User Details
- User Since
- Jun 2 2014, 4:20 PM (650 w, 2 d)
Today
So apart from the commit message, this is OK. It does add an additional bogus va == pa requirement for runtime entries, but we already do it elsewhere (which is also bogus: I ran for years on a LinuxBoot arm64 machine w/o this requirement... But I've lost access to that environment so since I can't test it, I can't ask someone else to fix something they can't test)....
Sun, Nov 15
Sat, Nov 14
Fri, Nov 13
This looks reasonable to me. I'd be interested in @mav 's take on it.
This looks good... but it might be good to have at least a reference FAT image with the mal-formed LFN entry for regression.
Thu, Nov 12
I'll take a look at this. I also have a series that has a fix for this in it. I'll see which is better.
Yea, I forgot to remove these.
I think it is time to axe those too.
codex also flags Three earlier defects also recur here: address-sign classification rejects valid low kernel addresses (line 154), unresolved frames are removed and entirely unresolved samples disappear (line 172), and L3-miss counting omits DRAM-served misses (lines 213–217).
codex flagged a couple spec vs code issues.
generally good (and cool functionality). Lots of nits.
So there's apparently a problem with --demangle
I know I'm being picky about the ordering here (and I should have flagged it a while ago), but there's a desire, when possible, to have each rev committed be bisectable. There's a proposal to allow 'loopback branches' to stack multiple commits and have only the last one be buildable, but that's not gone anywhere.
Thi
Wed, Nov 11
Tue, Nov 10
Yea. Life happened. I'll try to build this tonight
I'm not sure tagging every string in the kernel with __null_terminated is viable. Can you explain this a little more?
Oh! Right! this is what you needed all that for.
Thanks ken! Looks great now I think.
Code wise, this looks good.
Co-authored-by is reserved for humans. Use assisted-by instead.
Sun, Nov 8
Fri, Nov 6
Thu, Nov 5
This looks good, but I didn't verify all the details of clocks and baud generation.
I'm not going to be able to do the netflix testing after all, so go for it. If there's problem, then they can report them on upgrade.
Wed, Nov 4
This looks fairly clean. I'm not seeing anything really big. I'd chunk it smaller, but after going back and forth, I'm just going to mention it.
Somewhere in .9 since they are kernel only.
I'd look at tvtohz.9 as a model maybe?
I'd call the man page SEC_TO_NSEC.9 and have MLINKS for the other routines.
Shouldn't be a big add, if you need help, give a yell.
Tue, Nov 3
So I approved the last one, but not this one. We need man pages for these things.
Mon, Nov 2
woof! Way better.
I'm going to click OK here... but this can lead to 1ms or 2ms delays in processing other CAM devices since it's a busy wait. For MMC, that's likely OK. They don't usually have other devices on the system that are so performance critical that a stall like this would matter... However, as usage shifts, we may need to revisit one day. For now I think this is OK, with my mentioning the caveat in case people trace some future performance problem back to this commit and then to this review.
Smart. NVMe didn't used to need anything like this, but it might now with apple hardware support.
Similar comments: UFS mirrors NVMe closely in its setup, and nvme does this.
mirrors what nvme controller does. The different SIMs are different in how they handle this, but I'm partial to this method, and I think for the world of UFS, it's simplicity is better than other SIMs that need added complexity to deal with some things failing and some not.
This looks right to me....
After reading through everything else, I'm not sure why we can't load the kernel.elf into low memory. But that's code on arm64 since we have no userboot.so there. I'm less worried about this than with my first comment, but I'd be interested to see the boot time differences.
If it builds, I'll push it in. But NOTES is used by all the architectures, so it will take me a while to test.
Doesn't kernel.elf still load in this case? We've been shying away from kernel.bin in general, so I'd like to understand the need. The ELF loading code is small, especially when we don't have any drivers.
Sat, Oct 31
Fri, Oct 30
The code looks good, and make things safer. Not sure I understand the comment I flagged, but that's likely nothing.
My comments are generally small for this, so this is likely fine as is.
The tuning algorithm for the rockport I didn't audit beyond looking for obvious mistakes, everything else I put under the microscope.
Thu, Oct 22
This is a proper migration to a newer API.
Wed, Oct 21
Oopps! I thought I'd accepted this before now. Done.
Landed w/o proper review markings, my bad.
Oh, But I pushed w/o the reviewed stuff. My bad.
updated and pushed! Thanks for the feedback!