admin/LithosAnanake
1
1
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

FABRIC-3.10.8 §7.7: MC-3b plan -- per-core IPI receivers, boot round-trip self-test

Browse Source 
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Robert Allan James
2026-09-26 15:36:28 -04:00
co-authored by Claude Opus 5.5
parent 4440faf352
commit d7745b49be
1 changed files with 23 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
FABRIC-3.10.8.md
+23
View File
@@ -1081,3 +1081,26 @@ The boot hart HSM chose (0) is the one OpenSBI names as `Boot HART ID`.
| riscv64 | `logs/20260926-151755` | same; 0 `timed out` | | riscv64 | `logs/20260926-151755` | same; 0 `timed out` |
**MC-3a is complete.** Next: MC-3b, each worker core's interrupt controller and the IPI. **MC-3a is complete.** Next: MC-3b, each worker core's interrupt controller and the IPI.
### 7.7 MC-3b — waking a worker core (plan)
**Change** (§7.2 item 3; nothing runs on the worker core yet):
- **Before marking itself online, each worker core sets up its own interrupt receiver** (`arch_smp_secondary_init()`), for the IPI only:
- amd64: its LAPIC enabled (SIVR, TPR 0), with a new IPI vector 0x22 and its own ISR stub. Vectors past 33 otherwise all land in the "255" default stub (the same trap the keyboard's vector 33 hit, `isr.S`).
- aarch64: its banked GIC CPU interface and SGI 0 enabled. Its target mask is read from its own banked `GICD_ITARGETSR0`.
- riscv64: `sie.SSIE`.
- **Sending** (`arch_smp_send_ipi()`):
- amd64: a fixed IPI through the ICR.
- aarch64: `GICD_SGIR` with the target mask.
- riscv64: SBI IPI `send_ipi` for its hartid.
- **Receiving.** Each ISA's interrupt handler recognises the IPI, counts it in the receiving core's block (`ipi_received`) and acknowledges it: LAPIC EOI; the GIC's full IAR value to EOIR, since an SGI's IAR carries its source CPU; `sip.SSIP` cleared on riscv64. Core 0's existing handlers are unchanged.
- **Parking with a way out.** The park step now lets exactly one interrupt in, without a lost-wake window:
- amd64: `sti; hlt; cli` (the `sti` shadow covers the `hlt`).
- aarch64 and riscv64: `wfi` with interrupts masked (it wakes on any pending, enabled interrupt), then unmask and mask again to take it.
- **Boot self-test** in `sk_smp_init()`: for each worker core, core 0 sends 3 IPIs one at a time, each waited for under a 100 ms bound, and prints the round trips.
- **The idle-point report** now logs a core's wakes together with its IPI count, when either changes. A wake without an IPI is the "spurious" figure.
**Evidence (named in advance):**
- **All three ISAs:** `SMP: core 1: IPI round trips 3/3`, then one idle-point line with **3 wakes, 3 IPIs**, and no later line in the run (no further wake).
- **The standard sequence unchanged:** K-10 hashes, G byte-identical, all ok `3`, 0 B idle growth, 0 order violations, 0 double claims.
- **Host `ps -L` on aarch64 and riscv64:** `CPU 1/TCG` stays at about 0% after boot.