12:29 <dramforever> I've been going down the SiFive errata CIP-453 rabbit hole, and I have a *lot* of questions. So if you have any comments please share them. I mostly just want to know I'm not going insane. That being said...

12:31 <dramforever> 1. The CIP-453 patch in Linux. Why is this the assembly sign extension part this thing with a branch and not just slli/srai? Is it to avoid a redundant store when unnecessary? https://lore.kernel.org/all/1616423166-13857-5-git-send-email-vincent.chen@sifive.com/

12:34 <dramforever> 2. SiFive says in their errata docs to use ((mepc + mepc ^ mtval) & 2) for the actual mtval for instruction faults. But I think they meant (mepc + ((mepc ^ mtval) & 2)), right? https://www.sifive.com/boards/hifive-unmatched 'Freedom U740-C000 Errata'. They ask for contact information but you can just put anything

12:38 <dramforever> 3. This code looks like it would harmlessly work on anything that only has 16- and 32-bit instructions. But no it doesn't work on QEMU. I acknowledge that I was probably doing some weird stuff at the time, but I think I managed to get SIGSEGV with pc hundreds of bytes before si_addr.

12:39 <dramforever> I unfortunately forgot how to reproduce it because it was last year and just shrug it off because it didn't seem to matter, but I think I managed to reproduce basically the same thing by just running KVM: https://gitlab.com/qemu-project/qemu/-/issues/1155

12:42 <dramforever> Is this something that's known to occur? It seems to be related to a string of straight line code that crosses page boundaries, but I've otherwise been unable to figure it out.

21:40 <conchuod> mmind00: what's the name of the program/webserver you use to access the serial terminals on your boards?

21:48 <mmind00> conchuod: at the base it's just ser2net nodejs and an nginx ;-) ... + a lot of homegrown javascript voodoo: https://github.com/mmind/boardfarm-mgmt

21:49 <conchuod> Cool, thanks!

21:49 <conchuod> I think -12 years is before I even wrote my first program.

21:50 <mmind00> conchuod: tanks for making me feel old :-P ;-)

21:50 <conchuod> Sorry!

21:50 <mmind00> conchuod: hehe no worries

21:52 <conchuod> I might've bought a couple ykush things

21:52 <conchuod> Some relay ones and some of the usb

21:53 <conchuod> Wrote some wrapper stuff for them to integrate with my build scripts so next up is automating the serial terminal side of it

21:54 <mmind00> once you have automated some part of board development, you will never be able to work with a board on your desk again ;-)

21:54 <conchuod> Yeah.. I need to get it setup for work too - some of the lad have fancy netio setups but that's always seemed overkill to me

21:55 <mmind00> netio?

21:55 <conchuod> https://www.netio-products.com/en/device/powerpdu-4c

21:56 <mmind00> oh yeah, that really looks like overkill having like 230V outputs when you actually want 12V or 5V on your board :-)

21:58 <conchuod> Yup... I have a PSU on my desk at work so I'm just gonna do what I did for my Icicle at home & get a breakout for the 24 pin

21:58 <mmind00> my boardfarm is 19in hardware in a modified Ikea shelf, so one case holds the main unit + an flexATX power supply to gives me a number of 5V + 12V lines easily

21:59 <conchuod> then I can slot a ykush relay thing in the middle & use that to switch it on and off rather than have a netio -> kettle lead -> regular icicle external psu

22:00 <mmind00> yep ... so much easier

22:00 <conchuod> I already need the PSU for other boards that only take a 6pin so may as well reuse it :)

22:01 <conchuod> Ye I recently got some 19" rack hw, but I need to rebuild my machine into the case etc but it's too hot right now to want to

22:01 <dh`> dramforever: regarding what you posted a while back, my *guess* would be that the qemu behavior you've observed represents (a) it compiled a block beginning at address X, which crosses a page boundary at address Y; (b) it executes this compiled block, which likely happens as a unit; (c) the second page is not present; (d) it throws a trap with X as the restart address (so as to rerun the whole compiled block) and Y as the fault a

22:02 <dh`> whether this is allowable or correct, fiik

22:02 <conchuod> "fiik"

22:02 <conchuod> i like that one

22:02 <dh`> FIIK is a traditional internet acronym :-p

22:05 <dh`> anyway, if that analysis is correct not doing it will slow qemu down for what I'm sure they'll argue is no good reason

22:06 <jrtc27> that shouldn't happen

22:07 <jrtc27> a fetch page fault should result in the QEMU TB containing code to generate an exception at that exact PC, and terminate the TB

22:11 <dh`> possibly someone thought it was a good idea to prefault the page before executing the block?