Re: [PATCH] mlx4_core: Fix crash on uninitialized priv->cmd.slave_sem

[Or Gerlitz <ogerlitz-VPRAkNaXOzVWk0Htik3J/w@public.gmane.org>]

On 27/09/2012 08:46, Roland Dreier wrote:
> On Wed, Sep 26, 2012 at 2:51 PM, Or Gerlitz <or.gerlitz-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org> wrote:
>> What exactly did you mean by saying "for unlucky configurations"
>> above? what value did you use for mlx4_core's port_array_type module
>> param?
> I didn't set the parameter at all.  What I meant by "unlucky configs" was
> all the conditions in the following code in mlx4_dev_cap():
>
> 		/*
> 		 * Link sensing is allowed on the port if 3 conditions are true:
> 		 * 1. Both protocols are supported on the port.
> 		 * 2. Different types are supported on the port
> 		 * 3. FW declared that it supports link sensing
> 		 */
> 		mlx4_priv(dev)->sense.sense_allowed[i] =
> 			((dev->caps.supported_type[i] == MLX4_PORT_TYPE_AUTO) &&
> 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_DPDP) &&
> 			 (dev->caps.flags & MLX4_DEV_CAP_FLAG_SENSE_SUPPORT));
>
> 		/*
> 		 * If "default_sense" bit is set, we move the port to "AUTO" mode
> 		 * and perform sense_port FW command to try and set the correct
> 		 * port type from beginning
> 		 */
> 		if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i]) {
> 			enum mlx4_port_type sensed_port = MLX4_PORT_TYPE_NONE;
> 			dev->caps.possible_type[i] = MLX4_PORT_TYPE_AUTO;
> 			mlx4_SENSE_PORT(dev, i, &sensed_port);
> 			if (sensed_port != MLX4_PORT_TYPE_NONE)
> 				dev->caps.port_type[i] = sensed_port;
> 		} else {

Roland,

Looking deeper on the issues you reported, we noted that the bug is up 
here on the RHS of

if (mlx4_priv(dev)->sense.sense_allowed[i] && dev->caps.default_sense[i])

as when a firmware sets the “do_sense” bit as 1 (which is the default in 
2.11.0500), the driver begins to do sensing. This firmware somehow went 
out without enough mfunc checking, bad. A fast solution would be a one 
liner patch that forces the LHS of this line to be false under mfunc 
mode, that is add forth condition to the three conditions above which is 
&& !mfunc

In the overall driver architecture level, auto-sending can/will be abled 
in later state, when the PPF
can signal the VFs to reset/re-create their mlx4 device when the link 
type change, as done in native
mode.

Or.



> I'm beginning to think it might be that I had nothing hooked up
> to my adapter initially, which seems to lead to the fatal SENSE_PORT
> call here.  I notice that once I bring up the link, the stream of failing
> SENSE_PORT calls (which don't ever work in SR-IOV mode AFAICT) stops as well.
>
>

--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html