Can the SoftIRQ prempt currently running the same SoftIRQ with Linux kernel?

I am working on a performance improvement of a driver and should consider the possibility of deadlock. In a SoftIRQ context, spin_lock will be held and protect some variable. In this case, should I use spin_lock or spin_lock_bh? spin_lock_bh sounds like safer, but I have a general question about SoftIRQ implementation.

• the same softIRQ function can be scheduled on the same CPU during the execution of the SoftIRQ -> we need to use spin_lock_bh in the SoftIRQ
• the same softIRQ function can be scheduled on the other CPU, but not the same CPU -> we may use spin_lock in the SoftIRQ

What is true from the above statement?

From Linux Kernel Development (2nd edition):

A softirq never preempts another softirq.

The reason is simple: during softIRQ software interrupts are disabled.

So it is sufficient to use plain spin_lock() in a softIRQ function.