[PATCH 3/3] staging: vme: make match() driver specific to improve non-VME64x support
Emilio G. Cota
cota at braap.org
Wed Aug 31 21:20:47 UTC 2011
This was hard to review. There are references to functions that
are not committed in Greg's tree yet ("staging" tree @ git.kernel.org).
I assume this patch was applied before you wrote the v4 patchset:
https://lkml.org/lkml/2011/8/12/107
On Wed, Aug 31, 2011 at 12:05:46 +0200, Manohar Vanga wrote:
(snip)
> Another change introduced in this patch is that devices are now created
> within the VME driver structure rather than in the VME bridge structure.
> This way, things don't go haywire if the bridge driver is removed while
> a driver is using it (this is also additionally prevented by having
> reference counting of used bridge modules).
The mention to refcounting seems outdated. As I stated in my reply
to v0, we should just safely remove devices under the bus when
vme_unregister_bus() is called.
> ---
> drivers/staging/vme/devices/vme_user.c | 53 +++-----
> drivers/staging/vme/devices/vme_user.h | 2 +-
> drivers/staging/vme/vme.c | 201 +++++++++++++++-----------------
> drivers/staging/vme/vme.h | 22 +++-
> drivers/staging/vme/vme_api.txt | 60 +++++-----
> drivers/staging/vme/vme_bridge.h | 4 -
> 6 files changed, 163 insertions(+), 179 deletions(-)
>
> diff --git a/drivers/staging/vme/devices/vme_user.c b/drivers/staging/vme/devices/vme_user.c
> index bb33dc2..f85be2c 100644
> --- a/drivers/staging/vme/devices/vme_user.c
> +++ b/drivers/staging/vme/devices/vme_user.c
> @@ -43,7 +43,7 @@
> static DEFINE_MUTEX(vme_user_mutex);
> static const char driver_name[] = "vme_user";
>
> -static int bus[USER_BUS_MAX];
> +static int bus[VME_USER_BUS_MAX];
> static unsigned int bus_num;
>
> /* Currently Documentation/devices.txt defines the following for VME:
> @@ -135,6 +135,7 @@ static ssize_t vme_user_write(struct file *, const char __user *, size_t,
> static loff_t vme_user_llseek(struct file *, loff_t, int);
> static long vme_user_unlocked_ioctl(struct file *, unsigned int, unsigned long);
>
> +static int vme_user_match(struct vme_dev *);
> static int __devinit vme_user_probe(struct vme_dev *);
> static int __devexit vme_user_remove(struct vme_dev *);
>
> @@ -620,6 +621,7 @@ static void buf_unalloc(int num)
>
> static struct vme_driver vme_user_driver = {
> .name = driver_name,
> + .match = vme_user_match,
> .probe = vme_user_probe,
> .remove = __devexit_p(vme_user_remove),
> };
> @@ -628,8 +630,6 @@ static struct vme_driver vme_user_driver = {
> static int __init vme_user_init(void)
> {
> int retval = 0;
> - int i;
> - struct vme_device_id *ids;
>
> printk(KERN_INFO "VME User Space Access Driver\n");
>
> @@ -643,47 +643,36 @@ static int __init vme_user_init(void)
> /* Let's start by supporting one bus, we can support more than one
> * in future revisions if that ever becomes necessary.
> */
> - if (bus_num > USER_BUS_MAX) {
> + if (bus_num > VME_USER_BUS_MAX) {
> printk(KERN_ERR "%s: Driver only able to handle %d buses\n",
> - driver_name, USER_BUS_MAX);
> - bus_num = USER_BUS_MAX;
> - }
> -
> -
> - /* Dynamically create the bind table based on module parameters */
> - ids = kzalloc(sizeof(struct vme_device_id) * (bus_num + 1), GFP_KERNEL);
> - if (ids == NULL) {
> - printk(KERN_ERR "%s: Unable to allocate ID table\n",
> - driver_name);
> - retval = -ENOMEM;
> - goto err_id;
> - }
> -
> - for (i = 0; i < bus_num; i++) {
> - ids[i].bus = bus[i];
> - /*
> - * We register the driver against the slot occupied by *this*
> - * card, since it's really a low level way of controlling
> - * the VME bridge
> - */
> - ids[i].slot = VME_SLOT_CURRENT;
> + driver_name, VME_USER_BUS_MAX);
> + bus_num = VME_USER_BUS_MAX;
> }
>
> - vme_user_driver.bind_table = ids;
> -
> - retval = vme_register_driver(&vme_user_driver);
> + /*
> + * Here we just register the maximum number of devices we can and
> + * leave vme_user_match() to allow only 1 to go through to probe().
> + * This way, if we later want to allow multiple user access devices,
> + * we just change the code in vme_user_match().
> + */
> + retval = vme_register_driver(&vme_user_driver, VME_MAX_SLOTS);
> if (retval != 0)
> goto err_reg;
>
> return retval;
>
> err_reg:
> - kfree(ids);
> -err_id:
> err_nocard:
> return retval;
> }
>
> +static int vme_user_match(struct vme_dev *vdev)
> +{
> + if (vdev->id.num >= VME_USER_BUS_MAX)
> + return 0;
> + return 1;
> +}
> +
> /*
> * In this simple access driver, the old behaviour is being preserved as much
> * as practical. We will therefore reserve the buffers and request the images
> @@ -896,8 +885,6 @@ static int __devexit vme_user_remove(struct vme_dev *dev)
> static void __exit vme_user_exit(void)
> {
> vme_unregister_driver(&vme_user_driver);
> -
> - kfree(vme_user_driver.bind_table);
> }
>
>
> diff --git a/drivers/staging/vme/devices/vme_user.h b/drivers/staging/vme/devices/vme_user.h
> index 24bf4e5..d85a1e9 100644
> --- a/drivers/staging/vme/devices/vme_user.h
> +++ b/drivers/staging/vme/devices/vme_user.h
> @@ -1,7 +1,7 @@
> #ifndef _VME_USER_H_
> #define _VME_USER_H_
>
> -#define USER_BUS_MAX 1
> +#define VME_USER_BUS_MAX 1
>
> /*
> * VMEbus Master Window Configuration Structure
> diff --git a/drivers/staging/vme/vme.c b/drivers/staging/vme/vme.c
> index 76e08f3..2c3d47d 100644
> --- a/drivers/staging/vme/vme.c
> +++ b/drivers/staging/vme/vme.c
> @@ -42,10 +42,7 @@ static DEFINE_MUTEX(vme_buses_lock);
> static void __exit vme_exit(void);
> static int __init vme_init(void);
>
> -static struct vme_dev *dev_to_vme_dev(struct device *dev)
> -{
> - return container_of(dev, struct vme_dev, dev);
> -}
> +#define dev_to_vme_dev(dev) container_of(dev, struct vme_dev, dev)
The function was added in patch 2. In general, you should not add code
that then gets removed, unless there's a good reason. In this case just
add either the function or the macro in patch 2, and stick to it.
>
> /*
> * Find the bridge that the resource is associated with.
> @@ -1342,153 +1339,148 @@ static void vme_dev_release(struct device *dev)
>
> int vme_register_bridge(struct vme_bridge *bridge)
> {
> - struct vme_dev *vdev;
> - int retval;
> - int i;
> + return vme_add_bus(bridge);
> +}
> +EXPORT_SYMBOL(vme_register_bridge);
>
> - retval = vme_add_bus(bridge);
> - if (retval)
> - return retval;
> +void vme_unregister_bridge(struct vme_bridge *bridge)
> +{
> + vme_remove_bus(bridge);
> +}
> +EXPORT_SYMBOL(vme_unregister_bridge);
vme_remove_bus is in the separate patch:
https://lkml.org/lkml/2011/8/12/107
+static void vme_remove_bus(struct vme_bridge *bridge)
+{
+ mutex_lock(&vme_buses_lock);
+ vme_bus_numbers &= ~(1 << bridge->num);
+ list_del(&bridge->bus_list);
+ mutex_unlock(&vme_buses_lock);
+}
It's quite unfortunate that we cannot easily see the old vs the
new vme_unregister_bridge in this patch. I paste here for clarity:
> -void vme_unregister_bridge(struct vme_bridge *bridge)
> {
> - int i;
> - struct vme_dev *vdev;
> -
> -
> - for (i = 0; i < VME_SLOTS_MAX; i++) {
> - vdev = bridge->dev[i];
> - device_unregister(&vdev->dev);
> - }
> vme_remove_bus(bridge);
> }
So we're essentially leaving the devices there, even though the
bridge they're under will be removed. This doesn't seem right.
btw with the removal of the array of vme_dev's from struct vme_bridge,
the bridge cannot know which devices are under it.
We have to bear in mind that the drv->devices list needs to be
updated when devices come and go; possibly a bridge->devices list
could also be kept.
Helpers around device_register and _unregister may simplify the lists'
housekeeping.
>
> - /* This creates 32 vme "slot" devices. This equates to a slot for each
> - * ID available in a system conforming to the ANSI/VITA 1-1994
> - * specification.
> - */
> - for (i = 0; i < VME_SLOTS_MAX; i++) {
> - bridge->dev[i] = kzalloc(sizeof(struct vme_dev), GFP_KERNEL);
> - if (!bridge->dev[i]) {
> - retval = -ENOMEM;
> - goto err_devalloc;
> - }
> - vdev = bridge->dev[i];
> - memset(vdev, 0, sizeof(struct vme_dev));
> +/* - Driver Registration --------------------------------------------------- */
>
> +static int __vme_register_driver_bus(struct vme_driver *drv,
> + struct vme_bridge *bridge, unsigned int ndevs)
> +{
> + int err;
> + unsigned int i;
> + struct vme_dev *vdev;
> + struct vme_dev *tmp;
> +
> + for (i = 0; i < ndevs; i++) {
> + vdev = kzalloc(sizeof(struct vme_dev), GFP_KERNEL);
> + if (!vdev) {
> + err = -ENOMEM;
> + goto err_alloc;
> + }
> + vdev->id.num = i;
> vdev->id.bus = bridge->num;
> vdev->id.slot = i + 1;
> vdev->bridge = bridge;
> + vdev->dev.platform_data = drv;
> + vdev->dev.release = vme_dev_release;
> vdev->dev.parent = bridge->parent;
> vdev->dev.bus = &vme_bus_type;
> - vdev->dev.release = vme_dev_release;
> - /*
> - * We save a pointer to the bridge in platform_data so that we
> - * can get to it later. We keep driver_data for use by the
> - * driver that binds against the slot
> - */
> - vdev->dev.platform_data = bridge;
> - dev_set_name(&vdev->dev, "vme-%x.%x", bridge->num, i + 1);
> + dev_set_name(&vdev->dev, "%s.%u-%u", drv->name, vdev->id.bus,
> + vdev->id.num);
>
> - retval = device_register(&vdev->dev);
> - if (retval)
> - goto err_reg;
> - }
> + err = device_register(&vdev->dev);
> + if (err)
> + goto err_dev_reg;
>
> - return retval;
> + if (vdev->dev.platform_data) {
> + list_add_tail(&vdev->list, &drv->devices);
> + drv->ndev++;
Ok, so drv->ndev can only increase. In case a device is removed (when
a bus driver is removed) this may need to be decreased, which isn't
done in the corresponding list_del() calls (I've marked them).
In fact I wonder whether it is useful at all to have drv->ndev. What's
its purpose?
> + } else
> + device_unregister(&vdev->dev);
> + }
> + return 0;
>
> -err_reg:
> +err_dev_reg:
> kfree(vdev);
> -err_devalloc:
> - while (--i >= 0) {
> - vdev = bridge->dev[i];
> +err_alloc:
> + list_for_each_entry_safe(vdev, tmp, &drv->devices, list) {
> + list_del(&vdev->list);
missing drv->ndev-- ?
> device_unregister(&vdev->dev);
> }
> - vme_remove_bus(bridge);
> - return retval;
> + return err;
> }
> -EXPORT_SYMBOL(vme_register_bridge);
>
> -void vme_unregister_bridge(struct vme_bridge *bridge)
> +static int __vme_register_driver(struct vme_driver *drv, unsigned int ndevs)
> {
> - int i;
> - struct vme_dev *vdev;
> -
> + struct vme_bridge *bridge;
> + int err = 0;
>
> - for (i = 0; i < VME_SLOTS_MAX; i++) {
> - vdev = bridge->dev[i];
> - device_unregister(&vdev->dev);
> + mutex_lock(&vme_buses_lock);
> + list_for_each_entry(bridge, &vme_bus_list, bus_list) {
> + /*
> + * This cannot cause trouble as we already have vme_buses_lock
> + * and if the bridge is removed, it will have to go through
> + * vme_unregister_bridge() to do it (which calls remove() on
> + * the bridge which in turn tries to acquire vme_buses_lock and
> + * will have to wait). The probe() called after device
> + * registration in __vme_register_driver below will also fail
> + * as the bridge is being removed (since the probe() calls
> + * vme_bridge_get()).
> + */
> + err = __vme_register_driver_bus(drv, bridge, ndevs);
> + if (err)
> + break;
> }
> - vme_remove_bus(bridge);
> + mutex_unlock(&vme_buses_lock);
> + return err;
> }
> -EXPORT_SYMBOL(vme_unregister_bridge);
> -
>
> -/* - Driver Registration --------------------------------------------------- */
> -
> -int vme_register_driver(struct vme_driver *drv)
> +int vme_register_driver(struct vme_driver *drv, unsigned int ndevs)
> {
> + int err;
> +
> drv->driver.name = drv->name;
> drv->driver.bus = &vme_bus_type;
> + INIT_LIST_HEAD(&drv->devices);
> +
> + err = driver_register(&drv->driver);
> + if (err)
> + return err;
> +
> + err = __vme_register_driver(drv, ndevs);
> + if (err)
> + vme_unregister_driver(drv);
>
> - return driver_register(&drv->driver);
> + return err;
> }
> EXPORT_SYMBOL(vme_register_driver);
>
> void vme_unregister_driver(struct vme_driver *drv)
> {
> + struct vme_dev *dev, *dev_tmp;
> +
> + list_for_each_entry_safe(dev, dev_tmp, &drv->devices, list) {
> + device_unregister(&dev->dev);
> + list_del(&dev->list);
missing drv->ndev-- ?
Emilio
More information about the devel
mailing list