kvm tools: Generate SPAPR PPC64 guest device tree

The generated DT is the bare minimum structure required for SPAPR (on which
subsequent patches for VIO, XICS, PCI etc. will build); root node, cpus, memory.

The DT contains CPU-specific configuration; a very simple 'cpu info' mechanism
is added to recognise/differentiate DT entries for POWER7 and PPC970 host CPUs.
Future support of more CPUs is possible.

libfdt is included from scripts/dtc/libfdt.

Signed-off-by: Matt Evans <matt@ozlabs.org>
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Pekka Enberg <penberg@kernel.org>

Makefile

@@ -105,6 +105,8 @@ ifeq ($(uname_M),x86_64)
 	DEFINES      += -DCONFIG_X86_64
 endif
 
+LIBFDT_SRC = fdt.o fdt_ro.o fdt_wip.o fdt_sw.o fdt_rw.o fdt_strerror.o
+LIBFDT_OBJS = $(patsubst %,../../scripts/dtc/libfdt/%,$(LIBFDT_SRC))
 
 ### Arch-specific stuff
 
@@ -130,9 +132,14 @@ ifeq ($(uname_M), ppc64)
 	OBJS	+= powerpc/ioport.o
 	OBJS	+= powerpc/irq.o
 	OBJS	+= powerpc/kvm.o
+	OBJS	+= powerpc/cpu_info.o
 	OBJS	+= powerpc/kvm-cpu.o
+# We use libfdt, but it's sometimes not packaged 64bit.  It's small too,
+# so just build it in:
+	CFLAGS 	+= -I../../scripts/dtc/libfdt
+	OBJS	+= $(LIBFDT_OBJS)
 	ARCH_INCLUDE := powerpc/include
-	CFLAGS += -m64
+	CFLAGS 	+= -m64
 endif
 
 ###
@@ -222,10 +229,6 @@ DEFINES	+= -DBUILD_ARCH='"$(ARCH)"'
 KVM_INCLUDE := include
 CFLAGS	+= $(CPPFLAGS) $(DEFINES) -I$(KVM_INCLUDE) -I$(ARCH_INCLUDE) -I$(KINCL_PATH)/include -I$(KINCL_PATH)/arch/$(ARCH)/include/ -O2 -fno-strict-aliasing -g
 
-ifneq ($(WERROR),0)
-	WARNINGS += -Werror
-endif
-
 WARNINGS += -Wall
 WARNINGS += -Wcast-align
 WARNINGS += -Wformat=2
@@ -244,6 +247,13 @@ WARNINGS += -Wwrite-strings
 
 CFLAGS	+= $(WARNINGS)
 
+# Some targets may use 'external' sources that don't build totally cleanly.
+CFLAGS_EASYGOING := $(CFLAGS)
+
+ifneq ($(WERROR),0)
+	CFLAGS += -Werror
+endif
+
 all: arch_support_check $(PROGRAM) $(PROGRAM_ALIAS) $(GUEST_INIT) $(GUEST_INIT_S2)
 
 arch_support_check:
@@ -294,6 +304,12 @@ builtin-help.d: $(KVM_INCLUDE)/common-cmds.h
 
 $(OBJS):
 
+# This rule relaxes the -Werror on libfdt, since for now it still has
+# a bunch of warnings. :(
+../../scripts/dtc/libfdt/%.o: ../../scripts/dtc/libfdt/%.c
+	$(E) "  CC      " $@
+	$(Q) $(CC) -c $(CFLAGS_EASYGOING) $< -o $@
+
 util/rbtree.static.o util/rbtree.o: ../../lib/rbtree.c
 	$(E) "  CC      " $@
 	$(Q) $(CC) -c $(CFLAGS) $< -o $@

powerpc/cpu_info.c

@@ -0,0 +1,83 @@
+/*
+ * PPC CPU identification
+ *
+ * This is a very simple "host CPU info" struct to get us going.
+ * For the little host information we need, I don't want to grub about
+ * parsing stuff in /proc/device-tree so just match host PVR to differentiate
+ * PPC970 and POWER7 (which is all that's currently supported).
+ *
+ * Qemu does something similar but this is MUCH simpler!
+ *
+ * Copyright 2012 Matt Evans <matt@ozlabs.org>, IBM Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation.
+ */
+
+#include "cpu_info.h"
+#include "kvm/util.h"
+
+/* POWER7 */
+
+/*
+ * Basic set of pages for POWER7.  It actually supports more but there were some
+ * limitations as to which may be advertised to the guest.  FIXME when this
+ * settles down -- for now use basic set:
+ */
+static u32 power7_page_sizes_prop[] = {0xc, 0x0, 0x1, 0xc, 0x0, 0x18, 0x100, 0x1, 0x18, 0x0};
+/* POWER7 has 1T segments, so advertise these */
+static u32 power7_segment_sizes_prop[] = {0x1c, 0x28, 0xffffffff, 0xffffffff};
+
+static struct cpu_info cpu_power7_info = {
+	"POWER7",
+	power7_page_sizes_prop, sizeof(power7_page_sizes_prop),
+	power7_segment_sizes_prop, sizeof(power7_segment_sizes_prop),
+	32, 		/* SLB size */
+	512000000, 	/* TB frequency */
+	128,		/* d-cache block size */
+	128,		/* i-cache block size */
+	CPUINFO_FLAG_DFP | CPUINFO_FLAG_VSX | CPUINFO_FLAG_VMX
+};
+
+/* PPC970/G5 */
+
+static u32 g5_page_sizes_prop[] = {0xc, 0x0, 0x1, 0xc, 0x0, 0x18, 0x100, 0x1, 0x18, 0x0};
+
+static struct cpu_info cpu_970_info = {
+	"G5",
+	g5_page_sizes_prop, sizeof(g5_page_sizes_prop),
+	0 /* Null = no segment sizes prop, use defaults */, 0,
+	0, 		/* SLB size default */
+	33333333, 	/* TB frequency */
+	128,		/* d-cache block size */
+	128,		/* i-cache block size */
+	CPUINFO_FLAG_VMX
+};
+
+/* This is a default catchall for 'no match' on PVR: */
+static struct cpu_info cpu_dummy_info = { "unknown", 0, 0, 0, 0, 0, 0, 0, 0 };
+
+static struct pvr_info host_pvr_info[] = {
+	{ 0xffffffff, 0x0f000003, &cpu_power7_info },
+	{ 0xffff0000, 0x003f0000, &cpu_power7_info },
+	{ 0xffff0000, 0x004a0000, &cpu_power7_info },
+	{ 0xffff0000, 0x00390000, &cpu_970_info },
+	{ 0xffff0000, 0x003c0000, &cpu_970_info },
+        { 0xffff0000, 0x00440000, &cpu_970_info },
+        { 0xffff0000, 0x00450000, &cpu_970_info },
+};
+
+struct cpu_info *find_cpu_info(u32 pvr)
+{
+	unsigned int i;
+	for (i = 0; i < sizeof(host_pvr_info)/sizeof(struct pvr_info); i++) {
+		if ((pvr & host_pvr_info[i].pvr_mask) ==
+		    host_pvr_info[i].pvr) {
+			return host_pvr_info[i].cpu_info;
+		}
+	}
+	/* Didn't find anything? Rut-ro. */
+	pr_warning("Host CPU unsupported by kvmtool\n");
+	return &cpu_dummy_info;
+}

powerpc/cpu_info.h

@@ -0,0 +1,43 @@
+/*
+ * PPC CPU identification
+ *
+ * Copyright 2012 Matt Evans <matt@ozlabs.org>, IBM Corporation.
+ *
+ * This program is free software; you can redistribute it and/or modify it
+ * under the terms of the GNU General Public License version 2 as published
+ * by the Free Software Foundation.
+ */
+
+#ifndef CPU_INFO_H
+#define CPU_INFO_H
+
+#include <linux/types.h>
+#include <linux/kernel.h>
+
+struct cpu_info {
+	const char	*name;
+	u32 		*page_sizes_prop;
+	u32		page_sizes_prop_len;
+	u32 		*segment_sizes_prop;
+	u32		segment_sizes_prop_len;
+	u32		slb_size;
+	u32		tb_freq;
+	u32		d_bsize;
+	u32		i_bsize;
+	u32		flags;
+};
+
+struct pvr_info {
+	u32		pvr_mask;
+	u32		pvr;
+	struct cpu_info *cpu_info;
+};
+
+/* Misc capabilities/CPU properties */
+#define CPUINFO_FLAG_DFP	0x00000001
+#define CPUINFO_FLAG_VMX	0x00000002
+#define CPUINFO_FLAG_VSX	0x00000004
+
+struct cpu_info *find_cpu_info(u32 pvr);
+
+#endif

powerpc/include/kvm/kvm-arch.h

@@ -53,6 +53,7 @@ struct kvm {
 	void			*ram_start;
 
 	u64			sdr1;
+	u32			pvr;
 
 	bool			nmi_disabled;
 
@@ -70,4 +71,14 @@ struct kvm {
 	int			vm_state;
 };
 
+/* Helper for the various bits of code that generate FDT nodes */
+#define _FDT(exp)							\
+	do {								\
+		int ret = (exp);					\
+		if (ret < 0) {						\
+			die("Error creating device tree: %s: %s\n",	\
+			    #exp, fdt_strerror(ret));			\
+		}							\
+	} while (0)
+
 #endif /* KVM__KVM_ARCH_H */

powerpc/kvm-cpu.c

@@ -134,6 +134,7 @@ static void kvm_cpu__setup_sregs(struct kvm_cpu *vcpu)
 		die("KVM_GET_SREGS failed");
 
 	sregs.u.s.sdr1 = vcpu->kvm->sdr1;
+	sregs.pvr = vcpu->kvm->pvr;
 
 	if (ioctl(vcpu->vcpu_fd, KVM_SET_SREGS, &sregs) < 0)
 		die("KVM_SET_SREGS failed");

powerpc/kvm.c

@@ -3,6 +3,9 @@
  *
  * Copyright 2011 Matt Evans <matt@ozlabs.org>, IBM Corporation.
  *
+ * Portions of FDT setup borrowed from QEMU, copyright 2010 David Gibson, IBM
+ * Corporation.
+ *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms of the GNU General Public License version 2 as published
  * by the Free Software Foundation.
@@ -10,6 +13,8 @@
 
 #include "kvm/kvm.h"
 #include "kvm/util.h"
+#include "libfdt.h"
+#include "cpu_info.h"
 
 #include <linux/kvm.h>
 
@@ -26,7 +31,8 @@
 #include <errno.h>
 
 #include <linux/byteorder.h>
-#include <libfdt.h>
+
+#define HPT_ORDER 24
 
 #define HUGETLBFS_PATH "/var/lib/hugetlbfs/global/pagesize-16MB/"
 
@@ -36,6 +42,13 @@ struct kvm_ext kvm_req_ext[] = {
 	{ 0, 0 }
 };
 
+static uint32_t mfpvr(void)
+{
+	uint32_t r;
+	asm volatile ("mfpvr %0" : "=r"(r));
+	return r;
+}
+
 bool kvm__arch_cpu_supports_vm(void)
 {
 	return true;
@@ -106,6 +119,8 @@ void kvm__arch_init(struct kvm *kvm, const char *hugetlbfs_path, u64 ram_size)
 
 	kvm->sdr1 = ((hpt + 0x3ffffULL) & ~0x3ffffULL) | (HPT_ORDER-18);
 
+	kvm->pvr = mfpvr();
+
 	/* FIXME: This is book3s-specific */
 	cap_ppc_rma = ioctl(kvm->sys_fd, KVM_CHECK_EXTENSION, KVM_CAP_PPC_RMA);
 	if (cap_ppc_rma == 2)
@@ -183,9 +198,146 @@ bool load_bzimage(struct kvm *kvm, int fd_kernel,
 	return false;
 }
 
+#define SMT_THREADS 4
+
+/*
+ * Set up the FDT for the kernel: This function is currently fairly SPAPR-heavy,
+ * and whilst most PPC targets will require CPU/memory nodes, others like RTAS
+ * should eventually be added separately.
+ */
 static void setup_fdt(struct kvm *kvm)
 {
+	uint64_t 	mem_reg_property[] = { 0, cpu_to_be64(kvm->ram_size) };
+	int 		smp_cpus = kvm->nrcpus;
+	char 		hypertas_prop_kvm[] = "hcall-pft\0hcall-term\0"
+		"hcall-dabr\0hcall-interrupt\0hcall-tce\0hcall-vio\0"
+		"hcall-splpar\0hcall-bulk";
+	int 		i, j;
+	char 		cpu_name[30];
+	u8		staging_fdt[FDT_MAX_SIZE];
+	struct cpu_info *cpu_info = find_cpu_info(kvm->pvr);
 
+	/* Generate an appropriate DT at kvm->fdt_gra */
+	void *fdt_dest = guest_flat_to_host(kvm, kvm->fdt_gra);
+	void *fdt = staging_fdt;
+
+	_FDT(fdt_create(fdt, FDT_MAX_SIZE));
+	_FDT(fdt_finish_reservemap(fdt));
+
+	_FDT(fdt_begin_node(fdt, ""));
+
+	_FDT(fdt_property_string(fdt, "device_type", "chrp"));
+	_FDT(fdt_property_string(fdt, "model", "IBM pSeries (kvmtool)"));
+	_FDT(fdt_property_cell(fdt, "#address-cells", 0x2));
+	_FDT(fdt_property_cell(fdt, "#size-cells", 0x2));
+
+	/* /chosen */
+	_FDT(fdt_begin_node(fdt, "chosen"));
+	/* cmdline */
+	_FDT(fdt_property_string(fdt, "bootargs", kern_cmdline));
+	/* Initrd */
+	if (kvm->initrd_size != 0) {
+		uint32_t ird_st_prop = cpu_to_be32(kvm->initrd_gra);
+		uint32_t ird_end_prop = cpu_to_be32(kvm->initrd_gra +
+						    kvm->initrd_size);
+		_FDT(fdt_property(fdt, "linux,initrd-start",
+				   &ird_st_prop, sizeof(ird_st_prop)));
+		_FDT(fdt_property(fdt, "linux,initrd-end",
+				   &ird_end_prop, sizeof(ird_end_prop)));
+	}
+	_FDT(fdt_end_node(fdt));
+
+	/*
+	 * Memory: We don't alloc. a separate RMA yet.  If we ever need to
+	 * (CAP_PPC_RMA == 2) then have one memory node for 0->RMAsize, and
+	 * another RMAsize->endOfMem.
+	 */
+	_FDT(fdt_begin_node(fdt, "memory@0"));
+	_FDT(fdt_property_string(fdt, "device_type", "memory"));
+	_FDT(fdt_property(fdt, "reg", mem_reg_property,
+			  sizeof(mem_reg_property)));
+	_FDT(fdt_end_node(fdt));
+
+	/* CPUs */
+	_FDT(fdt_begin_node(fdt, "cpus"));
+	_FDT(fdt_property_cell(fdt, "#address-cells", 0x1));
+	_FDT(fdt_property_cell(fdt, "#size-cells", 0x0));
+
+	for (i = 0; i < smp_cpus; i += SMT_THREADS) {
+		int32_t pft_size_prop[] = { 0, HPT_ORDER };
+		uint32_t servers_prop[SMT_THREADS];
+		uint32_t gservers_prop[SMT_THREADS * 2];
+		int threads = (smp_cpus - i) >= SMT_THREADS ? SMT_THREADS :
+			smp_cpus - i;
+
+		sprintf(cpu_name, "PowerPC,%s@%d", cpu_info->name, i);
+		_FDT(fdt_begin_node(fdt, cpu_name));
+		sprintf(cpu_name, "PowerPC,%s", cpu_info->name);
+		_FDT(fdt_property_string(fdt, "name", cpu_name));
+		_FDT(fdt_property_string(fdt, "device_type", "cpu"));
+
+		_FDT(fdt_property_cell(fdt, "reg", i));
+		_FDT(fdt_property_cell(fdt, "cpu-version", kvm->pvr));
+
+		_FDT(fdt_property_cell(fdt, "dcache-block-size", cpu_info->d_bsize));
+		_FDT(fdt_property_cell(fdt, "icache-block-size", cpu_info->i_bsize));
+
+		_FDT(fdt_property_cell(fdt, "timebase-frequency", cpu_info->tb_freq));
+		/* Lies, but safeish lies! */
+		_FDT(fdt_property_cell(fdt, "clock-frequency", 0xddbab200));
+
+		if (cpu_info->slb_size)
+			_FDT(fdt_property_cell(fdt, "ibm,slb-size", cpu_info->slb_size));
+		/*
+		 * HPT size is hardwired; KVM currently fixes it at 16MB but the
+		 * moment that changes we'll need to read it out of the kernel.
+		 */
+		_FDT(fdt_property(fdt, "ibm,pft-size", pft_size_prop,
+				  sizeof(pft_size_prop)));
+
+		_FDT(fdt_property_string(fdt, "status", "okay"));
+		_FDT(fdt_property(fdt, "64-bit", NULL, 0));
+		/* A server for each thread in this core */
+		for (j = 0; j < SMT_THREADS; j++) {
+			servers_prop[j] = cpu_to_be32(i+j);
+			/*
+			 * Hack borrowed from QEMU, direct the group queues back
+			 * to cpu 0:
+			 */
+			gservers_prop[j*2] = cpu_to_be32(i+j);
+			gservers_prop[j*2 + 1] = 0;
+		}
+		_FDT(fdt_property(fdt, "ibm,ppc-interrupt-server#s",
+				   servers_prop, threads * sizeof(uint32_t)));
+		_FDT(fdt_property(fdt, "ibm,ppc-interrupt-gserver#s",
+				  gservers_prop,
+				  threads * 2 * sizeof(uint32_t)));
+		if (cpu_info->page_sizes_prop)
+			_FDT(fdt_property(fdt, "ibm,segment-page-sizes",
+					  cpu_info->page_sizes_prop,
+					  cpu_info->page_sizes_prop_len));
+		if (cpu_info->segment_sizes_prop)
+			_FDT(fdt_property(fdt, "ibm,processor-segment-sizes",
+					  cpu_info->segment_sizes_prop,
+					  cpu_info->segment_sizes_prop_len));
+		/* VSX / DFP options: */
+		if (cpu_info->flags & CPUINFO_FLAG_VMX)
+			_FDT(fdt_property_cell(fdt, "ibm,vmx",
+					       (cpu_info->flags &
+						CPUINFO_FLAG_VSX) ? 2 : 1));
+		if (cpu_info->flags & CPUINFO_FLAG_DFP)
+			_FDT(fdt_property_cell(fdt, "ibm,dfp", 0x1));
+		_FDT(fdt_end_node(fdt));
+	}
+	_FDT(fdt_end_node(fdt));
+
+	/* Finalise: */
+	_FDT(fdt_end_node(fdt)); /* Root node */
+	_FDT(fdt_finish(fdt));
+
+	_FDT(fdt_open_into(fdt, fdt_dest, FDT_MAX_SIZE));
+	_FDT(fdt_add_mem_rsv(fdt_dest, kvm->rtas_gra, kvm->rtas_size));
+	_FDT(fdt_pack(fdt_dest));
 }
 
 /**