kernel: support large 64-bit pci membars, increase pc64 VMAPSIZE to 1TB

This makes vmap()/vunmap() take a vlong size argument,
and change the type of Pci.mem[].size to vlong as well.

Even if vmap() wont support large mappings, it is nice to
get the original unruncated value for error checking.

pc64 needs a bigger VMAP window, as system76 pangolin
puts the framebuffer at a physical address > 512GB.

4 files changed, 17 insertions, 14 deletions

diff --git a/sys/src/9/pc64/fns.h b/sys/src/9/pc64/fns.h
index b58c7e3a4..687ec50af 100644
--- a/sys/src/9/pc64/fns.h
+++ b/sys/src/9/pc64/fns.h
@@ -162,14 +162,14 @@ int	tas(void*);
 uvlong	tscticks(uvlong*);
 ulong	umballoc(ulong, ulong, ulong);
 void	umbfree(ulong, ulong);
-uvlong	upaalloc(uvlong, ulong, ulong);
-uvlong	upaallocwin(uvlong, ulong, ulong, ulong);
-void	upafree(uvlong, ulong);
+uvlong	upaalloc(uvlong, uvlong, uvlong);
+uvlong	upaallocwin(uvlong, uvlong, uvlong, uvlong);
+void	upafree(uvlong, uvlong);
 void	vectortable(void);
 void	vmxprocrestore(Proc *);
 void	vmxshutdown(void);
-void*	vmap(uvlong, int);
-void	vunmap(void*, int);
+void*	vmap(uvlong, vlong);
+void	vunmap(void*, vlong);
 void	wbinvd(void);
 void	writeconf(void);
 int	wrmsr(int, vlong);
diff --git a/sys/src/9/pc64/mem.h b/sys/src/9/pc64/mem.h
index 4541a62a0..dc28517a2 100644
--- a/sys/src/9/pc64/mem.h
+++ b/sys/src/9/pc64/mem.h
@@ -53,10 +53,10 @@
 #define KZERO		(0xffffffff80000000ull)
 #define KTZERO		(KZERO+1*MiB+64*KiB)
 
-#define VMAP		(0xffffff0000000000ull)
-#define VMAPSIZE	(512ull*GiB)
+#define VMAP		(0xfffffe8000000000ull)
+#define VMAPSIZE	(1024ull*GiB)
 
-#define	KMAP		(0xfffffe8000000000ull)
+#define	KMAP		(0xfffffe0000000000ull)
 #define KMAPSIZE	(2*MiB)
 
 /*
diff --git a/sys/src/9/pc64/mmu.c b/sys/src/9/pc64/mmu.c
index 1f6560bfe..7f33ccc5a 100644
--- a/sys/src/9/pc64/mmu.c
+++ b/sys/src/9/pc64/mmu.c
@@ -584,13 +584,13 @@ kunmap(KMap *k)
  * synchronization is being done.
  */
 void*
-vmap(uvlong pa, int size)
+vmap(uvlong pa, vlong size)
 {
 	uintptr va;
 	int o;
 
 	if(pa < BY2PG || size <= 0 || -pa < size || pa+size > VMAPSIZE){
-		print("vmap pa=%llux size=%d pc=%#p\n", pa, size, getcallerpc(&pa));
+		print("vmap pa=%llux size=%lld pc=%#p\n", pa, size, getcallerpc(&pa));
 		return nil;
 	}
 	va = pa+VMAP;
@@ -607,7 +607,7 @@ vmap(uvlong pa, int size)
 }
 
 void
-vunmap(void *v, int)
+vunmap(void *v, vlong)
 {
 	paddr(v);	/* will panic on error */
 }
diff --git a/sys/src/9/pc64/squidboy.c b/sys/src/9/pc64/squidboy.c
index f7882820b..ea0f9a10a 100644
--- a/sys/src/9/pc64/squidboy.c
+++ b/sys/src/9/pc64/squidboy.c
@@ -32,7 +32,7 @@ squidboy(Apic* apic)
 void
 mpstartap(Apic* apic)
 {
-	uintptr *apbootp, *pml4, *pdp0;
+	uintptr *apbootp, *pml4, *pdp0, v;
 	Segdesc *gdt;
 	Mach *mach;
 	uchar *p;
@@ -65,8 +65,11 @@ mpstartap(Apic* apic)
 	 * map KZERO (note that we share the KZERO (and VMAP)
 	 * PDP between processors.
 	 */
-	pml4[PTLX(KZERO, 3)] = MACHP(0)->pml4[PTLX(KZERO, 3)];
-	pml4[PTLX(VMAP, 3)] = MACHP(0)->pml4[PTLX(VMAP, 3)];
+	*mmuwalk(pml4, KZERO, 3, 0) = *mmuwalk(m->pml4, KZERO, 3, 0);
+	for(v = VMAP; v < VMAP+VMAPSIZE; v += PGLSZ(3)){
+		mmuwalk(m->pml4, v, 2, 1);	/* force create */
+		*mmuwalk(pml4, v, 3, 0) = *mmuwalk(m->pml4, v, 3, 0);
+	}
 
 	/* double map */
 	pml4[0] = PADDR(pdp0) | PTEWRITE|PTEVALID;