Re: [PATCH v4 04/54] scsi: BusLogic: Pass the host pointer directly to several functions

[Bart Van Assche <bvanassche@acm.org>]

On 9/28/26 11:55 AM, Khalid Aziz wrote:
> I prefer not to pull in macro definitions from header files into the
> driver code. This opens up possibility for API inconsistencies if
> the macro in header file is ever updated.
The DEF_SCSI_QCMD() macro was introduced in 2010 and there have not
been any changes to its functionality in the 16 years that passed since
its introduction. Hence, I think it's unlikely that it will be changed
in the future.

> Why not extract struct *scsi_host in blogic_qcmd_lck() from
> struct *scsi_cmd instead and continue to use DEF_SCSI_QCMD()?

Because it is the only way to avoid __assume_ctx_lock() in
blogic_qcmd_lck(). __assume_ctx_lock() should be avoided if
possible because it defeats the purpose of compile-time thread-safety
analysis.

The BusLogic driver is one of the few drivers that releases and
reacquires shost->host_lock() from inside blogic_qcmd_lck(). Preventing
that the Clang compiler complains about the spin_unlock_irq() and
spin_lock_irq() calls in blogic_qcmd_lck() requires adding a
__must_hold(&shost->host_lock) annotation to the blogic_qcmd_lck()
function. And adding the __must_hold(&shost->host_lock) annotation
to the blogic_qcmd_lck() function is only possible if 'shost' is one of
the arguments of the blogic_qcmd_lck() function.

Thanks,

Bart.