Machines have started to appear that have the framebuffer at an address > 4GB.

On these machines we can't use the direct map since early on during boot
the direct map only covers the first 4GB of memory.  Instead, use a
special (and temporary) mapping until we remap the framebuffer near the
start of autoconf.  With lots of help from mlarkin@

tested by yasuoka@
ok mlarkin@

5 files changed, 116 insertions, 11 deletions

diff --git a/sys/arch/amd64/amd64/efifb.c b/sys/arch/amd64/amd64/efifb.c
index 44c2478463e..69d6a01bad4 100644
--- a/sys/arch/amd64/amd64/efifb.c
+++ b/sys/arch/amd64/amd64/efifb.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: efifb.c,v 1.26 2019/11/26 02:20:50 jsg Exp $	*/
+/*	$OpenBSD: efifb.c,v 1.27 2020/01/24 05:27:31 kettenis Exp $	*/
 
 /*
  * Copyright (c) 2015 YASUOKA Masahiko <yasuoka@yasuoka.net>
@@ -105,6 +105,8 @@ int	 efifb_load_font(void *, void *, struct wsdisplay_font *);
 void	 efifb_scrollback(void *, void *, int lines);
 void	 efifb_efiinfo_init(struct efifb *);
 void	 efifb_cnattach_common(void);
+vaddr_t	 efifb_early_map(paddr_t);
+void	 efifb_early_cleanup(void);
 
 struct cb_framebuffer *cb_find_fb(paddr_t);
 
@@ -430,7 +432,7 @@ efifb_cnattach_common(void)
 	struct rasops_info	*ri = &fb->rinfo;
 	long			 defattr = 0;
 
-	ri->ri_bits = (u_char *)PMAP_DIRECT_MAP(fb->paddr);
+	ri->ri_bits = (u_char *)efifb_early_map(fb->paddr);
 
 	efifb_rasops_preinit(fb);
 
@@ -459,14 +461,17 @@ efifb_cnremap(void)
 		return;
 
 	if (_bus_space_map(iot, fb->paddr, fb->psize,
-	    BUS_SPACE_MAP_PREFETCHABLE | BUS_SPACE_MAP_LINEAR, &ioh) == 0)
-		ri->ri_origbits = bus_space_vaddr(iot, ioh);
+	    BUS_SPACE_MAP_PREFETCHABLE | BUS_SPACE_MAP_LINEAR, &ioh))
+		panic("can't remap framebuffer");
+	ri->ri_origbits = bus_space_vaddr(iot, ioh);
 
 	efifb_rasops_preinit(fb);
 	ri->ri_flg &= ~RI_CLEAR;
 	ri->ri_flg |= RI_CENTER | RI_WRONLY;
 
 	rasops_init(ri, efifb_std_descr.nrows, efifb_std_descr.ncols);
+
+	efifb_early_cleanup();
 }
 
 int
@@ -637,3 +642,15 @@ efifb_stolen(void)
 	struct efifb *fb = &efifb_console;
 	return fb->psize;
 }
+
+vaddr_t
+efifb_early_map(paddr_t pa)
+{
+	return pmap_set_pml4_early(pa);
+}
+
+void
+efifb_early_cleanup(void)
+{
+	pmap_clear_pml4_early();
+}
diff --git a/sys/arch/amd64/amd64/locore0.S b/sys/arch/amd64/amd64/locore0.S
index 3be32d5718d..62d31e89917 100644
--- a/sys/arch/amd64/amd64/locore0.S
+++ b/sys/arch/amd64/amd64/locore0.S
@@ -1,4 +1,4 @@
-/*	$OpenBSD: locore0.S,v 1.17 2019/02/18 08:26:20 yasuoka Exp $	*/
+/*	$OpenBSD: locore0.S,v 1.18 2020/01/24 05:27:31 kettenis Exp $	*/
 /*	$NetBSD: locore.S,v 1.13 2004/03/25 18:33:17 drochner Exp $	*/
 
 /*
@@ -321,7 +321,7 @@ cont:
 #define PROC0_DMP2_OFF	(PROC0_DMP3_OFF + NDML3_ENTRIES * NBPG)
 #define TABLESIZE \
     ((NKL4_KIMG_ENTRIES + TABLE_L3_ENTRIES + TABLE_L2_ENTRIES + 1 + UPAGES + \
-	NDML3_ENTRIES + NDML2_ENTRIES) * NBPG)
+	NDML3_ENTRIES + NDML2_ENTRIES + 3) * NBPG)
 
 #define fillkpt \
 1:	movl	%eax,(%ebx)	;	/* store phys addr */ \
@@ -669,10 +669,11 @@ longmode_hi:
 	movw	%ax,%gs
 	movw	%ax,%fs
 
-	/* XXX merge these */
 	leaq	TABLESIZE(%rsi),%rdi
-	call	_C_LABEL(init_x86_64)
+	subq	$(NBPG*3), %rdi
 
+	/* XXX merge these */
+	call	_C_LABEL(init_x86_64)
 	call	_C_LABEL(main)
 
 	.section .codepatch,"a"
diff --git a/sys/arch/amd64/amd64/machdep.c b/sys/arch/amd64/amd64/machdep.c
index f2d96f949ea..21c88098743 100644
--- a/sys/arch/amd64/amd64/machdep.c
+++ b/sys/arch/amd64/amd64/machdep.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: machdep.c,v 1.260 2019/12/20 07:49:31 jsg Exp $	*/
+/*	$OpenBSD: machdep.c,v 1.261 2020/01/24 05:27:31 kettenis Exp $	*/
 /*	$NetBSD: machdep.c,v 1.3 2003/05/07 22:58:18 fvdl Exp $	*/
 
 /*-
@@ -1365,6 +1365,8 @@ map_tramps(void)
 typedef void (vector)(void);
 extern vector *IDTVEC(exceptions)[];
 
+paddr_t early_pte_pages;
+
 void
 init_x86_64(paddr_t first_avail)
 {
@@ -1373,6 +1375,15 @@ init_x86_64(paddr_t first_avail)
 	int x, ist;
 	uint64_t max_dm_size = ((uint64_t)512 * NUM_L4_SLOT_DIRECT) << 30;
 
+	/*
+	 * locore0 mapped 3 pages for use before the pmap is initialized
+	 * starting at first_avail. These pages are currently used by
+	 * efifb to create early-use VAs for the framebuffer before efifb
+	 * is attached.
+	 */
+	early_pte_pages = first_avail;
+	first_avail += 3 * NBPG;
+
 	cpu_init_msrs(&cpu_info_primary);
 
 	proc0.p_addr = proc0paddr;
diff --git a/sys/arch/amd64/amd64/pmap.c b/sys/arch/amd64/amd64/pmap.c
index 5ca512aadda..42f25b12000 100644
--- a/sys/arch/amd64/amd64/pmap.c
+++ b/sys/arch/amd64/amd64/pmap.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pmap.c,v 1.137 2019/12/19 17:42:17 mpi Exp $	*/
+/*	$OpenBSD: pmap.c,v 1.138 2020/01/24 05:27:31 kettenis Exp $	*/
 /*	$NetBSD: pmap.c,v 1.3 2003/05/08 18:13:13 thorpej Exp $	*/
 
 /*
@@ -544,6 +544,78 @@ pmap_kremove(vaddr_t sva, vsize_t len)
 }
 
 /*
+ * pmap_set_pml4_early
+ *
+ * Utility function to map 2GB of 2MB pages to 'pa'. The VA that is assigned
+ * is the pml4 entry for 'early mappings' (see pmap.h). This function is used
+ * by display drivers that need to map their framebuffers early, before the
+ * pmap is fully initialized (eg, to show panic messages).
+ *
+ * Users of this function must call pmap_clear_pml4_early to remove the
+ * mapping when finished.
+ *
+ * Parameters:
+ *  pa: phys addr to map
+ *
+ * Return value:
+ *  VA mapping to 'pa'. This mapping is 2GB in size and starts at the base
+ *   of the 2MB region containing 'va'.
+ */
+vaddr_t
+pmap_set_pml4_early(paddr_t pa)
+{
+	extern paddr_t early_pte_pages;
+	pt_entry_t *pml4e, *pte;
+	int i, j, off;
+	paddr_t curpa;
+	vaddr_t va;
+
+	pml4e = (pt_entry_t *)(proc0.p_addr->u_pcb.pcb_cr3 + KERNBASE);
+	pml4e[PDIR_SLOT_EARLY] = (pd_entry_t)early_pte_pages | PG_V | PG_RW;
+
+	off = pa & PAGE_MASK_L2;
+	curpa = pa & L2_FRAME;
+
+	pte = (pt_entry_t *)PMAP_DIRECT_MAP(early_pte_pages);
+	memset(pte, 0, 3 * NBPG);
+
+	pte[0] = (early_pte_pages + NBPG) | PG_V | PG_RW;
+	pte[1] = (early_pte_pages + 2 * NBPG) | PG_V | PG_RW;
+
+	pte = (pt_entry_t *)PMAP_DIRECT_MAP(early_pte_pages + NBPG);
+	for (i = 0; i < 2; i++) {
+		/* 2 early pages of mappings */
+		for (j = 0; j < 512; j++) {
+			/* j[0..511] : 2MB mappings per page */
+			pte[(i * 512) + j] = curpa | PG_V | PG_RW | PG_PS;
+			curpa += (2 * 1024 * 1024);
+		}
+	}
+
+	va = (vaddr_t)((PDIR_SLOT_EARLY * 512ULL) << L3_SHIFT) + off;
+	return VA_SIGN_NEG(va);
+}
+
+/*
+ * pmap_clear_pml4_early
+ *
+ * Clears the mapping previously established with pmap_set_pml4_early.
+ */
+void
+pmap_clear_pml4_early(void)
+{
+	extern paddr_t early_pte_pages;
+	pt_entry_t *pml4e, *pte;
+
+	pte = (pt_entry_t *)PMAP_DIRECT_MAP(early_pte_pages);
+	memset(pte, 0, 3 * NBPG);
+
+	pml4e = (pd_entry_t *)pmap_kernel()->pm_pdir;
+	pml4e[PDIR_SLOT_EARLY] = 0;
+	tlbflush();
+}
+
+/*
  * p m a p   i n i t   f u n c t i o n s
  *
  * pmap_bootstrap and pmap_init are called during system startup
diff --git a/sys/arch/amd64/include/pmap.h b/sys/arch/amd64/include/pmap.h
index a2180a05b1b..bcce28080ac 100644
--- a/sys/arch/amd64/include/pmap.h
+++ b/sys/arch/amd64/include/pmap.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: pmap.h,v 1.76 2019/12/19 17:42:17 mpi Exp $	*/
+/*	$OpenBSD: pmap.h,v 1.77 2020/01/24 05:27:32 kettenis Exp $	*/
 /*	$NetBSD: pmap.h,v 1.1 2003/04/26 18:39:46 fvdl Exp $	*/
 
 /*
@@ -143,10 +143,12 @@
 #define L4_SLOT_KERNBASE	511
 #define NUM_L4_SLOT_DIRECT	4
 #define L4_SLOT_DIRECT		(L4_SLOT_KERNBASE - NUM_L4_SLOT_DIRECT)
+#define L4_SLOT_EARLY		(L4_SLOT_DIRECT - 1)
 
 #define PDIR_SLOT_KERN		L4_SLOT_KERN
 #define PDIR_SLOT_PTE		L4_SLOT_PTE
 #define PDIR_SLOT_DIRECT	L4_SLOT_DIRECT
+#define PDIR_SLOT_EARLY		L4_SLOT_EARLY
 
 /*
  * the following defines give the virtual addresses of various MMU
@@ -401,6 +403,8 @@ void	pagezero(vaddr_t);
 
 int	pmap_convert(struct pmap *, int);
 void	pmap_enter_special(vaddr_t, paddr_t, vm_prot_t);
+vaddr_t	pmap_set_pml4_early(paddr_t pa);
+void	pmap_clear_pml4_early(void);
 
 /*
  * functions for flushing the cache for vaddrs and pages.