Re: [PATCH 13/62] firewire: convert to idr_alloc()

[Stefan Richter <stefanr@s5r6.in-berlin.de>]

On Feb 02 Tejun Heo wrote:
> Convert to the much saner new idr interface.
> 
> Only compile tested.
> 
> Signed-off-by: Tejun Heo <tj@kernel.org>
> Cc: Stefan Richter <stefanr@s5r6.in-berlin.de>
> Cc: linux1394-devel@lists.sourceforge.net
> ---
> This patch depends on an earlier idr changes and I think it would be
> best to route these together through -mm.  Please holler if there's
> any objection.  Thanks.
> 
>  drivers/firewire/core-cdev.c   | 16 +++++++---------
>  drivers/firewire/core-device.c |  5 ++---
>  2 files changed, 9 insertions(+), 12 deletions(-)
> 
> diff --git a/drivers/firewire/core-cdev.c b/drivers/firewire/core-cdev.c
> index 68c3138..46a1f88 100644
> --- a/drivers/firewire/core-cdev.c
> +++ b/drivers/firewire/core-cdev.c
> @@ -490,24 +490,22 @@ static int add_client_resource(struct client *client,
>  	unsigned long flags;
>  	int ret;
>  
> - retry:
> -	if (idr_pre_get(&client->resource_idr, gfp_mask) == 0)
> -		return -ENOMEM;
> -
> +	idr_preload(gfp_mask);
>  	spin_lock_irqsave(&client->lock, flags);
> +
>  	if (client->in_shutdown)
>  		ret = -ECANCELED;
>  	else
> -		ret = idr_get_new(&client->resource_idr, resource,
> -				  &resource->handle);
> +		ret = idr_alloc(&client->resource_idr, resource, 0, 0,
> +				GFP_NOWAIT);
>  	if (ret >= 0) {
> +		resource->handle = ret;
>  		client_get(client);
>  		schedule_if_iso_resource(resource);
>  	}
> -	spin_unlock_irqrestore(&client->lock, flags);
>  
> -	if (ret == -EAGAIN)
> -		goto retry;
> +	spin_unlock_irqrestore(&client->lock, flags);
> +	idr_preload_end();
>  
>  	return ret < 0 ? ret : 0;
>  }

Most of the add_client_resource() callers are process context (user
process doing an ioctl()), and for them this is going to work fine.  But
one caller is tasklet context (firewire-ohci's asynchronous reception DMA
IRQ handler "bottom half").

The kerneldoc comment in 05/62 "idr: implement idr_preload[_end]() and
idr_alloc()" tells me that idr_preload and idr_preload_end must be used in
process context only.  Can it work in tasklet context regardless?  I am
not entirely sure from looking at the implementation.

> diff --git a/drivers/firewire/core-device.c b/drivers/firewire/core-device.c
> index 3873d53..c137de6 100644
> --- a/drivers/firewire/core-device.c
> +++ b/drivers/firewire/core-device.c
> @@ -1017,13 +1017,12 @@ static void fw_device_init(struct work_struct *work)
>  
>  	fw_device_get(device);
>  	down_write(&fw_device_rwsem);
> -	ret = idr_pre_get(&fw_device_idr, GFP_KERNEL) ?
> -	      idr_get_new(&fw_device_idr, device, &minor) :
> -	      -ENOMEM;
> +	ret = idr_alloc(&fw_device_idr, device, 0, 0, GFP_KERNEL);
>  	up_write(&fw_device_rwsem);
>  
>  	if (ret < 0)
>  		goto error;
> +	minor = ret;
>  
>  	device->device.bus = &fw_bus_type;
>  	device->device.type = &fw_device_type;

This hunk is OK, of course.
-- 
Stefan Richter
-=====-===-= --=- ---==
http://arcgraph.de/sr/