From c62f205818c55b47faab29eafdcc5e930a2aa4ff Mon Sep 17 00:00:00 2001
From: jhibbits <jhibbits@FreeBSD.org>
Date: Tue, 25 Feb 2020 01:40:22 +0000
Subject: [PATCH] powerpc/booke: Use a pseudo-DMAP for the device mappings on
 booke64

Since powerpc64 has such a large virtual address space, significantly larger
than its physical address space, take advantage of this, and create yet
another DMAP-like instance for the device mappings.  In this case, the
device mapping "DMAP" is in the 0x8000000000000000 - 0xc000000000000000
range, so as not to overlap the physical memory DMAP.

This will allow us to add TLB1 entry coalescing in the future, especially
useful for things like the radeonkms driver, which maps parts of the GPU at
a time, but eventually maps all of it, using up a lot of TLB1 entries (~40).
---
 sys/powerpc/booke/pmap.c | 39 ++++++++++++++++++++++++++-------------
 1 file changed, 26 insertions(+), 13 deletions(-)

diff --git a/sys/powerpc/booke/pmap.c b/sys/powerpc/booke/pmap.c
index 5abb19102a2b..2bcb33c77c2a 100644
--- a/sys/powerpc/booke/pmap.c
+++ b/sys/powerpc/booke/pmap.c
@@ -221,7 +221,22 @@ uint32_t tlb1_entries;
 
 #define TLB1_ENTRIES (tlb1_entries)
 
-static vm_offset_t tlb1_map_base = (vm_offset_t)VM_MAXUSER_ADDRESS + PAGE_SIZE;
+/*
+ * Base of the pmap_mapdev() region.  On 32-bit it immediately follows the
+ * userspace address range.  On On 64-bit it's far above, at (1 << 63), and
+ * ranges up to the DMAP, giving 62 bits of PA allowed.  This is far larger than
+ * the widest Book-E address bus, the e6500 has a 40-bit PA space.  This allows
+ * us to map akin to the DMAP, with addresses identical to the PA, offset by the
+ * base.
+ */
+#ifdef __powerpc64__
+#define	VM_MAPDEV_BASE		0x8000000000000000
+#define	VM_MAPDEV_PA_MAX	0x4000000000000000 /* Don't encroach on DMAP */
+#else
+#define	VM_MAPDEV_BASE	(VM_MAXUSER_ADDRESS + PAGE_SIZE)
+#endif
+
+static vm_offset_t tlb1_map_base = VM_MAPDEV_BASE;
 
 static tlbtid_t tid_alloc(struct pmap *);
 static void tid_flush(tlbtid_t tid);
@@ -3475,8 +3490,10 @@ mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
 {
 	tlb_entry_t e;
 	vm_paddr_t tmppa;
-	void *res;
-	uintptr_t va, tmpva;
+#ifndef __powerpc64__
+	uintptr_t tmpva;
+#endif
+	uintptr_t va;
 	vm_size_t sz;
 	int i;
 	int wimge;
@@ -3512,6 +3529,11 @@ mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
 
 	size = roundup(size, PAGE_SIZE);
 
+#ifdef __powerpc64__
+	KASSERT(pa < VM_MAPDEV_PA_MAX,
+	    ("Unsupported physical address! %lx", pa));
+	va = VM_MAPDEV_BASE + pa;
+#else
 	/*
 	 * The device mapping area is between VM_MAXUSER_ADDRESS and
 	 * VM_MIN_KERNEL_ADDRESS.  This gives 1GB of device addressing.
@@ -3534,24 +3556,15 @@ mmu_booke_mapdev_attr(mmu_t mmu, vm_paddr_t pa, vm_size_t size, vm_memattr_t ma)
 	    sz = ffsl((~((1 << flsl(size-1)) - 1)) & pa);
 	    sz = sz ? min(roundup(sz + 3, 4), flsl(size) - 1) : flsl(size) - 1;
 	    va = roundup(tlb1_map_base, 1 << sz) | (((1 << sz) - 1) & pa);
-#ifdef __powerpc64__
-	} while (!atomic_cmpset_long(&tlb1_map_base, tmpva, va + size));
-#else
 	} while (!atomic_cmpset_int(&tlb1_map_base, tmpva, va + size));
-#endif
-#else
-#ifdef __powerpc64__
-	va = atomic_fetchadd_long(&tlb1_map_base, size);
-#else
 	va = atomic_fetchadd_int(&tlb1_map_base, size);
 #endif
 #endif
-	res = (void *)va;
 
 	if (tlb1_mapin_region(va, pa, size, tlb_calc_wimg(pa, ma)) != size)
 		return (NULL);
 
-	return (res);
+	return ((void *)va);
 }
 
 /*