Fatal paths write the console UART only under its registers, whose lock knows which CPU's fatal path holds it by Japabu · Pull Request #675 · ToyOSOrg/ToyOS

The nested-NMI report holds the console's registers, so another CPU's burst never lands inside it

Japabu

and

claude

committed

21e3fc8

Merge remote-tracking branch 'origin/main' into wt/toyos-nmiloud

Japabu

committed

742780b

Every fatal path writes the console through registers it asked for, and re-enters its own CPU's hold

Japabu

and

claude

committed

f9835df

Review round 3: the console lock reads its own CPU and refuses a burst under its fatal hold; the UART moves a byte only with Registers

Japabu

and

claude

committed

7d6a152

nested_nmi_verdict cuts a burst with no line end into each report line, so each of whole's clauses is held alone

Japabu

and

claude

committed

4a3e601

The seize model joins its holder before it asserts, so a seize that gives up fails the test instead of aborting it

Japabu

and

claude

committed

f71a5fd

Review round 4: each CPU's id read is held against its roster id at bring-up, and every UART touch takes the registers

Japabu

and

claude

committed

1f88a38

File that the AArch64 boot CPU's id read is held against nothing

Japabu

and

claude

committed

53dc49d

smp::commit says why the boot CPU refuses an AP's id read, not the AP

Japabu

and

claude

committed

1a9380f

Review round 5: an AP's id read rides in the roster's own echo, and Roster::commit refuses a mismatch

Japabu

and

claude

committed

d31faab

Review round 6: Roster::echo reads its own CPU's hardware id

Japabu

and

claude

committed

3b961d5

Navigation Menu

Repository navigation

Fatal paths write the console UART only under its registers, whose lock knows which CPU's fatal path holds it - #675

Fatal paths write the console UART only under its registers, whose lock knows which CPU's fatal path holds it#675
Japabu merged 11 commits into
mainfrom
wt/toyos-nmiloud

Commits on Oct 1, 2026

The nested-NMI report holds the console's registers, so another CPU's burst never lands inside it

Merge remote-tracking branch 'origin/main' into wt/toyos-nmiloud

Every fatal path writes the console through registers it asked for, and re-enters its own CPU's hold

Review round 3: the console lock reads its own CPU and refuses a burst under its fatal hold; the UART moves a byte only with Registers

nested_nmi_verdict cuts a burst with no line end into each report line, so each of whole's clauses is held alone

The seize model joins its holder before it asserts, so a seize that gives up fails the test instead of aborting it

Review round 4: each CPU's id read is held against its roster id at bring-up, and every UART touch takes the registers

File that the AArch64 boot CPU's id read is held against nothing

smp::commit says why the boot CPU refuses an AP's id read, not the AP

Review round 5: an AP's id read rides in the roster's own echo, and Roster::commit refuses a mismatch

Review round 6: Roster::echo reads its own CPU's hardware id