From 8ba4dbd71801a8a1f4ba58813cded56052cac39d Mon Sep 17 00:00:00 2001
From: miod <miod@openbsd.org>
Date: Tue, 23 Jun 2015 19:50:48 +0000
Subject: [PATCH] Disable some of the pmap_emulate_reference() DEBUG checks if
 option MULTIPROCESSOR, and quote the alpha ARM to explain why; while there,
 make the failure messages a bit more detailed.

---
 sys/arch/alpha/alpha/pmap.c | 62 ++++++++++++++++++++++++++++---------
 1 file changed, 47 insertions(+), 15 deletions(-)

diff --git a/sys/arch/alpha/alpha/pmap.c b/sys/arch/alpha/alpha/pmap.c
index 235f54f2425..9f9f42a006c 100644
--- a/sys/arch/alpha/alpha/pmap.c
+++ b/sys/arch/alpha/alpha/pmap.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: pmap.c,v 1.76 2015/05/02 20:50:08 miod Exp $ */
+/* $OpenBSD: pmap.c,v 1.77 2015/06/23 19:50:48 miod Exp $ */
 /* $NetBSD: pmap.c,v 1.154 2000/12/07 22:18:55 thorpej Exp $ */
 
 /*-
@@ -386,7 +386,7 @@ struct pmap_tlb_shootdown_job {
 struct pmap_tlb_shootdown_q {
 	TAILQ_HEAD(, pmap_tlb_shootdown_job) pq_head;
 	TAILQ_HEAD(, pmap_tlb_shootdown_job) pq_free;
-	int pq_pte;			/* aggregate PTE bits */
+	int pq_pte;			/* aggregate low PTE bits */
 	int pq_tbia;			/* pending global flush */
 	struct mutex pq_mtx;		/* queue lock */
 	struct pmap_tlb_shootdown_job pq_jobs[PMAP_TLB_SHOOTDOWN_MAXJOBS];
@@ -2536,17 +2536,49 @@ pmap_emulate_reference(struct proc *p, vaddr_t v, int user, int type)
 #ifdef DEBUG				/* These checks are more expensive */
 	if (!pmap_pte_v(pte))
 		panic("pmap_emulate_reference: invalid pte");
+#ifndef MULTIPROCESSOR
+	/*
+	 * Quoting the Alpha ARM 14.3.1.4/5/6:
+	 * ``The Translation Buffer may reload and cache the old PTE value
+	 *   between the time the FOR (resp. FOW, FOE) fault invalidates the
+	 *   old value from the Translation Buffer and the time software
+	 *   updates the PTE in memory.  Software that depends on the
+	 *   processor-provided invalidate must thus be prepared to take
+	 *   another FOR (resp. FOW, FOE) fault on a page after clearing the
+	 *   page's PTE<FOR(resp. FOW, FOE)> bit. The second fault will
+	 *   invalidate the stale PTE from the Translation Buffer, and the
+	 *   processor cannot load another stale copy. Thus, in the worst case,
+	 *   a multiprocessor system will take an initial FOR (resp. FOW, FOE)
+	 *   fault and then an additional FOR (resp. FOW, FOE) fault on each
+	 *   processor. In practice, even a single repetition is unlikely.''
+	 *
+	 * In practice, spurious faults on the other processors happen, at
+	 * least on fast 21264 or better processors.
+	 */
 	if (type == ALPHA_MMCSR_FOW) {
-		if (!(*pte & (user ? PG_UWE : PG_UWE | PG_KWE)))
-			panic("pmap_emulate_reference: write but unwritable");
-		if (!(*pte & PG_FOW))
-			panic("pmap_emulate_reference: write but not FOW");
+		if (!(*pte & (user ? PG_UWE : PG_UWE | PG_KWE))) {
+			panic("pmap_emulate_reference(%d,%d): "
+			    "write but unwritable pte 0x%lx",
+			    user, type, *pte);
+		}
+		if (!(*pte & PG_FOW)) {
+			panic("pmap_emulate_reference(%d,%d): "
+			    "write but not FOW pte 0x%lx",
+			    user, type, *pte);
+		}
 	} else {
-		if (!(*pte & (user ? PG_URE : PG_URE | PG_KRE)))
-			panic("pmap_emulate_reference: !write but unreadable");
-		if (!(*pte & (PG_FOR | PG_FOE)))
-			panic("pmap_emulate_reference: !write but not FOR|FOE");
+		if (!(*pte & (user ? PG_URE : PG_URE | PG_KRE))) {
+			panic("pmap_emulate_reference(%d,%d): "
+			    "!write but unreadable pte 0x%lx",
+			    user, type, *pte);
+		}
+		if (!(*pte & (PG_FOR | PG_FOE))) {
+			panic("pmap_emulate_reference(%d,%d): "
+			    "!write but not FOR|FOE pte 0x%lx",
+			    user, type, *pte);
+		}
 	}
+#endif /* MULTIPROCESSOR */
 	/* Other diagnostics? */
 #endif
 	pa = pmap_pte_pa(pte);
@@ -2559,9 +2591,11 @@ pmap_emulate_reference(struct proc *p, vaddr_t v, int user, int type)
 	pg = PHYS_TO_VM_PAGE(pa);
 
 #ifdef DIAGNOSTIC
-	if (pg == NULL)
+	if (pg == NULL) {
 		panic("pmap_emulate_reference(%p, 0x%lx, %d, %d): "
-		    "pa 0x%lx not managed", p, v, user, type, pa);
+		    "pa 0x%lx (pte %p 0x%08lx) not managed",
+		    p, v, user, type, pa, pte, *pte);
+	}
 #endif
 
 	/*
@@ -3515,9 +3549,8 @@ pmap_do_tlb_shootdown(struct cpu_info *ci, struct trapframe *framep)
 			    pj->pj_pmap->pm_cpus & cpu_mask, cpu_id);
 			pmap_tlb_shootdown_job_put(pq, pj);
 		}
-
-		pq->pq_pte = 0;
 	}
+	pq->pq_pte = 0;
 
 	PSJQ_UNLOCK(pq, s);
 }
@@ -3540,7 +3573,6 @@ pmap_tlb_shootdown_q_drain(struct pmap_tlb_shootdown_q *pq)
 		TAILQ_REMOVE(&pq->pq_head, pj, pj_list);
 		pmap_tlb_shootdown_job_put(pq, pj);
 	}
-	pq->pq_pte = 0;
 }
 
 /*
-- 
2.20.1