piątek, 18 kwietnia 2025 o 00:10:02 UTC+2 Keith Randall napisał(a): On Thursday, April 17, 2025 at 11:50:18 AM UTC-7 Wojciech Kaczmarek wrote:
Hi Gophers, I am experimenting with assembler, ABI0 and stuff. While I was learning the ABI0 frame structure from disassembled code, I noticed that the stack check happens before subtracting from RSP (I'm on arm64, but the same applies to SP on amd64, I mean the hardware register ofc). Then, what if the frame being allocated is big enough to cause trouble? Do you mean, what if the stack check fails? It calls into runtime.morestack. morestack will allocate a new, larger stack and copy the existing contents over. I wasn't quite asking about check fail scenario. The stack check is done before frame is even considered, and later the frame size is subtracted from RSP. But then there is no checking again. Also, after the CALL runtime.morestack*, what is the safe amount to subtract from RSP? There's no specific amount. Probably the frame that didn't fit at the start will fit after calling morestack. But not necessarily, if the frame is really large it will call into morestack again. (That's why after calling morestack we jump to the top of the function and redo the stack check.) Hmm, as above - it seems that g.stackguard0 is compared to the RSP before frame gets allocated on the stack, so the question remains - what frame size is safe? (I am also thinking about some form of limited "alloca", to have a NOFRAME function and make frame manually in the asm - the more it seems important to know what limit on modifying RSP is safe here). -- You received this message because you are subscribed to the Google Groups "golang-nuts" group. To unsubscribe from this group and stop receiving emails from it, send an email to golang-nuts+unsubscr...@googlegroups.com. To view this discussion visit https://groups.google.com/d/msgid/golang-nuts/5dd2b937-faab-43ca-bb43-e5c8cc48ed49n%40googlegroups.com.
