Separate book?

Alexander E. Patrakov patrakov at
Mon May 29 21:09:10 PDT 2006

Gerard Beekmans wrote:
>  > The problem is not "how" (which is already dealt with), but "why". I.e.,
>> one could say that some sysfs attributes for SCSI devices are created 
>> after the uevent is sent. That's why we have the WAIT_FOR_SYSFS rule 
>> that waits for "ioerr_cnt".
>> But it will be certainly better to show where exactly this takes place 
>> (i.e., the scsi_sysfs_add_sdev() function from 
>> drivers/scsi/scsi_sysfs.c). This function, in its very end, creates 
>> all sysfs attributes listed in the scsi_sysfs_sdev_attrs[] array. Upon 
>> examining the contents of that array, it is clearly seen that the 
>> "ioerr_cnt" is really the last attribute created.
> The "why" can be taken too far.

Indeed, it is not completely suitable for the LFS book (that's why a separate 
book has been proposed by Ag). My point is:

1) LFS should provide a working "how"
2) LFS editors should know enough of "why" not to break the "how" in the future 
(e.g., should have enough information to judge when workarounds such as 
WAIT_FOR_SYSFS rules become unneeded)

> Is understanding the 
> scsi_sysfs_add_sdev() function really of interest to most people? I 
> don't think most of us care to know the programming internals of the 
> Linux kernel to that degree. I wager to say that most people just like 
> to know when/why devices are created, but it sounds like this may be 
> going too in depth.

Correct for readers, but not for editors.

Alexander E. Patrakov

More information about the lfs-dev mailing list