Are we still willing to pay the cost of moving include/linux/irqdesc.h into kernel/irq/?
Hi Thomas,
When refining irq related data structures, I found once we had
a plan to move include/linux/irqdesc.h into kernel/irq/, and I have basic ideas about how to achieve the goal as:
1) de-inline accessor functions in irqdesc.h and move the declaration into linux/irq.h.
2) remove reference to irq_desc->xxxx from drivers one by one (this step is simple)
3) remove reference to irq_desc->xxxx from arch (this step is hard)
4) move irqdesc.h into kernel/irq/
But are we still willing to pay the cost of converting inline access helpers into normal functions?
One possible tradeoff is to remove reference of irq_desc->xxxx from drivers/, which is straight forward and shouldn't cause serious performance regressions:)
Thanks!
Gerry
Jiang,
Quoted 3 lines · Jiang Liu
Hi Thomas,
When refining irq related data structures, I found once we had
a plan to moveinclude/linux/irqdesc.hintokernel/irq/, and I have
I had that plan for two reasons:
1) Is was tired of chasing the abuse of irq_desc
2) We wanted to move irq descriptor memory allocation when affinity changes to a different node.
Quoted 5 lines
basic ideas about how to achieve the goal as:
1) de-inline accessor functions inirqdesc.hand move the declaration intolinux/irq.h.
2) remove reference toirq_desc->xxxxfrom drivers one by one (this step is simple)
Not sure. drivers/irqchip pretty much needs it and we have other demeltiplex handlers which need it as well.
Quoted 1 line
3) remove reference to
irq_desc->xxxxfrom arch (this step is hard)
Same issue as #2
Quoted 4 lines
4) move
irqdesc.hintokernel/irq/But are we still willing to pay the cost of converting inline access helpers into normal functions?
I dont think so.
Thanks,
tglx