@jonmasters You are assuming that I am not an AI. I am touched, that is the nicest thing anyone has said about me all day! ;-)

@penberg Four-in-a-box CEO and CTO? Interesting. ;-)

@baiwfg2 Glad you liked it, and yes, learning any large body of software, including the Linux kernel, does take quite a bit of time and effort. ;-)

@andthedropout @narenm07 @paulg @JohnathanBi For whatever it is worth, much later in life I did enjoy Karl Popper's "The Open Society and Its Enemies", which I (perhaps incorrectly) interpreted as a stinging critique of a number of schools of philosophical thought.

@narenm07 @paulg @andthedropout @JohnathanBi I tried to take philosophy twice in university, and failed both times. I didn't fail the class, I instead failed to take the class. I asked too many questions, so was given independent study projects and told me to stay out of the classrooms. In retrospect, this was optimal.

@joel_linux Much as I would like to apply that guardrails technique to curling, they have rules that rule out the obvious guardrails. ;-)

@esrtweet There are worse examples to follow than that of John Henry. ;-)

@nileshtrivedi There is an epub version of it posted there, if you would like. (Yes, I do need to do another release, don't I?)

@rexstjohn Use their collected works in your weightlifting routine?

@shachaf The more recent load-acquire and store-release approach works much better. Or, for barrier instructions, the powerpc lwsync instruction, which can implement both smp_rmb() and smp_wmb() along with load-acquire and store-release.

@shachaf Alpha has only full-memory-barrier ("mb") and write-memory-barrier ("wmb") instructions. The Linux-kernel smp_rmb() macro therefore expands to the "mb" instruction. Yes, READ_ONCE() could use rmb() instead of mb(), but it would result in the same "mb" instruction.

@shachaf DEC Alpha's hardware memory model, which has split caches that do not interact with each other without memory-barrier instructions (see Section 15.6.1). Note that the Linux kernel requires Alpha to use full memory barriers in READ_ONCE() and friends for exactly this reason.

@shachaf Hello, and glad you liked it! For more up-to-date material, please feel free to take a look at Chapters 3 and 15, along with Appendix C here:

Excellent LWN writeup of my Kangrejos talk:

@joel_linux As opposed to, say, unreliable people? Though you could reasonably argue that a great many people are indistinguishable from LLMs. But I am happy to let you make that argument. ;-)

@debasishg And that is worth a call-out to Dipankar Sarma, who did the heavy lifting required to get RCU accepted back in the day!