Turns out the cores on Apple's M1 SoC only support 8-bit ASIDs.

Thank you Apple (not)!

Add an initial attempt to support such systems.  This isn't good enough
since the kernel will hang once you create more than 127 processes.
But it makes things work reasonably well until you reach that limit
which is good enough to build things on the machine itself.

ok patrick@

diff --git a/sys/arch/arm64/arm64/pmap.c b/sys/arch/arm64/arm64/pmap.c
index f224fae..7e5c9d0 100644 (file)
--- a/sys/arch/arm64/arm64/pmap.c
+++ b/sys/arch/arm64/arm64/pmap.c
@@ -1,4 +1,4 @@
-/* $OpenBSD: pmap.c,v 1.71 2021/02/16 12:33:22 kettenis Exp $ */
+/* $OpenBSD: pmap.c,v 1.72 2021/03/04 18:36:52 kettenis Exp $ */
 /*
  * Copyright (c) 2008-2009,2014-2016 Dale Rahn <drahn@dalerahn.com>
  *
@@ -35,6 +35,7 @@
 #include <ddb/db_output.h>
 
 void pmap_setttb(struct proc *p);
+void pmap_allocate_asid(pmap_t);
 void pmap_free_asid(pmap_t pm);
 
 /* We run userland code with ASIDs that have the low bit set. */
@@ -82,8 +83,6 @@ void pmap_remove_pv(struct pte_desc *pted);
 void _pmap_kenter_pa(vaddr_t va, paddr_t pa, vm_prot_t prot, int flags,
     int cache);
 
-void pmap_allocate_asid(pmap_t);
-
 struct pmapvp0 {
        uint64_t l0[VP_IDX0_CNT];
        struct pmapvp1 *vp[VP_IDX0_CNT];
@@ -224,6 +223,56 @@ const uint64_t ap_bits_kern[8] = {
        [PROT_EXEC|PROT_WRITE|PROT_READ]        = ATTR_UXN|ATTR_AF|ATTR_AP(0),
 };
 
+/*
+ * We allocate ASIDs in pairs.  The first ASID is used to run the
+ * kernel and has both userland and the full kernel mapped.  The
+ * second ASID is used for running userland and has only the
+ * trampoline page mapped in addition to userland.
+ */
+
+#define MAX_NASID (1 << 16)
+uint32_t pmap_asid[MAX_NASID / 32];
+int pmap_nasid = (1 << 8);
+
+void
+pmap_allocate_asid(pmap_t pm)
+{
+       uint32_t bits;
+       int asid, bit;
+
+       for (;;) {
+               do {
+                       asid = arc4random() & (pmap_nasid - 2);
+                       bit = (asid & (32 - 1));
+                       bits = pmap_asid[asid / 32];
+               } while (asid == 0 || (bits & (3U << bit)));
+
+               if (atomic_cas_uint(&pmap_asid[asid / 32], bits,
+                   bits | (3U << bit)) == bits)
+                       break;
+       }
+       pm->pm_asid = asid;
+}
+
+void
+pmap_free_asid(pmap_t pm)
+{
+       uint32_t bits;
+       int bit;
+
+       KASSERT(pm != curcpu()->ci_curpm);
+       cpu_tlb_flush_asid_all((uint64_t)pm->pm_asid << 48);
+       cpu_tlb_flush_asid_all((uint64_t)(pm->pm_asid | ASID_USER) << 48);
+
+       bit = (pm->pm_asid & (32 - 1));
+       for (;;) {
+               bits = pmap_asid[pm->pm_asid / 32];
+               if (atomic_cas_uint(&pmap_asid[pm->pm_asid / 32], bits,
+                   bits & ~(3U << bit)) == bits)
+                       break;
+       }
+}
+
 /*
  * This is used for pmap_kernel() mappings, they are not to be removed
  * from the vp table because they were statically initialized at the
@@ -1158,6 +1207,7 @@ pmap_bootstrap(long kvo, paddr_t lpt1, long kernelstart, long kernelend,
        struct pmapvp3 *vp3;
        struct pte_desc *pted;
        vaddr_t vstart;
+       uint64_t id_aa64mmfr0;
        int i, j, k;
        int lb_idx2, ub_idx2;
 
@@ -1316,6 +1366,10 @@ pmap_bootstrap(long kvo, paddr_t lpt1, long kernelstart, long kernelend,
 
        curcpu()->ci_curpm = pmap_kernel();
 
+       id_aa64mmfr0 = READ_SPECIALREG(id_aa64mmfr0_el1);
+       if (ID_AA64MMFR0_ASID_BITS(id_aa64mmfr0) == ID_AA64MMFR0_ASID_BITS_16)
+               pmap_nasid = (1 << 16);
+
        vmmap = vstart;
        vstart += PAGE_SIZE;
 
@@ -2240,55 +2294,6 @@ pmap_map_early(paddr_t spa, psize_t len)
        __asm volatile("isb");
 }
 
-/*
- * We allocate ASIDs in pairs.  The first ASID is used to run the
- * kernel and has both userland and the full kernel mapped.  The
- * second ASID is used for running userland and has only the
- * trampoline page mapped in addition to userland.
- */
-
-#define NUM_ASID (1 << 16)
-uint32_t pmap_asid[NUM_ASID / 32];
-
-void
-pmap_allocate_asid(pmap_t pm)
-{
-       uint32_t bits;
-       int asid, bit;
-
-       for (;;) {
-               do {
-                       asid = arc4random() & (NUM_ASID - 2);
-                       bit = (asid & (32 - 1));
-                       bits = pmap_asid[asid / 32];
-               } while (asid == 0 || (bits & (3U << bit)));
-
-               if (atomic_cas_uint(&pmap_asid[asid / 32], bits,
-                   bits | (3U << bit)) == bits)
-                       break;
-       }
-       pm->pm_asid = asid;
-}
-
-void
-pmap_free_asid(pmap_t pm)
-{
-       uint32_t bits;
-       int bit;
-
-       KASSERT(pm != curcpu()->ci_curpm);
-       cpu_tlb_flush_asid_all((uint64_t)pm->pm_asid << 48);
-       cpu_tlb_flush_asid_all((uint64_t)(pm->pm_asid | ASID_USER) << 48);
-
-       bit = (pm->pm_asid & (32 - 1));
-       for (;;) {
-               bits = pmap_asid[pm->pm_asid / 32];
-               if (atomic_cas_uint(&pmap_asid[pm->pm_asid / 32], bits,
-                   bits & ~(3U << bit)) == bits)
-                       break;
-       }
-}
-
 void
 pmap_setttb(struct proc *p)
 {