From b2c7a8d84a0075fea104009115d0c34bffc39e06 Mon Sep 17 00:00:00 2001
From: cinap_lenrek <cinap_lenrek@felloff.net>
Date: Fri, 28 Jun 2019 18:12:13 +0200
Subject: [PATCH] pc64: preallocate mmupool page tables

preallocate 2% of user pages for page tables and MMU structures
and keep them mapped in the VMAP range. this leaves more space
in the KZERO window and avoids running out of kernel memory on
machines with large amounts of memory.
---
 sys/src/9/pc64/fns.h  |  1 +
 sys/src/9/pc64/main.c | 39 --------------------------
 sys/src/9/pc64/mmu.c  | 65 ++++++++++++++++++++++++++++++++++++++++++-
 3 files changed, 65 insertions(+), 40 deletions(-)

diff --git a/sys/src/9/pc64/fns.h b/sys/src/9/pc64/fns.h
index e993e60b1..d0d585ef3 100644
--- a/sys/src/9/pc64/fns.h
+++ b/sys/src/9/pc64/fns.h
@@ -157,6 +157,7 @@ void	pcmspecialclose(int);
 void	(*_pcmspecialclose)(int);
 void	pcmunmap(int, PCMmap*);
 void	pmap(uintptr *, uintptr, uintptr, vlong);
+void	preallocpages(void);
 void	procrestore(Proc*);
 void	procsave(Proc*);
 void	procsetup(Proc*);
diff --git a/sys/src/9/pc64/main.c b/sys/src/9/pc64/main.c
index e61840ea0..c8e705f49 100644
--- a/sys/src/9/pc64/main.c
+++ b/sys/src/9/pc64/main.c
@@ -103,45 +103,6 @@ confinit(void)
 	}
 }
 
-/*
- * The palloc.pages array can be a large chunk out of the 2GB
- * window above KZERO, so we allocate the array from
- * upages and map in the VMAP window before pageinit()
- */
-static void
-preallocpages(void)
-{
-	Pallocmem *pm;
-	uintptr va, base, top;
-	vlong size;
-	ulong np;
-	int i;
-
-	np = 0;
-	for(i=0; i<nelem(palloc.mem); i++){
-		pm = &palloc.mem[i];
-		np += pm->npage;
-	}
-	size = (uvlong)np * BY2PG;
-	size += sizeof(Page) + BY2PG;	/* round up */
-	size = (size / (sizeof(Page) + BY2PG)) * sizeof(Page);
-	size = ROUND(size, PGLSZ(1));
-
-	for(i=0; i<nelem(palloc.mem); i++){
-		pm = &palloc.mem[i];
-		base = ROUND(pm->base, PGLSZ(1));
-		top = pm->base + (uvlong)pm->npage * BY2PG;
-		if((base + size) <= VMAPSIZE && (vlong)(top - base) >= size){
-			va = base + VMAP;
-			pmap(m->pml4, base | PTEGLOBAL|PTEWRITE|PTEVALID, va, size);
-			palloc.pages = (Page*)va;
-			pm->base = base + size;
-			pm->npage = (top - pm->base)/BY2PG;
-			break;
-		}
-	}
-}
-
 void
 machinit(void)
 {
diff --git a/sys/src/9/pc64/mmu.c b/sys/src/9/pc64/mmu.c
index 9261354e8..dfb165ef5 100644
--- a/sys/src/9/pc64/mmu.c
+++ b/sys/src/9/pc64/mmu.c
@@ -271,7 +271,11 @@ mmuwalk(uintptr* table, uintptr va, int level, int create)
 		if(pte & PTEVALID){
 			if(pte & PTESIZE)
 				return 0;
-			table = KADDR(PPN(pte));
+			pte = PPN(pte);
+			if(pte >= (uintptr)-KZERO)
+				table = (void*)(pte + VMAP);
+			else
+				table = (void*)(pte + KZERO);
 		} else {
 			if(!create)
 				return 0;
@@ -570,3 +574,62 @@ patwc(void *a, int n)
 		*pte = (*pte & ~mask) | (attr & mask);
 	}
 }
+
+/*
+ * The palloc.pages array and mmupool can be a large chunk
+ * out of the 2GB window above KZERO, so we allocate from
+ * upages and map in the VMAP window before pageinit()
+ */
+void
+preallocpages(void)
+{
+	Pallocmem *pm;
+	uintptr va, base, top;
+	vlong tsize, psize;
+	ulong np, nt;
+	int i;
+
+	np = 0;
+	for(i=0; i<nelem(palloc.mem); i++){
+		pm = &palloc.mem[i];
+		np += pm->npage;
+	}
+	nt = np / 50;	/* 2% for mmupool */
+	np -= nt;
+
+	nt = (uvlong)nt*BY2PG / (sizeof(MMU)+PTSZ);
+	tsize = (uvlong)nt * (sizeof(MMU)+PTSZ);
+
+	psize = (uvlong)np * BY2PG;
+	psize += sizeof(Page) + BY2PG;
+	psize = (psize / (sizeof(Page)+BY2PG)) * sizeof(Page);
+
+	psize += tsize;
+	psize = ROUND(psize, PGLSZ(1));
+
+	for(i=0; i<nelem(palloc.mem); i++){
+		pm = &palloc.mem[i];
+		base = ROUND(pm->base, PGLSZ(1));
+		top = pm->base + (uvlong)pm->npage * BY2PG;
+		if((base + psize) <= VMAPSIZE && (vlong)(top - base) >= psize){
+			pm->base = base + psize;
+			pm->npage = (top - pm->base)/BY2PG;
+
+			va = base + VMAP;
+			pmap(m->pml4, base | PTEGLOBAL|PTEWRITE|PTEVALID, va, psize);
+
+			palloc.pages = (void*)(va + tsize);
+
+			mmupool.nfree = mmupool.nalloc = nt;
+			mmupool.free = (void*)(va + (uvlong)nt*PTSZ);
+			for(i=0; i<nt; i++){
+				mmupool.free[i].page = (uintptr*)va;
+				mmupool.free[i].next = &mmupool.free[i+1];
+				va += PTSZ;
+			}
+			mmupool.free[i-1].next = nil;
+
+			break;
+		}
+	}
+}