Re: [PATCH] soc: renesas: mfis: add hwspinlock support

[Wolfram Sang <wsa+renesas@sang-engineering.com>]

[-- Attachment #1: Type: text/plain, Size: 972 bytes --]


> [Severity: High]
> Will this sleep in atomic context on preempt-rt kernels?
> 
> The hwspinlock core can invoke unlock callbacks from raw atomic contexts
> (for example, via hwspin_unlock_in_atomic). However, this callback invokes
> mfis_write(), which acquires the unprotect_lock:
> 
> mfis_write()
>     spin_lock_irqsave(&priv->unprotect_lock, flags);
> 
> On preempt-rt builds, a spinlock_t becomes a preemptible, sleepable rt_mutex.
> Acquiring it inside an atomic context could lead to a sleeping in atomic
> regression. On standard kernels, it also violates the lockdep wait-type
> hierarchy since we are nesting a spinlock inside a raw atomic context.
> 
> Should unprotect_lock be converted to a raw_spinlock_t to avoid this
> regression?

The spinlock was a best effort thing at the time. It should be a
hwspinlock even to protect against accesses from non-AP cores. I thought
this could be postponed to later, but it seems it cannot.


[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]