- User Since
- Mar 11 2014, 8:46 PM (201 w, 4 d)
Fri, Jan 19
Thu, Jan 18
I think should_yield is pretty broken regardless of this change. I think the idea is to demote to PRI_USER if the thread has been running for more than hogticks without yielding the CPU. Rather than using 'ticks' which counts walltime and not CPU time, I wonder if it would be better to have a 'td_swvoltime' (for lack of a better name) which is set to 'td_runtime' in the SW_VOL case and left alone in the SW_INVOL case. To check for a 'hog' you would do something like:
More tweaks to active.
- Use 'mapped' instead of 'active'.
- Add MC# stacks to mini-kernel map for PTI.
- Adjust 1 label so we always load curpcb.
- Don't save FS/GS bases for MC#.
- Remove labels not used since r190620.
Wed, Jan 17
Committed in rS328099.
I attempted to update the comment a bit to not claim that there is only one correct way, but that mapping ELF objects directly is a grey area. I still prefer the approach in the patch, but hopefully the message is now more accurate?
- Further tweaking and update similarly stale comments in the vmmeter header.
Tue, Jan 16
Possibly s/understood/supported/ in the commit message, but that's a minor nit.
- Don't setup progtab entries for sections not loaded by the loader.
Yes, though we already use anonymous unions in the tree now (e.g. struct mbuf uses them).
I'm not quite sure what you would like to do for makecontext. It is already using the right value (16 for newabi), but I don't know if it's really safe to expose STACK_ALIGN to userland (perhaps I can include <machine/md_var.h> to get it in this file?) If the comment is really stale and there is no gap for newabi, then I don't think we can collapse the #ifdef's down any further in makecontext() aside from perhaps using STACK_ALIGN.
- Use STACK_ALIGN in cpu_set_upcall() as well.
- Update comments in __makecontext().
Mon, Jan 15
Sun, Jan 14
Fri, Jan 12
Is llvm-objdump far enough along to implement the functionality needed by ports?
Thu, Jan 11
crd_iv is only used after the copyin() if it succeeds. It shouldn't have any different data as a result of this change once a request is submitted to the driver.
Wed, Jan 10
I am a bit surprised it worked fine without the patch. Perhaps if you used watchpoints both outside and inside at the same time you might get some inconsistent results? But I'm glad that I don't seem to have broken anything for SVM. Can someone confirm that setting "DR" in the VMCB cache flags is correct (i.e. equivalent to the VT-x changes)?
In general this looks ok to me. As we discussed on IRC I think it would be useful to split up into multiple commits. (e.g. ifunc, fpu/npx usage, pmap usage, fast_syscall changes that don't use SMAP (remove cld and shared function), and then SMAP itself).
- Drop stale comments.
Tue, Jan 9
Mon, Jan 8
Fri, Jan 5
biospci_readconfig() doesn't look for a matching entry, it just invokes a BIOS routine that probably just translates the register arguments into the right values for writing to 0xcf8 and reading from 0xcfc.