len,0); // make room for data
+ // resize request
+ if (!palacios_bigenough_reqresp(dev->req,len)) {
+ // not enough room.
+ // we drop the lock, turn on interrupts, resize, and then retry
+ DEEP_DEBUG_PRINT("palacios: request not big enough, dropping lock to resize on device \"%s\"\n",dev->url);
+
+ spin_unlock_irqrestore(&(dev->lock),f);
+
+ if (palacios_resize_reqresp(&(dev->req),len,0)) {
+ ERROR("palacios: cannot resize for request on device \"%s\"\n",dev->url);
+ return 0;
+ } else {
+ // reacquire the lock
+ // There shouldn't be a race here since there should not be another
+ // request from palacios until this one finishes
+ spin_lock_irqsave(&(dev->lock),f);
+ DEEP_DEBUG_PRINT("palacios: reacquired lock on device \"%s\"\n",dev->url);
+ }
+ }
dev->req->type=PALACIOS_HOST_DEV_HOST_REQUEST_WRITE_CONF;
dev->req->port=0;
spin_unlock_irqrestore(&(dev->lock),f);
return op_len;
- }
+}
+static int palacios_host_dev_ack_irq(v3_host_dev_t hostdev, uint8_t irq)
+{
+ // we don't care
+ return 0;
+}
+
-int palacios_init_host_dev( void ) {
+int palacios_init_host_dev( void )
+{
V3_Init_Host_Device_Support(&palacios_host_dev_hooks);
return 0;
virtio->virtio_cfg.pci_isr = 0;
- virtio->mergeable_rx_bufs = 0;
+ virtio->mergeable_rx_bufs = 1;
virtio->virtio_cfg.host_features = 0 | (1 << VIRTIO_NET_F_MAC);
if(virtio->mergeable_rx_bufs) {
}
+/*copy data into ring buffer */
static inline int copy_data_to_desc(struct guest_info * core,
- struct virtio_net_state * virtio_state,
- struct vring_desc * desc,
- uchar_t * buf,
- uint_t buf_len,
- uint_t dst_offset){
+ struct virtio_net_state * virtio_state,
+ struct vring_desc * desc,
+ uchar_t * buf,
+ uint_t buf_len,
+ uint_t dst_offset){
uint32_t len;
uint8_t * desc_buf = NULL;
PrintDebug("Could not translate buffer address\n");
return -1;
}
- len = (desc->length < buf_len)?(desc->length - dst_offset):buf_len;
- memcpy(desc_buf+dst_offset, buf, len);
+ len = (desc->length < (buf_len+dst_offset))?(desc->length - dst_offset):buf_len;
+ memcpy(desc_buf + dst_offset, buf, len);
return len;
}
uint16_t desc_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
int desc_cnt = get_desc_count(q, desc_idx);
- if(desc_cnt > 2){
+ if(desc_cnt != 2){
PrintError("VNIC: merged rx buffer not supported, desc_cnt %d\n", desc_cnt);
goto exit_error;
}
goto exit_error;
}
if(buf_desc->next & VIRTIO_NEXT_FLAG){
- V3_Net_Print(2, "Virtio NIC: TX more buffer need to read\n");
+ PrintError("Virtio NIC: TX more buffer need to read\n");
}
q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
v3_unlock_irqrestore(virtio_state->tx_lock, flags);
if (txed && !(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
- v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus, 0, virtio_state->pci_dev);
+ v3_pci_raise_irq(virtio_state->virtio_dev->pci_bus,
+ 0, virtio_state->pci_dev);
virtio_state->virtio_cfg.pci_isr = 0x1;
-
virtio_state->stats.rx_interrupts ++;
}
struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
int port_idx = port % virtio->io_range_size;
- PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n", private_data,
- port, port_idx, length, *(uint32_t *)src);
+ PrintDebug("VIRTIO NIC %p Write for port %d (index=%d) len=%d, value=%x\n",
+ private_data, port, port_idx,
+ length, *(uint32_t *)src);
switch (port_idx) {
case GUEST_FEATURES_PORT:
uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
switch (queue_idx) {
case 0:
- virtio_setup_queue(core, virtio, &virtio->rx_vq, pfn, page_addr);
+ virtio_setup_queue(core, virtio,
+ &virtio->rx_vq,
+ pfn, page_addr);
break;
case 1:
- virtio_setup_queue(core, virtio, &virtio->tx_vq, pfn, page_addr);
+ virtio_setup_queue(core, virtio,
+ &virtio->tx_vq,
+ pfn, page_addr);
if(virtio->tx_notify == 0){
disable_cb(&virtio->tx_vq);
vnet_thread_wakeup(virtio->poll_thread);
}
break;
case 2:
- virtio_setup_queue(core, virtio, &virtio->ctrl_vq, pfn, page_addr);
+ virtio_setup_queue(core, virtio,
+ &virtio->ctrl_vq,
+ pfn, page_addr);
break;
default:
break;
int port_idx = port % virtio->io_range_size;
uint16_t queue_idx = virtio->virtio_cfg.vring_queue_selector;
- PrintDebug("Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n", private_data,
- port, port_idx, length);
+ PrintDebug("Virtio NIC %p: Read for port 0x%x (index =%d), length=%d\n",
+ private_data, port, port_idx, length);
switch (port_idx) {
case HOST_FEATURES_PORT:
case VIRTIO_ISR_PORT:
*(uint8_t *)dst = virtio->virtio_cfg.pci_isr;
virtio->virtio_cfg.pci_isr = 0;
- v3_pci_lower_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
+ v3_pci_lower_irq(virtio->virtio_dev->pci_bus,
+ 0, virtio->pci_dev);
break;
case VIRTIO_NET_CONFIG ... VIRTIO_NET_CONFIG + ETH_ALEN:
break;
default:
- PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n", port_idx);
+ PrintError("Virtio NIC: Read of Unhandled Virtio Read:%d\n",
+ port_idx);
return -1;
}
struct virtio_net_state * virtio = (struct virtio_net_state *)private_data;
struct virtio_queue * q = &(virtio->rx_vq);
struct virtio_net_hdr_mrg_rxbuf hdr;
- uint32_t data_len;
unsigned long flags;
+ uint8_t kick_guest = 0;
V3_Net_Print(2, "Virtio-NIC: virtio_rx: size: %d\n", size);
- if(net_debug >= 4){
- v3_hexdump(buf, size, NULL, 0);
- }
-
- flags = v3_lock_irqsave(virtio->rx_lock);
-
- data_len = size;
- memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
- if (q->ring_avail_addr == 0) {
+ if (!q->ring_avail_addr) {
V3_Net_Print(2, "Virtio NIC: RX Queue not set\n");
virtio->stats.rx_dropped ++;
- goto err_exit;
+
+ return -1;
}
+ memset(&hdr, 0, sizeof(struct virtio_net_hdr_mrg_rxbuf));
+
+ flags = v3_lock_irqsave(virtio->rx_lock);
+
if (q->cur_avail_idx != q->avail->index){
- addr_t hdr_addr = 0;
- uint16_t buf_idx = 0;
- uint16_t hdr_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
- struct vring_desc * hdr_desc = NULL;
- struct vring_desc * buf_desc = NULL;
- uint32_t hdr_len = 0;
- uint32_t len;
-
- hdr_desc = &(q->desc[hdr_idx]);
- if (v3_gpa_to_hva(&(virtio->virtio_dev->vm->cores[0]), hdr_desc->addr_gpa, &(hdr_addr)) == -1) {
- V3_Net_Print(2, "Virtio NIC: Could not translate receive buffer address\n");
- virtio->stats.rx_dropped ++;
- goto err_exit;
- }
+ uint16_t buf_idx;
+ struct vring_desc * buf_desc;
+ uint32_t hdr_len, len;
+ uint32_t offset = 0;
- hdr_len = sizeof(struct virtio_net_hdr_mrg_rxbuf);
+ hdr_len = (virtio->mergeable_rx_bufs)?
+ sizeof(struct virtio_net_hdr_mrg_rxbuf):
+ sizeof(struct virtio_net_hdr);
if(virtio->mergeable_rx_bufs){/* merged buffer */
- uint32_t offset = 0;
- len = 0;
- hdr.num_buffers = 0;
+ struct vring_desc * hdr_desc;
+ uint16_t old_idx = q->cur_avail_idx;
+ buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
hdr_desc = &(q->desc[buf_idx]);
- hdr_desc->flags &= ~VIRTIO_NEXT_FLAG;
- len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, buf, data_len, hdr_len);
+ len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
+ virtio, hdr_desc, buf, size, hdr_len);
+ if(len < 0){
+ goto err_exit;
+ }
offset += len;
- hdr.num_buffers ++;
q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
- q->used->ring[q->used->index % q->queue_size].length = hdr_len + len;
+ q->used->ring[q->used->index % q->queue_size].length = hdr_len + offset;
q->cur_avail_idx ++;
+ hdr.num_buffers ++;
- while(offset < data_len) {
+ while(offset < size) {
buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
buf_desc = &(q->desc[buf_idx]);
- len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf + offset, data_len - offset, 0);
- if (len <= 0){
- V3_Net_Print(2, "Virtio NIC:merged buffer, %d buffer size %d\n", hdr.num_buffers, data_len);
- virtio->stats.rx_dropped ++;
+ len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
+ virtio, buf_desc, buf+offset, size-offset, 0);
+ if (len < 0){
+ V3_Net_Print(2, "Virtio NIC:merged buffer, %d buffer size %d\n",
+ hdr.num_buffers, len);
+ q->cur_avail_idx = old_idx;
goto err_exit;
}
offset += len;
buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
- hdr.num_buffers ++;
q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
q->used->ring[(q->used->index + hdr.num_buffers) % q->queue_size].length = len;
q->cur_avail_idx ++;
+
+ hdr.num_buffers ++;
}
+
+ copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
+ virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0);
q->used->index += hdr.num_buffers;
- copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0);
}else{
- hdr_desc = &(q->desc[buf_idx]);
- copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, hdr_desc, (uchar_t *)&hdr, hdr_len, 0);
-
- buf_idx = hdr_desc->next;
+ buf_idx = q->avail->ring[q->cur_avail_idx % q->queue_size];
buf_desc = &(q->desc[buf_idx]);
- len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]), virtio, buf_desc, buf, data_len, 0);
- if (len < data_len) {
- V3_Net_Print(2, "Virtio NIC: ring buffer len less than pkt size, merged buffer not supported, buffer size %d\n", len);
- virtio->stats.rx_dropped ++;
-
- goto err_exit;
+
+ /* copy header */
+ len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
+ virtio, buf_desc, (uchar_t *)&(hdr.hdr), hdr_len, 0);
+ if(len < hdr_len){
+ V3_Net_Print(2, "Virtio NIC: rx copy header error %d, hdr_len %d\n",
+ len, hdr_len);
+ goto err_exit;
+ }
+
+ len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
+ virtio, buf_desc, buf, size, hdr_len);
+ if(len < 0){
+ V3_Net_Print(2, "Virtio NIC: rx copy data error %d\n", len);
+ goto err_exit;
+ }
+ offset += len;
+
+ /* copy rest of data */
+ while(offset < size &&
+ (buf_desc->flags & VIRTIO_NEXT_FLAG)){
+ buf_desc = &(q->desc[buf_desc->next]);
+ len = copy_data_to_desc(&(virtio->virtio_dev->vm->cores[0]),
+ virtio, buf_desc, buf+offset, size-offset, 0);
+ if (len < 0) {
+ break;
+ }
+ offset += len;
}
buf_desc->flags &= ~VIRTIO_NEXT_FLAG;
+
+ if(offset < size){
+ V3_Net_Print(2, "Virtio NIC: rx not enough ring buffer, buffer size %d\n",
+ len);
+ goto err_exit;
+ }
q->used->ring[q->used->index % q->queue_size].id = q->avail->ring[q->cur_avail_idx % q->queue_size];
- q->used->ring[q->used->index % q->queue_size].length = data_len + hdr_len; /* This should be the total length of data sent to guest (header+pkt_data) */
- q->used->index++;
- q->cur_avail_idx++;
+ q->used->ring[q->used->index % q->queue_size].length = size + hdr_len; /* This should be the total length of data sent to guest (header+pkt_data) */
+ q->used->index ++;
+ q->cur_avail_idx ++;
}
virtio->stats.rx_pkts ++;
virtio->stats.rx_bytes += size;
} else {
V3_Net_Print(2, "Virtio NIC: Guest RX queue is full\n");
- virtio->stats.rx_dropped ++;
+ virtio->stats.rx_dropped ++;
- /* kick guest to refill the queue */
- virtio->virtio_cfg.pci_isr = 0x1;
- v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
- v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].pcpu_id, 0);
- virtio->stats.rx_interrupts ++;
-
- goto err_exit;
+ /* kick guest to refill RX queue */
+ kick_guest = 1;
}
- if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG)) {
- V3_Net_Print(2, "Raising IRQ %d\n", virtio->pci_dev->config_header.intr_line);
+ v3_unlock_irqrestore(virtio->rx_lock, flags);
+
+ if (!(q->avail->flags & VIRTIO_NO_IRQ_FLAG) || kick_guest) {
+ V3_Net_Print(2, "Virtio NIC: RX Raising IRQ %d\n",
+ virtio->pci_dev->config_header.intr_line);
virtio->virtio_cfg.pci_isr = 0x1;
v3_pci_raise_irq(virtio->virtio_dev->pci_bus, 0, virtio->pci_dev);
-
virtio->stats.rx_interrupts ++;
}
- v3_unlock_irqrestore(virtio->rx_lock, flags);
-
/* notify guest if it is in guest mode */
- if(virtio->rx_notify == 1 &&
+ if((kick_guest || virtio->rx_notify == 1) &&
V3_Get_CPU() != virtio->virtio_dev->vm->cores[0].pcpu_id){
- v3_interrupt_cpu(virtio->virtio_dev->vm, virtio->virtio_dev->vm->cores[0].pcpu_id, 0);
+ v3_interrupt_cpu(virtio->virtio_dev->vm,
+ virtio->virtio_dev->vm->cores[0].pcpu_id,
+ 0);
}
return 0;
err_exit:
-
+ virtio->stats.rx_dropped ++;
v3_unlock_irqrestore(virtio->rx_lock, flags);
return -1;
static int virtio_tx_flush(void * args){
struct virtio_net_state *virtio = (struct virtio_net_state *)args;
- V3_Print("Virtio TX Poll Thread Starting for %s\n", virtio->vm->name);
+ V3_Print("Virtio TX Poll Thread Starting for %s\n",
+ virtio->vm->name);
while(1){
if(virtio->tx_notify == 0){
int tmp_ports = num_ports;
int i;
- // This gets the number of ports, rounded up to a power of 2
+ /* This gets the number of ports, rounded up to a power of 2 */
net_state->io_range_size = 1; // must be a power of 2
while (tmp_ports > 0) {
tmp_ports >>= 1;
}
/* this is to account for any low order bits being set in num_ports
- * if there are none, then num_ports was already a power of 2 so we shift right to reset it
- */
+ * if there are none, then num_ports was already a power of 2 so we shift right to reset it
+ */
if ((num_ports & ((net_state->io_range_size >> 1) - 1)) == 0) {
net_state->io_range_size >>= 1;
}
bars[i].type = PCI_BAR_NONE;
}
- PrintDebug("Virtio-NIC io_range_size = %d\n", net_state->io_range_size);
+ PrintDebug("Virtio-NIC io_range_size = %d\n",
+ net_state->io_range_size);
bars[0].type = PCI_BAR_IO;
bars[0].default_base_port = -1;
net_state->tx_notify = 0;
net_state->rx_notify = 0;
- net_state->timer = v3_add_timer(&(info->cores[0]),&timer_ops,net_state);
+ net_state->timer = v3_add_timer(&(info->cores[0]),
+ &timer_ops,net_state);
PrintError("net_state 0x%p\n", (void *)net_state);
ops->frontend_data = net_state;
memcpy(ops->fnt_mac, virtio->mac, ETH_ALEN);
- net_state->poll_thread = vnet_start_thread(virtio_tx_flush, (void *)net_state, "Virtio_Poll");
+ net_state->poll_thread = vnet_start_thread(virtio_tx_flush,
+ (void *)net_state, "Virtio_Poll");
net_state->status = 1;