* Wei Yang <richard.weiyang@xxxxxxxxx> [250123 20:44]: > On Thu, Jan 23, 2025 at 12:52:40PM -0500, Liam R. Howlett wrote: > >* Wei Yang <richard.weiyang@xxxxxxxxx> [250117 00:49]: > >> On Wed, Nov 27, 2024 at 08:31:13AM -0500, Liam R. Howlett wrote: > >> >* Wei Yang <richard.weiyang@xxxxxxxxx> [241126 20:28]: > >> >> Here is some cleanup related to spanning write. > >> > > >> >None of these fix anything, but do fiddle with code that's pretty > >> >critical to the kernel. Most of the changes will be immeasurable in > >> >change but carry risk to causing subtle changes. > >> > > >> >Some are simple removal of returns that aren't used while others change > >> >things because you think they are probably the equivalent. This seems > >> >like unnecessary chrun at this point. I'm all for efficient code but > >> >this is getting a bit much, some of these are just preference of what to > >> >use that will already exist in the cpu cache. > >> > > >> >I'll get back to you when I dig through them, as some need a deeper look > >> >for sure. > >> > > >> >Liam > >> > > >> > >> Hi, Liam > >> > >> Would you mind taking a look when you have time? > > > >Yes, I'll have a look soon. I don't love changes that dive deep into > >complex code that results in no gains (performance or feature wise). > > > >It's also odd to have simple "this return isn't use" and things moving > >code blocks to be executed only in certain scenarios, as the difficulty > >to verify the latter is much higher. > > > >Can we please limit changes to areas where there is a performance change > >or coupled with a change that is needed? ie: stop sending patches that > >change things unless it's with a feature or improvement (performance or > >otherwise). I'm just not convinced some of these are worth the > >cost vs risk. > > > > Ok. > > So you would drop this patch set or still want to take a look? I was going to look at it, but after I send my reply, I received a report of an issue caused in a certain configuration that caused the stack frame to grow out of the configured 1024 limit, which was tracked to a patch you added to simplify a previous function. So, I think we should drop these patches since they don't make a measurable difference and are not without risk. Thanks, Liam
