Extent sgi bus_dma to cope with different views of memory: non-contiguous

for the cpu, contiguous from different bases for devices. This allows
memory above 256MB to be used with bus_dma (and we had really been lucky
with the first few large-memory builds).
Information about memory accesses taken from Linux.

7 files changed, 92 insertions, 26 deletions

diff --git a/sys/arch/mips64/conf/files.mips64 b/sys/arch/mips64/conf/files.mips64
index c70c94557b8..3c3ed75da03 100644
--- a/sys/arch/mips64/conf/files.mips64
+++ b/sys/arch/mips64/conf/files.mips64
@@ -1,4 +1,4 @@
-#	$OpenBSD: files.mips64,v 1.8 2007/06/18 21:46:41 deraadt Exp $
+#	$OpenBSD: files.mips64,v 1.9 2007/06/21 20:17:12 miod Exp $
 
 file	arch/mips64/mips64/mem.c
 file	arch/mips64/mips64/process_machdep.c
@@ -8,7 +8,6 @@ file	arch/mips64/mips64/vm_machdep.c
 file	arch/mips64/mips64/arcbios.c		arcbios
 file	arch/mips64/mips64/clock.c
 file	arch/mips64/mips64/cpu.c
-file	arch/mips64/mips64/busdma.c
 file	arch/mips64/mips64/interrupt.c
 file	arch/mips64/mips64/mainbus.c
 file	arch/mips64/mips64/pmap.c
diff --git a/sys/arch/sgi/conf/files.sgi b/sys/arch/sgi/conf/files.sgi
index d6404ef02f5..0b9fd20ada1 100644
--- a/sys/arch/sgi/conf/files.sgi
+++ b/sys/arch/sgi/conf/files.sgi
@@ -1,4 +1,4 @@
-#	$OpenBSD: files.sgi,v 1.12 2007/06/18 21:46:41 deraadt Exp $
+#	$OpenBSD: files.sgi,v 1.13 2007/06/21 20:17:12 miod Exp $
 #
 # maxpartitions must be first item in files.${ARCH}
 #
@@ -10,6 +10,7 @@ maxusers 2 8 64
 
 file	arch/sgi/sgi/disksubr.c			disk
 file	arch/sgi/sgi/autoconf.c
+file	arch/sgi/sgi/bus_dma.c
 file	arch/sgi/sgi/conf.c
 file	arch/sgi/sgi/machdep.c
 file	arch/sgi/sgi/mutex.c
diff --git a/sys/arch/sgi/include/bus.h b/sys/arch/sgi/include/bus.h
index 1d84ced752b..4e191878a71 100644
--- a/sys/arch/sgi/include/bus.h
+++ b/sys/arch/sgi/include/bus.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: bus.h,v 1.5 2006/05/27 00:29:55 krw Exp $	*/
+/*	$OpenBSD: bus.h,v 1.6 2007/06/21 20:17:12 miod Exp $	*/
 
 /*
  * Copyright (c) 2003-2004 Opsycon AB Sweden.  All rights reserved.
@@ -345,7 +345,13 @@ struct machine_bus_dma_tag {
 	void	(*_dmamem_unmap)(bus_dma_tag_t, caddr_t, size_t);
 	paddr_t	(*_dmamem_mmap)(bus_dma_tag_t, bus_dma_segment_t *,
 		    int, off_t, int, int);
-	paddr_t	dma_offs;
+
+	/*
+	 * internal memory address translation information.
+	 */
+	bus_addr_t (*_pa_to_device)(paddr_t);
+	paddr_t	(*_device_to_pa)(bus_addr_t);
+	bus_addr_t _dma_mask;
 };
 
 #define	bus_dmamap_create(t, s, n, m, b, f, p)			\
diff --git a/sys/arch/sgi/localbus/crimebus.h b/sys/arch/sgi/localbus/crimebus.h
index 8eb35227c9b..67f7f7b4719 100644
--- a/sys/arch/sgi/localbus/crimebus.h
+++ b/sys/arch/sgi/localbus/crimebus.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: crimebus.h,v 1.5 2007/05/03 19:34:00 miod Exp $	*/
+/*	$OpenBSD: crimebus.h,v 1.6 2007/06/21 20:17:12 miod Exp $	*/
 
 /*
  * Copyright (c) 2003-2004 Opsycon AB (www.opsycon.se).
@@ -100,6 +100,9 @@
 #define  CRIME_MEM_BANK_128MB	   0x100
 #define CRIME_MAX_BANKS		 8
 
+#define	CRIME_MEMORY_OFFSET	0x40000000	/* 1GB */
+#define	CRIME_MEMORY_MASK	0x3fffffff
+
 #define	CRIME_MEM_ERROR_STAT	 0x0250
 #define	CRIME_MEM_ERROR_ADDR	 0x0258
 
diff --git a/sys/arch/sgi/localbus/macebus.c b/sys/arch/sgi/localbus/macebus.c
index 93e50b71b8e..555275d380d 100644
--- a/sys/arch/sgi/localbus/macebus.c
+++ b/sys/arch/sgi/localbus/macebus.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: macebus.c,v 1.24 2007/06/20 20:47:34 miod Exp $ */
+/*	$OpenBSD: macebus.c,v 1.25 2007/06/21 20:17:12 miod Exp $ */
 
 /*
  * Copyright (c) 2000-2004 Opsycon AB  (www.opsycon.se)
@@ -69,6 +69,9 @@ void macebus_do_pending_int(int);
 intrmask_t macebus_iointr(intrmask_t, struct trap_frame *);
 intrmask_t macebus_aux(intrmask_t, struct trap_frame *);
 
+bus_addr_t macebus_pa_to_device(paddr_t);
+paddr_t	macebus_device_to_pa(bus_addr_t);
+
 int maceticks;		/* Time tracker for special events */
 
 struct cfattach macebus_ca = {
@@ -121,7 +124,9 @@ struct machine_bus_dma_tag mace_bus_dma_tag = {
 	_dmamem_map,
 	_dmamem_unmap,
 	_dmamem_mmap,
-	NULL
+	macebus_pa_to_device,
+	macebus_device_to_pa,
+	CRIME_MEMORY_MASK
 };
 
 /*
@@ -430,6 +435,28 @@ mace_space_region(bus_space_tag_t t, bus_space_handle_t bsh,
 }
 
 /*
+ * Macebus bus_dma helpers.
+ * Mace accesses memory contiguously at 0x40000000 onwards.
+ */
+
+bus_addr_t
+macebus_pa_to_device(paddr_t pa)
+{
+	return (pa | CRIME_MEMORY_OFFSET);
+}
+
+paddr_t
+macebus_device_to_pa(bus_addr_t addr)
+{
+	paddr_t pa = (paddr_t)addr & CRIME_MEMORY_MASK;
+
+	if (pa >= 256 * 1024 * 1024)
+		pa |= CRIME_MEMORY_OFFSET;
+
+	return (pa);
+}
+
+/*
  *  Macebus interrupt handler driver.
  */
 
diff --git a/sys/arch/sgi/pci/macepcibridge.c b/sys/arch/sgi/pci/macepcibridge.c
index ae9801c6e40..64830ea9f26 100644
--- a/sys/arch/sgi/pci/macepcibridge.c
+++ b/sys/arch/sgi/pci/macepcibridge.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: macepcibridge.c,v 1.11 2006/12/14 17:36:12 kettenis Exp $ */
+/*	$OpenBSD: macepcibridge.c,v 1.12 2007/06/21 20:17:12 miod Exp $ */
 
 /*
  * Copyright (c) 2001-2004 Opsycon AB (www.opsycon.se)
@@ -49,6 +49,7 @@
 #include <dev/pci/pcivar.h>
 
 #include <mips64/archtype.h>
+#include <sgi/localbus/crimebus.h>
 #include <sgi/localbus/macebus.h>
 #include <sgi/pci/macepcibrvar.h>
 
@@ -77,6 +78,8 @@ void     *mace_pcibr_intr_establish(void *, pci_intr_handle_t,
 	    int, int (*func)(void *), void *, char *);
 void     mace_pcibr_intr_disestablish(void *, void *);
 
+bus_addr_t mace_pcibr_pa_to_device(paddr_t);
+paddr_t	mace_pcibr_device_to_pa(bus_addr_t);
 
 struct cfattach macepcibr_ca = {
 	sizeof(struct mace_pcibr_softc), mace_pcibrmatch, mace_pcibrattach,
@@ -132,7 +135,9 @@ struct machine_bus_dma_tag pci_bus_dma_tag = {
 	_dmamem_map,
 	_dmamem_unmap,
 	_dmamem_mmap,
-	NULL
+	mace_pcibr_pa_to_device,
+	mace_pcibr_device_to_pa,
+	CRIME_MEMORY_MASK
 };
 
 struct _perr_map {
@@ -239,7 +244,6 @@ mace_pcibrattach(struct device *parent, struct device *self, void *aux)
 	pba.pba_iot = sc->sc_io_bus_space;
 	pba.pba_memt = sc->sc_mem_bus_space;
 	pba.pba_dmat = malloc(sizeof(pci_bus_dma_tag), M_DEVBUF, M_NOWAIT);
-	pci_bus_dma_tag.dma_offs = 0x00000000;
 	*pba.pba_dmat = pci_bus_dma_tag;
 	pba.pba_pc = &sc->sc_pc;
 	pba.pba_domain = pci_ndomains++;
@@ -563,3 +567,25 @@ mace_pcib_space_region(bus_space_tag_t t, bus_space_handle_t bsh,
 	*nbshp = bsh + offset;
 	return (0);
 }
+
+/*
+ * Mace PCI bus_dma helpers.
+ * The PCI bus accesses memory contiguously at 0x00000000 onwards.
+ */
+
+bus_addr_t
+mace_pcibr_pa_to_device(paddr_t pa)
+{
+	return (pa & CRIME_MEMORY_MASK);
+}
+
+paddr_t
+mace_pcibr_device_to_pa(bus_addr_t addr)
+{
+	paddr_t pa = (paddr_t)addr & CRIME_MEMORY_MASK;
+
+	if (pa >= 256 * 1024 * 1024)
+		pa |= CRIME_MEMORY_OFFSET;
+
+	return (pa);
+}
diff --git a/sys/arch/mips64/mips64/busdma.c b/sys/arch/sgi/sgi/bus_dma.c
index 426130d7884..0a0fd0faca6 100644
--- a/sys/arch/mips64/mips64/busdma.c
+++ b/sys/arch/sgi/sgi/bus_dma.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: busdma.c,v 1.13 2007/05/30 19:44:26 miod Exp $ */
+/*	$OpenBSD: bus_dma.c,v 1.1 2007/06/21 20:17:10 miod Exp $ */
 
 /*
  * Copyright (c) 2003-2004 Opsycon AB  (www.opsycon.se / www.opsycon.com)
@@ -135,6 +135,7 @@ _dmamap_load(t, map, buf, buflen, p, flags)
 
 	lastaddr = ~0;		/* XXX gcc */
 	bmask  = ~(map->_dm_boundary - 1);
+	bmask &= t->_dma_mask;
 
 	saved_buflen = buflen;
 	for (first = 1, seg = 0; buflen > 0; ) {
@@ -167,7 +168,8 @@ _dmamap_load(t, map, buf, buflen, p, flags)
 		 * previous segment if possible.
 		 */
 		if (first) {
-			map->dm_segs[seg].ds_addr = curaddr + t->dma_offs;
+			map->dm_segs[seg].ds_addr =
+			    (*t->_pa_to_device)(curaddr);
 			map->dm_segs[seg].ds_len = sgsize;
 			map->dm_segs[seg].ds_vaddr = (vaddr_t)vaddr;
 			first = 0;
@@ -176,13 +178,14 @@ _dmamap_load(t, map, buf, buflen, p, flags)
 			    (map->dm_segs[seg].ds_len + sgsize) <=
 			     map->_dm_maxsegsz &&
 			     (map->_dm_boundary == 0 ||
-			     ((map->dm_segs[seg].ds_addr - t->dma_offs) & bmask) ==
+			     (map->dm_segs[seg].ds_addr & bmask) ==
 			     (curaddr & bmask)))
 				map->dm_segs[seg].ds_len += sgsize;
 			else {
 				if (++seg >= map->_dm_segcnt)
 					break;
-				map->dm_segs[seg].ds_addr = curaddr + t->dma_offs;
+				map->dm_segs[seg].ds_addr =
+				    (*t->_pa_to_device)(curaddr);
 				map->dm_segs[seg].ds_len = sgsize;
 				map->dm_segs[seg].ds_vaddr = (vaddr_t)vaddr;
 			}
@@ -236,15 +239,15 @@ _dmamap_load_mbuf(t, map, m, flags)
 			if (pmap_extract(pmap_kernel(), vaddr, &pa) == FALSE)
 				panic("_dmamap_load_mbuf: pmap_extract(%x, %x) failed!",
 				    pmap_kernel(), vaddr);
-			pa += t->dma_offs;
 
-			if (i > 0 && pa == (map->dm_segs[i-1].ds_addr + map->dm_segs[i-1].ds_len)
+			if (i > 0 && pa == (*t->_device_to_pa)(map->dm_segs[i-1].ds_addr + map->dm_segs[i-1].ds_len)
 			    && ((map->dm_segs[i-1].ds_len + incr) < map->_dm_maxsegsz)) {
 				/* Hey, waddyaknow, they're contiguous */
 				map->dm_segs[i-1].ds_len += incr;
 				continue;
 			}
-			map->dm_segs[i].ds_addr = pa;
+			map->dm_segs[i].ds_addr =
+			    (*t->_pa_to_device)(pa);
 			map->dm_segs[i].ds_vaddr = vaddr;
 			map->dm_segs[i].ds_len = incr;
 			i++;
@@ -297,6 +300,7 @@ _dmamap_load_raw(t, map, segs, nsegs, size, flags)
 		bus_addr_t bmask = ~(map->_dm_boundary - 1);
 		int i;
 
+		bmask &= t->_dma_mask;
 		for (i = 0; i < nsegs; i++) {
 			if (segs[i].ds_len > map->_dm_maxsegsz)
 				return (EINVAL);
@@ -457,7 +461,7 @@ _dmamem_free(t, segs, nsegs)
 		for (addr = segs[curseg].ds_addr;
 		    addr < (segs[curseg].ds_addr + segs[curseg].ds_len);
 		    addr += PAGE_SIZE) {
-			m = PHYS_TO_VM_PAGE(addr - t->dma_offs);
+			m = PHYS_TO_VM_PAGE((*t->_device_to_pa)(addr));
 			TAILQ_INSERT_TAIL(&mlist, m, pageq);
 		}
 	}
@@ -495,14 +499,14 @@ _dmamem_map(t, segs, nsegs, size, kvap, flags)
 		    addr += NBPG, va += NBPG, size -= NBPG) {
 			if (size == 0)
 				panic("_dmamem_map: size botch");
-			pmap_enter(pmap_kernel(), va, addr - t->dma_offs,
+			pmap_enter(pmap_kernel(), va, (*t->_device_to_pa)(addr),
 			    VM_PROT_READ | VM_PROT_WRITE,
 			    VM_PROT_READ | VM_PROT_WRITE | PMAP_WIRED);
 			segs[curseg].ds_vaddr = va;
 
 			if (flags & BUS_DMA_COHERENT &&
 			    sys_config.system_type == SGI_O2) 
-				pmap_page_cache(PHYS_TO_VM_PAGE(addr - t->dma_offs), PV_UNCACHED);
+				pmap_page_cache(PHYS_TO_VM_PAGE((*t->_device_to_pa)(addr)), PV_UNCACHED);
 		}
 	}
 
@@ -558,7 +562,7 @@ _dmamem_mmap(t, segs, nsegs, off, prot, flags)
 			continue;
 		}
 
-		return (atop(segs[i].ds_addr + off - t->dma_offs));
+		return (atop((*t->_device_to_pa)(segs[i].ds_addr) + off));
 	}
 
 	/* Page not found. */
@@ -608,13 +612,13 @@ _dmamem_alloc_range(t, size, alignment, boundary, segs, nsegs, rsegs,
 	 */
 	m = TAILQ_FIRST(&mlist);
 	curseg = 0;
-	lastaddr = segs[curseg].ds_addr = VM_PAGE_TO_PHYS(m);
-	segs[curseg].ds_addr += t->dma_offs;
+	lastaddr = segs[curseg].ds_addr =
+	    (*t->_pa_to_device)(VM_PAGE_TO_PHYS(m));
 	segs[curseg].ds_len = PAGE_SIZE;
 	m = TAILQ_NEXT(m, pageq);
 
 	for (; m != TAILQ_END(&mlist); m = TAILQ_NEXT(m, pageq)) {
-		curaddr = VM_PAGE_TO_PHYS(m);
+		curaddr = (*t->_pa_to_device)(VM_PAGE_TO_PHYS(m));
 #ifdef DIAGNOSTIC
 		if (curaddr < low || curaddr >= high) {
 			printf("vm_page_alloc_memory returned non-sensical"
@@ -626,7 +630,7 @@ _dmamem_alloc_range(t, size, alignment, boundary, segs, nsegs, rsegs,
 			segs[curseg].ds_len += PAGE_SIZE;
 		else {
 			curseg++;
-			segs[curseg].ds_addr = curaddr + t->dma_offs;
+			segs[curseg].ds_addr = curaddr;
 			segs[curseg].ds_len = PAGE_SIZE;
 		}
 		lastaddr = curaddr;