Shrikanth Hegde <[email protected]> writes: > On PPC32, the code for hard_irq_disable has diverged and it doesn't > do lockdep/tracing anymore, leading to use of both local_irq_disable > and hard_irq_disable on PPC32 to overcome those. > > That is un-necessary, instead simply calling local_irq_disable is > enough. It disables EE bit and does lockdep tracking. > > Subsequent patch uses this API in KVM code to avoid the warnings. > Wider use of it is possible, but it will be taken subsequently. > > Fixes: 5d38902c4838 ("powerpc: Add irqtrace support for 32-bit powerpc") > Suggested-by: Ritesh Harjani (IBM) <[email protected]> > Signed-off-by: Shrikanth Hegde <[email protected]> > --- > PS: Fixes tag is based on the historical analysis done by Ritesh. > I haven't got a chance to delve deep into it. Note that it is old > commit and lot of code interleave would have happened since > including local_irq_disable at some would have added proper > lockdep support. >
Yes, not sure whether we really want this this fix to go back to that old change? Maybe we can add CC: stable and keep the minimum kernel version to it, so that it gets applied >= vX.Y+ sort of. Instead of stable backports picking this change and taking it all the way to a very a old kernel release - which in my opinion is not worth it. Thoughts? Do we know from when we started hitting this problem? Should this be after we merge generic entry/exit? Or was there any previous kernel version too, where we were hitting this? > arch/powerpc/include/asm/hw_irq.h | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > BTW, thanks for taking it forward. The change looks good to me. So, please Feel free to add: Reviewed-by: Ritesh Harjani (IBM) <[email protected]> However, note that we should still get the stable tag / fixes thing sorted, for stable backports to do it's right thing.
