comex

Or, rather, you can, but it’s slower and more complicated; you need to do “IPC” from inside the VM to your helper function outside it. Another big downside: some machines don’t support HW virt at all, and nested virtualization is still iffy. Situation is improving though.

- No need to worry about Windows expecting %rsp to be in a given range - Lots of other fun tricks. The big downside: if JIT code runs in an isolated environment, you can’t just have it call helper functions implemented as normal userland code that can use host OS APIs.

…it gives you full control over the address space and register usage. So - You can identity map the target address space if you want (slight code size decrease) - You can point %fs or %gs to common structures / base pointers you have to index into, rather than reserving a GPR

Another example: if you’re writing a JIT for an emulator, you could run the JITted code in HW virt under your own tiny “kernel”. This lets you handle traps and modify page tables orders of magnitude faster than under a standard kernel. Also…

No validation required -> less chance to screw up. Faster. And in retrospect, easier to guard against Spectre type attacks. Or even if you still want to validate machine code (say, to ban CLFLUSH), HW virt can add an extra layer of protection.

For example, remember NaCl and its machine code validator that allowed untrusted programs to run in the same address space as trusted code? Well, that’s nice, but… with HW virt you could just run them in their own address space but still control all access to outside world.

Yay, Windows is adding an unprivileged hardware virtualization API to match Linux (/dev/kvm) and more recently macOS (Hypervisor.framework): https://news.ycombinator.com/item?id=17492466 … This should make it more feasible to write emulators, VMs, etc. that rely on HW virtualization internally!

clarification - that doesn’t mean I think she deserved to be fired, or that there weren’t legitimate factors predisposing her (among many others) towards being... unimpressed with rando replies.

(which may also have objective aspects), without meaning to imply that the objective aspects are wrong. </monologue>

but from the streamer’s perspective, I suspect the wording was influenced by a sense of “I disagree subjectively about how much to value player choice”. not that that’s what he said, exactly. but the subjective aspect makes it possible to “disagree” with a statement [..]

which I suspect contributed to the conflict. I can see how responding to a *factual* exposition with “allow me to disagree” could come off as very patronizing - as a suggestion that you don’t know what you’re talking about. [..]

ultimately, this is to a large extent a matter of opinion: do you prioritize player choice enough to spend a lot of time/budget making a complex branching narrative, or do you spend that time/budget making something else cool instead? [..]

and there’s no doubt that Price (like anyone else with experience in the matter) was well aware of the possibility of branching dialogue, and just didn’t mention it because it would be a large discussion by itself. nevertheless, not mentioning it makes it a valid objection. [..]

this matters because she started out with the broad claim that “I'm not sure if it's possible to make an MMORPG character compelling”, and the explanation does ignore that possibility. now, obviously there are good reasons to limit branching dialogue - it adds a lot of cost.

it’s not a, quote, “explanation of how narrative design influences player expression in the sort of games that Price narratively designs.” it’s very specifically a criticism that the original tweets don’t consider the possibility of branching dialogue. [..]

- reading the same memory address on two different cores (both in a loop) yields two different results (depending on the core) - adding a cache flush instruction fixes it - the mapping is marked as cache coherent - wtf