News for hackers. Stuff that matters.
Re:Ways of Implementing a System Call by pcr910303 on Friday January 17, 2020 @03:19AM (#22072854)
OK, I really don't know anything about assembly/OSes, but if the price of going to kernel mode with sysenter/syscall is smaller than a not that pricey(the article says it's less than a single 64-bit integer division), does that mean microkernels can utilize that to improve performance?

What are the shortcomings of the method compared to interrupts?

Re:Ways of Implementing a System Call by dfox on Friday January 17, 2020 @08:14AM (#22074298)
On microkernels the issue is usually not with syscall performance but with essentially every operation causing a task switch which on i386 causes TLB flush (which in post-Spectre days seems somewhat moot) and in many cases involves interacting with task scheduler.

(QNX uses interesting hack to sidestep this: the message passing interface does not really pass messages, but is simply cross-addressspace call/return that in essence temporarily moves a thread to different process without rescheduling anything)

Re:Ways of Implementing a System Call by monocasa on Friday January 17, 2020 @08:06AM (#22074240)
The l4 variants think so. Unfortunately Spectre was sort of a pain and now they have to switch page table roots a bunch.