Hello Greg,
On Tue, Sep 10, 2024 at 03:15:21PM +0200, Andrea Calabrese wrote:
Code refactoring using the recent guard and scoped_guard macros
for automatic cleanup of the spinlocks. This does not change the
effective behaviour of the kernel, but guarantees a cleaned-up exit from
each lock, automatically avoiding potential deadlocks.
The "guarantee" is there today, right? So this isn't really anything
other than a "convert to use new apis", right?
Basically, yes. That is why I thought it was trivial...
If so, I need to see a LOT of verification that the output is the same,
and that it has been properly tested. Converting working code for no
real reason other than "let's change this!" isn't always a good idea.
I agree that this has to be tested more. Right now, I tested it on my
laptop, on virtme-ng and qemu and had no problem. If you have any
suggestion on increasing the number of tests, I will test it also on that
(the more, the merrier).
Regarding the "let's change this", well, I thought it had to be done at
some point, since the API for RAII-like declarations are all there.
Signed-off-by: Andrea Calabrese <andrea.calabrese@xxxxxxxxxxxxxxxxxxxx>
---
Diff from V3: as Greg KH and Lucas Stach noticed, there was a
behavioural change between the two versions: I used guard(spinlock),
while the expected behaviour should have come from a spinlock_irqsave
guard. This has been fixed.
No diffstat?
Diffstat should be:
devres.c | 109
+++++++++++++++++++++++----------------------------------------
1 file changed, 40 insertions(+), 69 deletions(-)
sorry for not including it (although the patch was generated
automatically...
oh well).
Best regards,
Andrea Calabrese
