add pmap_active_cpus()

For amd64, i386, arm, and riscv, i.e. all architectures except arm64,
the custom implementation is provided since we maintain the bitmask of
active CPUs anyway.

Arm64 uses somewhat naive iteration over CPUs and match current vmspace'
pmap with the argument. It is not guaranteed that vmspace->pmap is the
same as the active pmap, but the inaccuracy should be toleratable.

Sponsored by:	The FreeBSD Foundation
MFC after:	1 week
Differential revision:	https://reviews.freebsd.org/D32360

sys/amd64/amd64/pmap.c

@@ -10256,6 +10256,12 @@ pmap_activate_boot(pmap_t pmap)
 	PCPU_SET(ucr3, PMAP_NO_CR3);
 }
 
+void
+pmap_active_cpus(pmap_t pmap, cpuset_t *res)
+{
+	*res = pmap->pm_active;
+}
+
 void
 pmap_sync_icache(pmap_t pm, vm_offset_t va, vm_size_t sz)
 {

sys/arm/arm/pmap-v6.c

@@ -6194,6 +6194,12 @@ pmap_activate(struct thread *td)
 	critical_exit();
 }
 
+void
+pmap_active_cpus(pmap_t pmap, cpuset_t *res)
+{
+	*res = pmap->pm_active;
+}
+
 /*
  * Perform the pmap work for mincore(2).  If the page is not both referenced and
  * modified by this pmap, returns its physical address so that the caller can

sys/arm64/include/pmap.h

@@ -129,6 +129,8 @@ extern struct pmap	kernel_pmap_store;
 	(uint64_t)(asid) << TTBR_ASID_SHIFT;			\
 })
 
+#define	PMAP_WANT_ACTIVE_CPUS_NAIVE
+
 extern vm_offset_t virtual_avail;
 extern vm_offset_t virtual_end;
 

sys/i386/i386/pmap_base.c

@@ -944,6 +944,12 @@ pmap_kremove(vm_offset_t va)
 	pmap_methods_ptr->pm_kremove(va);
 }
 
+void
+pmap_active_cpus(pmap_t pmap, cpuset_t *res)
+{
+	*res = pmap->pm_active;
+}
+
 extern struct pmap_methods pmap_pae_methods, pmap_nopae_methods;
 int pae_mode;
 SYSCTL_INT(_vm_pmap, OID_AUTO, pae_mode, CTLFLAG_RDTUN | CTLFLAG_NOFETCH,

sys/powerpc/powerpc/pmap_dispatch.c

@@ -253,3 +253,9 @@ pmap_is_valid_memattr(pmap_t pmap __unused, vm_memattr_t mode)
 		return (FALSE);
 	}
 }
+
+void
+pmap_active_cpus(pmap_t pmap, cpuset_t *res)
+{
+	*res = pmap->pm_active;
+}

sys/riscv/riscv/pmap.c

@@ -4687,6 +4687,12 @@ pmap_activate_boot(pmap_t pmap)
 	PCPU_SET(curpmap, pmap);
 }
 
+void
+pmap_active_cpus(pmap_t pmap, cpuset_t *res)
+{
+	*res = pmap->pm_active;
+}
+
 void
 pmap_sync_icache(pmap_t pmap, vm_offset_t va, vm_size_t sz)
 {

sys/vm/pmap.h

@@ -90,6 +90,7 @@ typedef struct pmap_statistics *pmap_statistics_t;
 #include <machine/pmap.h>
 
 #ifdef _KERNEL
+#include <sys/_cpuset.h>
 struct thread;
 
 /*
@@ -118,6 +119,7 @@ extern vm_offset_t kernel_vm_end;
 #define	PMAP_TS_REFERENCED_MAX	5
 
 void		 pmap_activate(struct thread *td);
+void		 pmap_active_cpus(pmap_t pmap, cpuset_t *res);
 void		 pmap_advise(pmap_t pmap, vm_offset_t sva, vm_offset_t eva,
 		    int advice);
 void		 pmap_align_superpage(vm_object_t, vm_ooffset_t, vm_offset_t *,

sys/vm/vm_kern.c

@@ -78,6 +78,7 @@
 #include <sys/msan.h>
 #include <sys/proc.h>
 #include <sys/rwlock.h>
+#include <sys/smp.h>
 #include <sys/sysctl.h>
 #include <sys/vmem.h>
 #include <sys/vmmeter.h>
@@ -907,6 +908,31 @@ kmem_bootstrap_free(vm_offset_t start, vm_size_t size)
 #endif
 }
 
+#ifdef PMAP_WANT_ACTIVE_CPUS_NAIVE
+void
+pmap_active_cpus(pmap_t pmap, cpuset_t *res)
+{
+	struct thread *td;
+	struct proc *p;
+	struct vmspace *vm;
+	int c;
+
+	CPU_ZERO(res);
+	CPU_FOREACH(c) {
+		td = cpuid_to_pcpu[c]->pc_curthread;
+		p = td->td_proc;
+		if (p == NULL)
+			continue;
+		vm = vmspace_acquire_ref(p);
+		if (vm == NULL)
+			continue;
+		if (pmap == vmspace_pmap(vm))
+			CPU_SET(c, res);
+		vmspace_free(vm);
+	}
+}
+#endif
+
 /*
  * Allow userspace to directly trigger the VM drain routine for testing
  * purposes.