Spinlocks and hartlocks by auctumnus · Pull Request #87 · UMN-Kernel-Object/ukoos

auctumnus · 2026-01-30T15:31:20Z

This PR adds spinlock and hartlock functionality.

Somewhat supersedes part of #33.

remexre · 2026-01-30T20:30:01Z

src/kernel/include/arch/riscv64/irq.h

+// TODO: move this over to insns.h?
+#define RISCV64_CSR_SSTATUS_SIE (1 << 1)
+
+// TODO: consider making these common across architectures somehow?


i kinda want these to be private to hartlocks, but they'd still need to be per-arch...

tbh, could you prototype them in hartlock.c and implement them in arch/riscv64/hartlock.c or smth?

private to hartlocks as in the spinlocks would take a hartlock instead?

or, well, in addition to being a spinlock, they would also take the hartlock

src/kernel/include/arch/riscv64/insns.h

remexre · 2026-01-30T20:43:19Z

src/kernel/spinlock.c

+  // Otherwise, we might get interrupted and then attempt to re-acquire the same
+  // spinlock, leading to a deadlock.


i'm missing what the problematic sequence of events here is; shouldn't the atomics make this not a problem?

hmm .... i guess on second thought it shouldn't be a huge problem, i forgot that if we get interrupted we would end up dropping the lock before returning back

(e.g. if we went

loop trying to acquire it while it's locked

get interrupted

try to acquire the same lock
then this is fine, it's locked by a different hart which should eventually drop it)

src/kernel/spinlock.c

remexre · 2026-01-30T20:46:59Z

src/kernel/include/arch/riscv64/irq.h

+#include <insns.h>
+#include <types.h>
+
+// TODO: move this over to insns.h?


yes; make it an enum blah : u64 too, so we get warnings if we try to use it in a too-small type

ig i didn't write down my problem here which is that nominally it's more of a struct than an enum; i.e. there are fields in sstatus that are more than 1 bit in size, and i'm unsure how to nicely represent that

remexre · 2026-02-19T02:15:32Z

still todo:

hartlocks
bitfields
terrible awful ub at the end of the file

Signed-off-by: Amy Ringo <me@remexre.com>

These functions are for mapping IO memory. Signed-off-by: Amy Ringo <me@remexre.com>

Signed-off-by: Amy Ringo <me@remexre.com>

Closes UMN-Kernel-Object#90. Note that this depends on GCC 15 (UMN-Kernel-Object#82) as well. GCC 16 will let us simplify the -march line to -march=rva23s64, but it's not released at the time of writing. Signed-off-by: Amy Ringo <me@remexre.com>

This makes it easier to test concurrency-related code. Signed-off-by: Amy Ringo <me@remexre.com>

This is used by make watch. Signed-off-by: Amy Ringo <me@remexre.com>

The {bytes:?} behavior doesn't match what Rust does, but it's a useful behavior for "typically a string" fields. Signed-off-by: Amy Ringo <me@remexre.com>

Signed-off-by: Amy Ringo <me@remexre.com>

We need to ensure that tasks are only scheduled on harts with the same register_save_area size, meaning we need to track which harts have which size. It's easiest to keep track of this if we hook that tracking into ordinary device detection, so this device class exists for that purpose. Because init_boothart_hart_locals_late panics if it cannot find the hart device for the boothart, this will not boot at the moment, though it will build. The next patch fixes this. Signed-off-by: Amy Ringo <me@remexre.com>

This restores the ability to boot, since the hart will now be properly detected. Signed-off-by: Amy Ringo <me@remexre.com>

Signed-off-by: Amy Ringo <me@remexre.com>

Signed-off-by: Autumn <auctumnus@pm.me>

Co-authored-by: remexre <me@remexre.com> Signed-off-by: autumn <auctumnus@pm.me> Signed-off-by: Autumn <auctumnus@pm.me>

Signed-off-by: Autumn <auctumnus@pm.me>

auctumnus · 2026-03-14T21:01:30Z

responsible ai disclosure: i got claudeパイセン to thread struct hartlock *hartlock thru the call graph in the alloc init stuff

auctumnus force-pushed the remexre/tasks branch from 8dd242b to 1732f04 Compare January 30, 2026 15:32

auctumnus changed the title ~~Remexre/tasks~~ Spinlocks Jan 30, 2026

remexre reviewed Jan 30, 2026

View reviewed changes

auctumnus force-pushed the remexre/tasks branch from 9208837 to 4b5a80c Compare January 31, 2026 19:33

remexre force-pushed the remexre/tasks branch 6 times, most recently from 241654e to cf06aaf Compare February 14, 2026 17:07

remexre force-pushed the remexre/tasks branch 7 times, most recently from ba047e3 to e277ac4 Compare February 28, 2026 19:21

remexre added 11 commits February 28, 2026 13:23

Adds support for pointers to align_down and align_up.

87e4f04

Signed-off-by: Amy Ringo <me@remexre.com>

Implements iomem_map and iomem_unmap.

0f036e7

These functions are for mapping IO memory. Signed-off-by: Amy Ringo <me@remexre.com>

Makes devicetree_print prints the path to the node.

7b8642c

Signed-off-by: Amy Ringo <me@remexre.com>

Adds initializers and documentation about them.

2a50ea8

Signed-off-by: Amy Ringo <me@remexre.com>

Adds a notion of device objects.

9372ec8

Signed-off-by: Amy Ringo <me@remexre.com>

Adds support for enumerating devices in a Devicetree.

0287b44

Signed-off-by: Amy Ringo <me@remexre.com>

Adds the UART device class.

df910ec

Signed-off-by: Amy Ringo <me@remexre.com>

Adds the ns16550a driver.

42c8ab0

Signed-off-by: Amy Ringo <me@remexre.com>

Adds a script that build-tests each commit since trunk.

9e91c26

Signed-off-by: Amy Ringo <me@remexre.com>

Switches qemu-riscv64 target to RVA23.

dacbac9

Closes UMN-Kernel-Object#90. Note that this depends on GCC 15 (UMN-Kernel-Object#82) as well. GCC 16 will let us simplify the -march line to -march=rva23s64, but it's not released at the time of writing. Signed-off-by: Amy Ringo <me@remexre.com>

Gives the qemu-riscv64 target a second hart.

7e5eb72

This makes it easier to test concurrency-related code. Signed-off-by: Amy Ringo <me@remexre.com>

remexre added 4 commits February 28, 2026 13:24

Adds watchexec to the devcontainer and flake.

c887d76

This is used by make watch. Signed-off-by: Amy Ringo <me@remexre.com>

Adds documentation for the bytes format directive, adds {bytes:?}.

1819a64

The {bytes:?} behavior doesn't match what Rust does, but it's a useful behavior for "typically a string" fields. Signed-off-by: Amy Ringo <me@remexre.com>

Adds the strnlen builtin.

5d91e35

Signed-off-by: Amy Ringo <me@remexre.com>

Adds a page documenting lists.

bc35bb0

Signed-off-by: Amy Ringo <me@remexre.com>

remexre force-pushed the remexre/tasks branch 2 times, most recently from 0d69730 to 56e119a Compare February 28, 2026 20:41

remexre added 8 commits February 28, 2026 14:57

Adds a driver for riscv64 harts.

475db86

This restores the ability to boot, since the hart will now be properly detected. Signed-off-by: Amy Ringo <me@remexre.com>

Adds a page documenting guards.

a68132c

Signed-off-by: Amy Ringo <me@remexre.com>

Moves "Threads and Harts" docs page to the start of a tasks section.

7cdb80a

Signed-off-by: Amy Ringo <me@remexre.com>

Adds the refcount_t type.

5e0a3c9

Signed-off-by: Amy Ringo <me@remexre.com>

ckpt tasks impl

2c74d27

mutex ckpt

3fb19c0

checkpoint of tasks; gonna need cpu drivers

2b91acc

Signed-off-by: Amy Ringo <me@remexre.com>

remexre force-pushed the remexre/tasks branch from 56e119a to 2b91acc Compare February 28, 2026 21:09

auctumnus changed the title ~~Spinlocks~~ Spinlocks and hartlocks Mar 14, 2026

auctumnus force-pushed the remexre/tasks branch from 0811e54 to 21501b2 Compare March 14, 2026 20:46

remexre and others added 7 commits March 14, 2026 15:46

ckpt tasks impl

51f9084

Signed-off-by: Autumn <auctumnus@pm.me>

feat: spinlocks

f3fd82a

Signed-off-by: Autumn <auctumnus@pm.me>

docs: add license for spinlock.c

fbc1d16

Signed-off-by: Autumn <auctumnus@pm.me>

fix: use <stdatomic.h>

404ff17

Signed-off-by: Autumn <auctumnus@pm.me>

Update src/kernel/include/arch/riscv64/insns.h

bd33691

Co-authored-by: remexre <me@remexre.com> Signed-off-by: autumn <auctumnus@pm.me> Signed-off-by: Autumn <auctumnus@pm.me>

review comments from 02/18

5cb8cec

Signed-off-by: Autumn <auctumnus@pm.me>

feat: hartlocks

c924289

Signed-off-by: Autumn <auctumnus@pm.me>

auctumnus force-pushed the remexre/tasks branch from 21501b2 to c924289 Compare March 14, 2026 20:47

remexre force-pushed the remexre/tasks branch 4 times, most recently from c62a624 to 3709edb Compare March 28, 2026 17:47

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Spinlocks and hartlocks#87

Spinlocks and hartlocks#87
auctumnus wants to merge 32 commits intoUMN-Kernel-Object:remexre/tasksfrom
auctumnus:remexre/tasks

auctumnus commented Jan 30, 2026 •

edited

Loading

Uh oh!

remexre Jan 30, 2026

Uh oh!

auctumnus Jan 31, 2026

Uh oh!

auctumnus Jan 31, 2026

Uh oh!

Uh oh!

remexre Jan 30, 2026

Uh oh!

auctumnus Jan 31, 2026

Uh oh!

auctumnus Jan 31, 2026

Uh oh!

Uh oh!

remexre Jan 30, 2026

Uh oh!

auctumnus Jan 31, 2026

Uh oh!

remexre commented Feb 19, 2026 •

edited

Loading

Uh oh!

auctumnus commented Mar 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

		// Otherwise, we might get interrupted and then attempt to re-acquire the same
		// spinlock, leading to a deadlock.

Conversation

auctumnus commented Jan 30, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

remexre commented Feb 19, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

auctumnus commented Mar 14, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

auctumnus commented Jan 30, 2026 •

edited

Loading

remexre commented Feb 19, 2026 •

edited

Loading