| Age | Commit message (Collapse) | Author |
|---|
|
|
|
+ Saves a multiplication, very minor
|
|
+ Gives a hint to the LTO optimizer to more agressively inline the functions,
seems to bring a slight performance increase
|
|
+ Speeds up fork a little bit
|
|
|
|
|
|
|
|
+ Worked in qemu, failed in visionfive2, fixed.
|
|
+ Processes won't be able to read previous stack frames etc
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
+ Anything extern is right out as LLVM doesn't produce correct code for
them. Maybe if I added some extra attributes but I'm skeptical.
Replaced with static variables and getters/setters.
+ Inline ASM is apparently a bit buggy, so use an assembly stub when jumping to init.
+ Minimize work done in main() to minimize chance of LLVM doing something silly.
Still not 100% certain that I shouldn't just write the main() as an assembly stub
in arch/riscv64 to be absolutely sure everything works as intended.
+ Make .kernel.start section SHF_ALLOC, otherwise lld complains about
pc-relative addressing
Probably some other stuff as well that I'm forgetting right now. But at least with
LLVM14 LTO seems to work, which is pretty cool?
|
|
+ Align kernel to 2MiB boundary in u-boot
+ Optimize alignment functions a little bit, should still have to check
on real hardware but 'feels' more clean
+ Add early boot debugging
+ Put extra cores we aren't ready to account for to sleep if/when they
boot.
+ Make BASE_PAGE_SIZE constant on riscv64/32, helps the compiler with
some alignment checks among other things.
|
|
|
|
+ No real point having multiple different levels of assertions, just say
you assert something and be done with it
|
|
+ Had some minor issues with a wraparound of size_t that effectively
meant that some regions were allocated twice. Also, booting should be
a bit more reliable now, turned out that the previous iteration of the
booting was just accidentally working due to the kernel being placed
'close enough' in RAM to where it was linked to. Fixed by allocating a
vmem of O1 that maps the kernel to a 2MiB boundary at boot, pretty
nifty.
|
|
+ dbg_fdt() is apparently broken, possibly due to UB or something, but
it causes some issues with optimizations enabled. Remove it
temporarily
|
|
|
|
+ The system is now a bit simpler and hopefully easier to understand, while also
extending the shared memory to be 1:N, where there is one owner who
may become a zombie while waiting for the N to die.
|
|
|
|
+ Took some fairly significant changes, for one the kernel is no longer
relocated at the start of a boot, instead it sits wherever the user
decides the kernel should sit. Similarly, the initial kernel stack and
page table are stored within the binary, slightly bloating the size
but making it much safer to boot since there's really no chance of us
overwriting the fdt or initrd in memory.
|
|
|
|
+ Allow threads to make themselves become orphants
|
|
+ Should write this down somewhere but the idea is that when a process
gets killed, it frees all the memory it can, making all threads within
that process orphans. Orphaned threads are assigned to the init
process, which will generally call exit() on each one. Zombie threads
are threads that own some bit of shared data, and whose reference
count is above zero. They may not be swapped to or called, even though
they take up space in thread map and reserve their thread ID.
|
|
|
|
|
|
+ Gives a very slight improvement to RPC speeds, but mostly cleans up
code a little bit.
|
|
+ Fairly considerable speedup, as we don't have to look up the rpc pte
every time separately, instead cacheing them. Adds an architecture
specific limitation to total rpc stack size, though.
|
|
|
|
|
|
+ Make 8250 serial driver more generic
+ Still TODO: write a tutorial on how to boot on the visionfive2
|
|
|
|
+ Not bootable quite yet. Among other things, I couldn't
get the current starfive u-boot fork to boot, so
try adding support for booting with precompiled u-boot
via the `go` command. Initial testing with qemu shows that
this should be possible, and if it works, might be useful
in the (far) future with other slightly janky SBCs.
Also, NS16550 is 8250-based, and I'm really only using the base
8250, so rename and add visionfive 2 uart to list of compatibles.
Visionfive 2 is still completely untested.
|
|
+ Both kind of go hand in hand, made sense to do both at the same time.
Some parts feel slightly hacky, the loader works by placing everything
on the stack and avoiding global values. Still, seems to work?
|
|
|
|
|
|
|
|
|
|
+ The secret is using gravestones. I'll have to write up full
documentation for the feature but essentially riscv lets us encode
whatever we want into page table entries, as long as they're not
active. We use this to encode highest user address that is not a zero,
in that all entries in the top level are either active or gravestones.
When an active entry is removed, it is either a gravestone (if there
are other active entries above it) or it starts a cascade of removing
entries that have been previously removed
Slight runtime overhead to page mapping, pretty major advantage in rpc
calls. Feature will need to be tested more thorougly, and the init
program is sort of a best scenario with just one top level userspace
page table entry active at a time, leading to incredibly fast context
switches.
Current implementation limits a process' max virtual memory to 248 GiB
(in Sv39), but I don't think the missing 8 GiB is that big of a deal.
|
|
+ It really only works with Sv39, I suppose similar structures should be
added for Sv4} etc. if I ever get around to it.
|
|
+ Both easier to look at and now the userspace doesn't play as big of a
role in the 'benchmarking' with optimizations turned on.
|
|
|
|
|
|
|
|
|
