i386: Always bounce DMA requests above 4G for !PAE kernels

i386 kernels without 'options PAE' will still use PAE page tables if
the CPU supports PAE both to support larger amounts of RAM and for
PG_NX permissions.  However, to avoid changing the API, bus_addr_t and
related constants (e.g. BUS_SPACE_MAXADDR) are still limited to
32 bits.

To cope with this, the x86 bus_dma code included an extra check to
bounce requests for addresses above BUS_SPACE_MAXADDR.  This check was
elided (probably because it looks always-true on its face and had no
comment explaining its purpose) in recent refactoring.  To fix,
restore a custom address-validation function for i386 kernels without
options PAE that includes this check.

Reported by:	ci.freebsd.org
Reviewed by:	markj
Fixes:		3933ff5 busdma: tidy bus_dma_run_filter() functions
Differential Revision:	https://reviews.freebsd.org/D43277

sys/x86/x86/busdma_bounce.c

@@ -117,6 +117,23 @@ static MALLOC_DEFINE(M_BUSDMA, "busdma", "busdma metadata");
 
 #include "../../kern/subr_busdma_bounce.c"
 
+/*
+ * On i386 kernels without 'options PAE' we need to also bounce any
+ * physical addresses above 4G.
+ *
+ * NB: vm_paddr_t is required here since bus_addr_t is only 32 bits in
+ * i386 kernels without 'options PAE'.
+ */
+static __inline bool
+must_bounce(bus_dma_tag_t dmat, vm_paddr_t paddr)
+{
+#if defined(__i386__) && !defined(PAE)
+	if (paddr > BUS_SPACE_MAXADDR)
+		return (true);
+#endif
+	return (addr_needs_bounce(dmat, paddr));
+}
+
 static int
 bounce_bus_dma_zone_setup(bus_dma_tag_t dmat)
 {
@@ -491,7 +508,7 @@ _bus_dmamap_pagesneeded(bus_dma_tag_t dmat, vm_paddr_t buf, bus_size_t buflen,
 	curaddr = buf;
 	while (buflen != 0) {
 		sgsize = MIN(buflen, dmat->common.maxsegsz);
-		if (addr_needs_bounce(dmat, curaddr)) {
+		if (must_bounce(dmat, curaddr)) {
 			sgsize = MIN(sgsize,
 			    PAGE_SIZE - (curaddr & PAGE_MASK));
 			if (pagesneeded == NULL)
@@ -547,7 +564,7 @@ _bus_dmamap_count_pages(bus_dma_tag_t dmat, bus_dmamap_t map, pmap_t pmap,
 				paddr = pmap_kextract(vaddr);
 			else
 				paddr = pmap_extract(pmap, vaddr);
-			if (addr_needs_bounce(dmat, paddr)) {
+			if (must_bounce(dmat, paddr)) {
 				sg_len = roundup2(sg_len,
 				    dmat->common.alignment);
 				map->pagesneeded++;
@@ -584,7 +601,7 @@ _bus_dmamap_count_ma(bus_dma_tag_t dmat, bus_dmamap_t map, struct vm_page **ma,
 			sg_len = PAGE_SIZE - ma_offs;
 			max_sgsize = MIN(buflen, dmat->common.maxsegsz);
 			sg_len = MIN(sg_len, max_sgsize);
-			if (addr_needs_bounce(dmat, paddr)) {
+			if (must_bounce(dmat, paddr)) {
 				sg_len = roundup2(sg_len,
 				    dmat->common.alignment);
 				sg_len = MIN(sg_len, max_sgsize);
@@ -685,7 +702,7 @@ bounce_bus_dmamap_load_phys(bus_dma_tag_t dmat, bus_dmamap_t map,
 		sgsize = MIN(buflen, dmat->common.maxsegsz);
 		if ((dmat->bounce_flags & BUS_DMA_COULD_BOUNCE) != 0 &&
 		    map->pagesneeded != 0 &&
-		    addr_needs_bounce(dmat, curaddr)) {
+		    must_bounce(dmat, curaddr)) {
 			sgsize = MIN(sgsize, PAGE_SIZE - (curaddr & PAGE_MASK));
 			curaddr = add_bounce_page(dmat, map, 0, curaddr, 0,
 			    sgsize);
@@ -753,7 +770,7 @@ bounce_bus_dmamap_load_buffer(bus_dma_tag_t dmat, bus_dmamap_t map, void *buf,
 		sgsize = PAGE_SIZE - (curaddr & PAGE_MASK);
 		if ((dmat->bounce_flags & BUS_DMA_COULD_BOUNCE) != 0 &&
 		    map->pagesneeded != 0 &&
-		    addr_needs_bounce(dmat, curaddr)) {
+		    must_bounce(dmat, curaddr)) {
 			sgsize = roundup2(sgsize, dmat->common.alignment);
 			sgsize = MIN(sgsize, max_sgsize);
 			curaddr = add_bounce_page(dmat, map, kvaddr, curaddr, 0,
@@ -820,7 +837,7 @@ bounce_bus_dmamap_load_ma(bus_dma_tag_t dmat, bus_dmamap_t map,
 		sgsize = PAGE_SIZE - ma_offs;
 		if ((dmat->bounce_flags & BUS_DMA_COULD_BOUNCE) != 0 &&
 		    map->pagesneeded != 0 &&
-		    addr_needs_bounce(dmat, paddr)) {
+		    must_bounce(dmat, paddr)) {
 			sgsize = roundup2(sgsize, dmat->common.alignment);
 			sgsize = MIN(sgsize, max_sgsize);
 			KASSERT(vm_addr_align_ok(sgsize,