On 21/01/2022 19:49, Krzysztof Kozlowski wrote: > On 21/01/2022 13:31, Dmitry Osipenko wrote: >> ... >>> @@ -529,12 +536,44 @@ static irqreturn_t tegra30_mc_handle_irq(int irq, void *data) >>> u8 id, type; >>> u32 value; >>> >>> - value = mc_readl(mc, MC_ERR_STATUS); >>> + switch (bit) { >> >> Again, I see that the code wasn't tested :/ Shouldn't be too difficult >> to create memory-read errors to check that at least basics work >> properly. Please always test your changes next time. >> >> So it must be "switch(BIT(bit))" here, please write it like this: >> >> u32 intmask = BIT(bit); >> ... >> switch(intmask) { >> > > Also, please build your changes with W=1... It's the second try of > sending un-tested and not-working code. This time also with a compiler > warning. This looks very bad :( I am afraid this might be taken too literally and W=1 build will replace other required steps, so let me be explicit: We not only expect to compile it but also compile with W=1, run sparse, smatch and coccicheck. Then also test. > > For big companies with a lot of engineers, like nVidia, it is useful if > some internal review happens. It is a nice way to offload community > reviewers which are - like maintainers - a scarce resource. Doing > internal review is not a requirement, but helps to find such mistakes > earlier, before using the community. It is simply nice to us. > Best regards, Krzysztof