EPROBE_DEFER and how it is supposed to work

[Real Name]

On Tue, Sep 09, 2014 at 06:26:27PM +0000, Rajat Jain wrote:
> Hi,
> 
> I'm not a newbie, but I am trying to understand the semantics of deferred probing.
> 
> My question is generic, but for an example:
> 
> Let's say I have a platform driver "A" for a device "a", that requires device "b" (controlled by driver "B") to be operational first. Both A &B can be built as part of kernel, or as modules independently.  As far as I could tell, there is no way to specify the ORDER in which the drivers' probe routines should be called. I took a look at modprobe / depmod, but it seems to be of relevance for modules only that are loaded from user space. My questions:
> 
> 1) Is there a way to specify that "Kernel should call A's probe routine only after B's probe routine"?

You can ask module A to load the required module B.

For example, lib/textsearch.c:textsearch_prepare -> lookup_ts_algo

lookup_ts_algo will load ts_bm.ko/ts_fsm.ko/ts_kmp.ko as required.

> 
> 2) How I currently do this: In A's probe routine, I currently check if the device "b" is available. If not, I return -EPROBE_DEFER. This mostly works ok. However, I'm curious about the case where the driver B is not built as part of the kernel but A is. After getting an _EPROB_DEFER from a B's probe routine, when will the kernel try again? And for how long would it keep on trying? I verified that if I load the driver B manually from user space, the A's probe routine does not get called (if it gets called now - it would have returned success!).
> 
> Thanks,
> 
> Rajat
> 
> _______________________________________________
> Kernelnewbies mailing list
> Kernelnewbies at kernelnewbies.org
> http://lists.kernelnewbies.org/mailman/listinfo/kernelnewbies