xref: /src/sys/arch/x86/pci/pci_machdep.c

/*	$NetBSD: pci_machdep.c,v 1.70 2015/04/27 07:03:58 knakahara Exp $	*/

/*-
 * Copyright (c) 1997, 1998 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This code is derived from software contributed to The NetBSD Foundation
 * by Jason R. Thorpe of the Numerical Aerospace Simulation Facility,
 * NASA Ames Research Center.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

/*
 * Copyright (c) 1996 Christopher G. Demetriou.  All rights reserved.
 * Copyright (c) 1994 Charles M. Hannum.  All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. All advertising materials mentioning features or use of this software
 *    must display the following acknowledgement:
 *	This product includes software developed by Charles M. Hannum.
 * 4. The name of the author may not be used to endorse or promote products
 *    derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

/*
 * Machine-specific functions for PCI autoconfiguration.
 *
 * On PCs, there are two methods of generating PCI configuration cycles.
 * We try to detect the appropriate mechanism for this machine and set
 * up a few function pointers to access the correct method directly.
 *
 * The configuration method can be hard-coded in the config file by
 * using `options PCI_CONF_MODE=N', where `N' is the configuration mode
 * as defined in section 3.6.4.1, `Generating Configuration Cycles'.
 */

#include <sys/cdefs.h>
__KERNEL_RCSID(0, "$NetBSD: pci_machdep.c,v 1.70 2015/04/27 07:03:58 knakahara Exp $");

#include <sys/types.h>
#include <sys/param.h>
#include <sys/time.h>
#include <sys/systm.h>
#include <sys/errno.h>
#include <sys/device.h>
#include <sys/bus.h>
#include <sys/cpu.h>
#include <sys/kmem.h>

#include <uvm/uvm_extern.h>

#include <machine/bus_private.h>

#include <machine/pio.h>
#include <machine/lock.h>

#include <dev/isa/isareg.h>
#include <dev/isa/isavar.h>
#include <dev/pci/pcivar.h>
#include <dev/pci/pcireg.h>
#include <dev/pci/pccbbreg.h>
#include <dev/pci/pcidevs.h>
#include <dev/pci/genfb_pcivar.h>

#include <dev/wsfb/genfbvar.h>
#include <arch/x86/include/genfb_machdep.h>
#include <dev/ic/vgareg.h>

#include "acpica.h"
#include "genfb.h"
#include "isa.h"
#include "opt_acpi.h"
#include "opt_ddb.h"
#include "opt_mpbios.h"
#include "opt_puc.h"
#include "opt_vga.h"
#include "pci.h"
#include "wsdisplay.h"
#include "com.h"

#ifdef DDB
#include <machine/db_machdep.h>
#include <ddb/db_sym.h>
#include <ddb/db_extern.h>
#endif

#ifdef VGA_POST
#include <x86/vga_post.h>
#endif

#include <x86/cpuvar.h>

#include <machine/autoconf.h>
#include <machine/bootinfo.h>

#ifdef MPBIOS
#include <machine/mpbiosvar.h>
#endif

#if NACPICA > 0
#include <machine/mpacpi.h>
#endif

#include <machine/mpconfig.h>

#if NCOM > 0
#include <dev/pci/puccn.h>
#endif

#include "opt_pci_conf_mode.h"

#ifdef PCI_CONF_MODE
#if (PCI_CONF_MODE == 1) || (PCI_CONF_MODE == 2)
static int pci_mode = PCI_CONF_MODE;
#else
#error Invalid PCI configuration mode.
#endif
#else
static int pci_mode = -1;
#endif

struct pci_conf_lock {
	uint32_t cl_cpuno;	/* 0: unlocked
				 * 1 + n: locked by CPU n (0 <= n)
				 */
	uint32_t cl_sel;	/* the address that's being read. */
};

static void pci_conf_unlock(struct pci_conf_lock *);
static uint32_t pci_conf_selector(pcitag_t, int);
static unsigned int pci_conf_port(pcitag_t, int);
static void pci_conf_select(uint32_t);
static void pci_conf_lock(struct pci_conf_lock *, uint32_t);
static void pci_bridge_hook(pci_chipset_tag_t, pcitag_t, void *);
struct pci_bridge_hook_arg {
	void (*func)(pci_chipset_tag_t, pcitag_t, void *);
	void *arg;
};

#define	PCI_MODE1_ENABLE	0x80000000UL
#define	PCI_MODE1_ADDRESS_REG	0x0cf8
#define	PCI_MODE1_DATA_REG	0x0cfc

#define	PCI_MODE2_ENABLE_REG	0x0cf8
#define	PCI_MODE2_FORWARD_REG	0x0cfa

#define _tag(b, d, f) \
	{.mode1 = PCI_MODE1_ENABLE | ((b) << 16) | ((d) << 11) | ((f) << 8)}
#define _qe(bus, dev, fcn, vend, prod) \
	{_tag(bus, dev, fcn), PCI_ID_CODE(vend, prod)}
const struct {
	pcitag_t tag;
	pcireg_t id;
} pcim1_quirk_tbl[] = {
	_qe(0, 0, 0, PCI_VENDOR_INVALID, 0x0000), /* patchable */
	_qe(0, 0, 0, PCI_VENDOR_COMPAQ, PCI_PRODUCT_COMPAQ_TRIFLEX1),
	/* XXX Triflex2 not tested */
	_qe(0, 0, 0, PCI_VENDOR_COMPAQ, PCI_PRODUCT_COMPAQ_TRIFLEX2),
	_qe(0, 0, 0, PCI_VENDOR_COMPAQ, PCI_PRODUCT_COMPAQ_TRIFLEX4),
	/* Triton needed for Connectix Virtual PC */
	_qe(0, 0, 0, PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82437FX),
	/* Connectix Virtual PC 5 has a 440BX */
	_qe(0, 0, 0, PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82443BX_NOAGP),
	/* Parallels Desktop for Mac */
	_qe(0, 2, 0, PCI_VENDOR_PARALLELS, PCI_PRODUCT_PARALLELS_VIDEO),
	_qe(0, 3, 0, PCI_VENDOR_PARALLELS, PCI_PRODUCT_PARALLELS_TOOLS),
	/* SIS 740 */
	_qe(0, 0, 0, PCI_VENDOR_SIS, PCI_PRODUCT_SIS_740),
	/* SIS 741 */
	_qe(0, 0, 0, PCI_VENDOR_SIS, PCI_PRODUCT_SIS_741),
	/* VIA Technologies VX900 */
	_qe(0, 0, 0, PCI_VENDOR_VIATECH, PCI_PRODUCT_VIATECH_VX900_HB)
};
#undef _tag
#undef _qe

/* arch/xen does not support MSI/MSI-X yet. */
#ifdef __HAVE_PCI_MSI_MSIX
#define PCI_QUIRK_DISABLE_MSI	1 /* Neigher MSI nor MSI-X work */
#define PCI_QUIRK_DISABLE_MSIX	2 /* MSI-X does not work */
#define PCI_QUIRK_ENABLE_MSI_VM	3 /* Older chipset in VM where MSI and MSI-X works */

#define _dme(vend, prod) \
	{ PCI_QUIRK_DISABLE_MSI, PCI_ID_CODE(vend, prod) }
#define _dmxe(vend, prod) \
	{ PCI_QUIRK_DISABLE_MSIX, PCI_ID_CODE(vend, prod) }
#define _emve(vend, prod) \
	{ PCI_QUIRK_ENABLE_MSI_VM, PCI_ID_CODE(vend, prod) }
const struct {
	int type;
	pcireg_t id;
} pci_msi_quirk_tbl[] = {
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_PCMC),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82437FX),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82437MX),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82437VX),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82439HX),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82439TX),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82443GX),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82443GX_AGP),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82440MX),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82441FX),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82443BX),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82443BX_AGP),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82443BX_NOAGP),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82443GX_NOAGP),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82443LX),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82443LX_AGP),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82810_MCH),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82810E_MCH),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82815_FULL_HUB),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82820_MCH),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82830MP_IO_1),
	_dme(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82840_HB),
	_dme(PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE_PCHB),
	_dme(PCI_VENDOR_NVIDIA, PCI_PRODUCT_NVIDIA_NFORCE2_PCHB),
	_dme(PCI_VENDOR_AMD, PCI_PRODUCT_AMD_SC751_SC),
	_dme(PCI_VENDOR_AMD, PCI_PRODUCT_AMD_SC761_SC),
	_dme(PCI_VENDOR_AMD, PCI_PRODUCT_AMD_SC762_NB),

	_emve(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82441FX), /* QEMU */
	_emve(PCI_VENDOR_INTEL, PCI_PRODUCT_INTEL_82443BX), /* VMWare */
};
#undef _dme
#undef _dmxe
#undef _emve
#endif /* __HAVE_PCI_MSI_MSIX */

/*
 * PCI doesn't have any special needs; just use the generic versions
 * of these functions.
 */
struct x86_bus_dma_tag pci_bus_dma_tag = {
	._tag_needs_free	= 0,
#if defined(_LP64) || defined(PAE)
	._bounce_thresh		= PCI32_DMA_BOUNCE_THRESHOLD,
	._bounce_alloc_lo	= ISA_DMA_BOUNCE_THRESHOLD,
	._bounce_alloc_hi	= PCI32_DMA_BOUNCE_THRESHOLD,
#else
	._bounce_thresh		= 0,
	._bounce_alloc_lo	= 0,
	._bounce_alloc_hi	= 0,
#endif
	._may_bounce		= NULL,
};

#ifdef _LP64
struct x86_bus_dma_tag pci_bus_dma64_tag = {
	._tag_needs_free	= 0,
	._bounce_thresh		= 0,
	._bounce_alloc_lo	= 0,
	._bounce_alloc_hi	= 0,
	._may_bounce		= NULL,
};
#endif

static struct pci_conf_lock cl0 = {
	  .cl_cpuno = 0UL
	, .cl_sel = 0UL
};

static struct pci_conf_lock * const cl = &cl0;

#if NGENFB > 0 && NACPICA > 0 && defined(VGA_POST)
extern int acpi_md_vbios_reset;
extern int acpi_md_vesa_modenum;
#endif

static struct genfb_colormap_callback gfb_cb;
static struct genfb_pmf_callback pmf_cb;
static struct genfb_mode_callback mode_cb;
#ifdef VGA_POST
static struct vga_post *vga_posth = NULL;
#endif

static void
pci_conf_lock(struct pci_conf_lock *ocl, uint32_t sel)
{
	uint32_t cpuno;

	KASSERT(sel != 0);

	kpreempt_disable();
	cpuno = cpu_number() + 1;
	/* If the kernel enters pci_conf_lock() through an interrupt
	 * handler, then the CPU may already hold the lock.
	 *
	 * If the CPU does not already hold the lock, spin until
	 * we can acquire it.
	 */
	if (cpuno == cl->cl_cpuno) {
		ocl->cl_cpuno = cpuno;
	} else {
		u_int spins;

		ocl->cl_cpuno = 0;

		spins = SPINLOCK_BACKOFF_MIN;
		while (atomic_cas_32(&cl->cl_cpuno, 0, cpuno) != 0) {
			SPINLOCK_BACKOFF(spins);
#ifdef LOCKDEBUG
			if (SPINLOCK_SPINOUT(spins)) {
				panic("%s: cpu %" PRId32
				    " spun out waiting for cpu %" PRId32,
				    __func__, cpuno, cl->cl_cpuno);
			}
#endif	/* LOCKDEBUG */
		}
	}

	/* Only one CPU can be here, so an interlocked atomic_swap(3)
	 * is not necessary.
	 *
	 * Evaluating atomic_cas_32_ni()'s argument, cl->cl_sel,
	 * and applying atomic_cas_32_ni() is not an atomic operation,
	 * however, any interrupt that, in the middle of the
	 * operation, modifies cl->cl_sel, will also restore
	 * cl->cl_sel.  So cl->cl_sel will have the same value when
	 * we apply atomic_cas_32_ni() as when we evaluated it,
	 * before.
	 */
	ocl->cl_sel = atomic_cas_32_ni(&cl->cl_sel, cl->cl_sel, sel);
	pci_conf_select(sel);
}

static void
pci_conf_unlock(struct pci_conf_lock *ocl)
{
	atomic_cas_32_ni(&cl->cl_sel, cl->cl_sel, ocl->cl_sel);
	pci_conf_select(ocl->cl_sel);
	if (ocl->cl_cpuno != cl->cl_cpuno)
		atomic_cas_32(&cl->cl_cpuno, cl->cl_cpuno, ocl->cl_cpuno);
	kpreempt_enable();
}

static uint32_t
pci_conf_selector(pcitag_t tag, int reg)
{
	static const pcitag_t mode2_mask = {
		.mode2 = {
			  .enable = 0xff
			, .forward = 0xff
		}
	};

	switch (pci_mode) {
	case 1:
		return tag.mode1 | reg;
	case 2:
		return tag.mode1 & mode2_mask.mode1;
	default:
		panic("%s: mode %d not configured", __func__, pci_mode);
	}
}

static unsigned int
pci_conf_port(pcitag_t tag, int reg)
{
	switch (pci_mode) {
	case 1:
		return PCI_MODE1_DATA_REG;
	case 2:
		return tag.mode2.port | reg;
	default:
		panic("%s: mode %d not configured", __func__, pci_mode);
	}
}

static void
pci_conf_select(uint32_t sel)
{
	pcitag_t tag;

	switch (pci_mode) {
	case 1:
		outl(PCI_MODE1_ADDRESS_REG, sel);
		return;
	case 2:
		tag.mode1 = sel;
		outb(PCI_MODE2_ENABLE_REG, tag.mode2.enable);
		if (tag.mode2.enable != 0)
			outb(PCI_MODE2_FORWARD_REG, tag.mode2.forward);
		return;
	default:
		panic("%s: mode %d not configured", __func__, pci_mode);
	}
}

#ifdef __HAVE_PCI_MSI_MSIX
static int
pci_has_msi_quirk(pcireg_t id, int type)
{
	int i;

	for (i = 0; i < __arraycount(pci_msi_quirk_tbl); i++) {
		if (id == pci_msi_quirk_tbl[i].id &&
		    type == pci_msi_quirk_tbl[i].type)
			return 1;
	}

	return 0;
}
#endif

void
pci_attach_hook(device_t parent, device_t self, struct pcibus_attach_args *pba)
{
#ifdef __HAVE_PCI_MSI_MSIX
	pci_chipset_tag_t pc = pba->pba_pc;
	pcitag_t tag;
	pcireg_t id, class;
#endif

	if (pba->pba_bus == 0)
		aprint_normal(": configuration mode %d", pci_mode);
#ifdef MPBIOS
	mpbios_pci_attach_hook(parent, self, pba);
#endif
#if NACPICA > 0
	mpacpi_pci_attach_hook(parent, self, pba);
#endif

#ifdef __HAVE_PCI_MSI_MSIX
	/*
	 * In order to decide whether the system supports MSI we look
	 * at the host bridge, which should be device 0 function 0 on
	 * bus 0.  It is better to not enable MSI on systems that
	 * support it than the other way around, so be conservative
	 * here.  So we don't enable MSI if we don't find a host
	 * bridge there.  We also deliberately don't enable MSI on
	 * chipsets from low-end manifacturers like VIA and SiS.
	 */
	tag = pci_make_tag(pc, 0, 0, 0);
	id = pci_conf_read(pc, tag, PCI_ID_REG);
	class = pci_conf_read(pc, tag, PCI_CLASS_REG);

	if (PCI_CLASS(class) != PCI_CLASS_BRIDGE ||
	    PCI_SUBCLASS(class) != PCI_SUBCLASS_BRIDGE_HOST)
		return;

	if (pci_has_msi_quirk(id, PCI_QUIRK_DISABLE_MSI)) {
		pba->pba_flags &= ~PCI_FLAGS_MSI_OKAY;
		pba->pba_flags &= ~PCI_FLAGS_MSIX_OKAY;
	} else if (pci_has_msi_quirk(id, PCI_QUIRK_DISABLE_MSIX)) {
		pba->pba_flags |= PCI_FLAGS_MSI_OKAY;
		pba->pba_flags &= ~PCI_FLAGS_MSIX_OKAY;
	} else {
		pba->pba_flags |= PCI_FLAGS_MSI_OKAY;
		pba->pba_flags |= PCI_FLAGS_MSIX_OKAY;
	}

	/* VMware and KVM use old chipset, but they can use MSI/MSI-X */
	if (cpu_feature[1] & CPUID2_RAZ) {
		if (pci_has_msi_quirk(id, PCI_QUIRK_ENABLE_MSI_VM)) {
			pba->pba_flags |= PCI_FLAGS_MSI_OKAY;
			pba->pba_flags |= PCI_FLAGS_MSIX_OKAY;
		}
	}

	/*
	 * Don't enable MSI on a HyperTransport bus.  In order to
	 * determine that bus 0 is a HyperTransport bus, we look at
	 * device 24 function 0, which is the HyperTransport
	 * host/primary interface integrated on most 64-bit AMD CPUs.
	 * If that device has a HyperTransport capability, bus 0 must
	 * be a HyperTransport bus and we disable MSI.
	 */
	tag = pci_make_tag(pc, 0, 24, 0);
	if (pci_get_capability(pc, tag, PCI_CAP_LDT, NULL, NULL)) {
		pba->pba_flags &= ~PCI_FLAGS_MSI_OKAY;
		pba->pba_flags &= ~PCI_FLAGS_MSIX_OKAY;
	}
#endif /* __HAVE_PCI_MSI_MSIX */
}

int
pci_bus_maxdevs(pci_chipset_tag_t pc, int busno)
{
	/*
	 * Bus number is irrelevant.  If Configuration Mechanism 2 is in
	 * use, can only have devices 0-15 on any bus.  If Configuration
	 * Mechanism 1 is in use, can have devices 0-32 (i.e. the `normal'
	 * range).
	 */
	if (pci_mode == 2)
		return (16);
	else
		return (32);
}

pcitag_t
pci_make_tag(pci_chipset_tag_t pc, int bus, int device, int function)
{
	pci_chipset_tag_t ipc;
	pcitag_t tag;

	for (ipc = pc; ipc != NULL; ipc = ipc->pc_super) {
		if ((ipc->pc_present & PCI_OVERRIDE_MAKE_TAG) == 0)
			continue;
		return (*ipc->pc_ov->ov_make_tag)(ipc->pc_ctx,
		    pc, bus, device, function);
	}

	switch (pci_mode) {
	case 1:
		if (bus >= 256 || device >= 32 || function >= 8)
			panic("%s: bad request(%d, %d, %d)", __func__,
			    bus, device, function);

		tag.mode1 = PCI_MODE1_ENABLE |
			    (bus << 16) | (device << 11) | (function << 8);
		return tag;
	case 2:
		if (bus >= 256 || device >= 16 || function >= 8)
			panic("%s: bad request(%d, %d, %d)", __func__,
			    bus, device, function);

		tag.mode2.port = 0xc000 | (device << 8);
		tag.mode2.enable = 0xf0 | (function << 1);
		tag.mode2.forward = bus;
		return tag;
	default:
		panic("%s: mode %d not configured", __func__, pci_mode);
	}
}

void
pci_decompose_tag(pci_chipset_tag_t pc, pcitag_t tag,
    int *bp, int *dp, int *fp)
{
	pci_chipset_tag_t ipc;

	for (ipc = pc; ipc != NULL; ipc = ipc->pc_super) {
		if ((ipc->pc_present & PCI_OVERRIDE_DECOMPOSE_TAG) == 0)
			continue;
		(*ipc->pc_ov->ov_decompose_tag)(ipc->pc_ctx,
		    pc, tag, bp, dp, fp);
		return;
	}

	switch (pci_mode) {
	case 1:
		if (bp != NULL)
			*bp = (tag.mode1 >> 16) & 0xff;
		if (dp != NULL)
			*dp = (tag.mode1 >> 11) & 0x1f;
		if (fp != NULL)
			*fp = (tag.mode1 >> 8) & 0x7;
		return;
	case 2:
		if (bp != NULL)
			*bp = tag.mode2.forward & 0xff;
		if (dp != NULL)
			*dp = (tag.mode2.port >> 8) & 0xf;
		if (fp != NULL)
			*fp = (tag.mode2.enable >> 1) & 0x7;
		return;
	default:
		panic("%s: mode %d not configured", __func__, pci_mode);
	}
}

pcireg_t
pci_conf_read(pci_chipset_tag_t pc, pcitag_t tag, int reg)
{
	pci_chipset_tag_t ipc;
	pcireg_t data;
	struct pci_conf_lock ocl;

	KASSERT((reg & 0x3) == 0);

	for (ipc = pc; ipc != NULL; ipc = ipc->pc_super) {
		if ((ipc->pc_present & PCI_OVERRIDE_CONF_READ) == 0)
			continue;
		return (*ipc->pc_ov->ov_conf_read)(ipc->pc_ctx, pc, tag, reg);
	}

	pci_conf_lock(&ocl, pci_conf_selector(tag, reg));
	data = inl(pci_conf_port(tag, reg));
	pci_conf_unlock(&ocl);
	return data;
}

void
pci_conf_write(pci_chipset_tag_t pc, pcitag_t tag, int reg, pcireg_t data)
{
	pci_chipset_tag_t ipc;
	struct pci_conf_lock ocl;

	KASSERT((reg & 0x3) == 0);

	for (ipc = pc; ipc != NULL; ipc = ipc->pc_super) {
		if ((ipc->pc_present & PCI_OVERRIDE_CONF_WRITE) == 0)
			continue;
		(*ipc->pc_ov->ov_conf_write)(ipc->pc_ctx, pc, tag, reg,
		    data);
		return;
	}

	pci_conf_lock(&ocl, pci_conf_selector(tag, reg));
	outl(pci_conf_port(tag, reg), data);
	pci_conf_unlock(&ocl);
}

void
pci_mode_set(int mode)
{
	KASSERT(pci_mode == -1 || pci_mode == mode);

	pci_mode = mode;
}

int
pci_mode_detect(void)
{
	uint32_t sav, val;
	int i;
	pcireg_t idreg;
	extern char cpu_brand_string[];

	if (pci_mode != -1)
		return pci_mode;

	/*
	 * We try to divine which configuration mode the host bridge wants.
	 */

	sav = inl(PCI_MODE1_ADDRESS_REG);

	pci_mode = 1; /* assume this for now */
	/*
	 * catch some known buggy implementations of mode 1
	 */
	for (i = 0; i < __arraycount(pcim1_quirk_tbl); i++) {
		pcitag_t t;

		if (PCI_VENDOR(pcim1_quirk_tbl[i].id) == PCI_VENDOR_INVALID)
			continue;
		t.mode1 = pcim1_quirk_tbl[i].tag.mode1;
		idreg = pci_conf_read(NULL, t, PCI_ID_REG); /* needs "pci_mode" */
		if (idreg == pcim1_quirk_tbl[i].id) {
#ifdef DEBUG
			printf("%s: known mode 1 PCI chipset (%08x)\n",
			    __func__, idreg);
#endif
			return (pci_mode);
		}
	}

	const char *reason, *system_vendor, *system_product;
	if (memcmp(cpu_brand_string, "QEMU", 4) == 0)
		/* PR 45671, https://bugs.launchpad.net/qemu/+bug/897771 */
		reason = "QEMU";
	else if ((system_vendor = pmf_get_platform("system-vendor")) != NULL &&
	    strcmp(system_vendor, "Xen") == 0 &&
	    (system_product = pmf_get_platform("system-product")) != NULL &&
	    strcmp(system_product, "HVM domU") == 0)
		reason = "Xen";
	else
		reason = NULL;

	if (reason) {
#ifdef DEBUG
		printf("%s: forcing PCI mode 1 for %s\n", __func__, reason);
#endif
		return (pci_mode);
	}

	/*
	 * Strong check for standard compliant mode 1:
	 * 1. bit 31 ("enable") can be set
	 * 2. byte/word access does not affect register
	 */
	outl(PCI_MODE1_ADDRESS_REG, PCI_MODE1_ENABLE);
	outb(PCI_MODE1_ADDRESS_REG + 3, 0);
	outw(PCI_MODE1_ADDRESS_REG + 2, 0);
	val = inl(PCI_MODE1_ADDRESS_REG);
	if ((val & 0x80fffffc) != PCI_MODE1_ENABLE) {
#ifdef DEBUG
		printf("%s: mode 1 enable failed (%x)\n", __func__, val);
#endif
		goto not1;
	}
	outl(PCI_MODE1_ADDRESS_REG, 0);
	val = inl(PCI_MODE1_ADDRESS_REG);
	if ((val & 0x80fffffc) != 0)
		goto not1;
	return (pci_mode);
not1:
	outl(PCI_MODE1_ADDRESS_REG, sav);

	/*
	 * This mode 2 check is quite weak (and known to give false
	 * positives on some Compaq machines).
	 * However, this doesn't matter, because this is the
	 * last test, and simply no PCI devices will be found if
	 * this happens.
	 */
	outb(PCI_MODE2_ENABLE_REG, 0);
	outb(PCI_MODE2_FORWARD_REG, 0);
	if (inb(PCI_MODE2_ENABLE_REG) != 0 ||
	    inb(PCI_MODE2_FORWARD_REG) != 0)
		goto not2;
	return (pci_mode = 2);
not2:

	return (pci_mode = 0);
}

void
pci_device_foreach(pci_chipset_tag_t pc, int maxbus,
	void (*func)(pci_chipset_tag_t, pcitag_t, void *), void *context)
{
	pci_device_foreach_min(pc, 0, maxbus, func, context);
}

void
pci_device_foreach_min(pci_chipset_tag_t pc, int minbus, int maxbus,
	void (*func)(pci_chipset_tag_t, pcitag_t, void *), void *context)
{
	const struct pci_quirkdata *qd;
	int bus, device, function, maxdevs, nfuncs;
	pcireg_t id, bhlcr;
	pcitag_t tag;

	for (bus = minbus; bus <= maxbus; bus++) {
		maxdevs = pci_bus_maxdevs(pc, bus);
		for (device = 0; device < maxdevs; device++) {
			tag = pci_make_tag(pc, bus, device, 0);
			id = pci_conf_read(pc, tag, PCI_ID_REG);

			/* Invalid vendor ID value? */
			if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
				continue;
			/* XXX Not invalid, but we've done this ~forever. */
			if (PCI_VENDOR(id) == 0)
				continue;

			qd = pci_lookup_quirkdata(PCI_VENDOR(id),
				PCI_PRODUCT(id));

			bhlcr = pci_conf_read(pc, tag, PCI_BHLC_REG);
			if (PCI_HDRTYPE_MULTIFN(bhlcr) ||
			     (qd != NULL &&
			     (qd->quirks & PCI_QUIRK_MULTIFUNCTION) != 0))
				nfuncs = 8;
			else
				nfuncs = 1;

			for (function = 0; function < nfuncs; function++) {
				tag = pci_make_tag(pc, bus, device, function);
				id = pci_conf_read(pc, tag, PCI_ID_REG);

				/* Invalid vendor ID value? */
				if (PCI_VENDOR(id) == PCI_VENDOR_INVALID)
					continue;
				/*
				 * XXX Not invalid, but we've done this
				 * ~forever.
				 */
				if (PCI_VENDOR(id) == 0)
					continue;
				(*func)(pc, tag, context);
			}
		}
	}
}

void
pci_bridge_foreach(pci_chipset_tag_t pc, int minbus, int maxbus,
	void (*func)(pci_chipset_tag_t, pcitag_t, void *), void *ctx)
{
	struct pci_bridge_hook_arg bridge_hook;

	bridge_hook.func = func;
	bridge_hook.arg = ctx;

	pci_device_foreach_min(pc, minbus, maxbus, pci_bridge_hook,
		&bridge_hook);
}

static void
pci_bridge_hook(pci_chipset_tag_t pc, pcitag_t tag, void *ctx)
{
	struct pci_bridge_hook_arg *bridge_hook = (void *)ctx;
	pcireg_t reg;

	reg = pci_conf_read(pc, tag, PCI_CLASS_REG);
	if (PCI_CLASS(reg) == PCI_CLASS_BRIDGE &&
	    (PCI_SUBCLASS(reg) == PCI_SUBCLASS_BRIDGE_PCI ||
		PCI_SUBCLASS(reg) == PCI_SUBCLASS_BRIDGE_CARDBUS)) {
		(*bridge_hook->func)(pc, tag, bridge_hook->arg);
	}
}

static const void *
bit_to_function_pointer(const struct pci_overrides *ov, uint64_t bit)
{
	switch (bit) {
	case PCI_OVERRIDE_CONF_READ:
		return ov->ov_conf_read;
	case PCI_OVERRIDE_CONF_WRITE:
		return ov->ov_conf_write;
	case PCI_OVERRIDE_INTR_MAP:
		return ov->ov_intr_map;
	case PCI_OVERRIDE_INTR_STRING:
		return ov->ov_intr_string;
	case PCI_OVERRIDE_INTR_EVCNT:
		return ov->ov_intr_evcnt;
	case PCI_OVERRIDE_INTR_ESTABLISH:
		return ov->ov_intr_establish;
	case PCI_OVERRIDE_INTR_DISESTABLISH:
		return ov->ov_intr_disestablish;
	case PCI_OVERRIDE_MAKE_TAG:
		return ov->ov_make_tag;
	case PCI_OVERRIDE_DECOMPOSE_TAG:
		return ov->ov_decompose_tag;
	default:
		return NULL;
	}
}

void
pci_chipset_tag_destroy(pci_chipset_tag_t pc)
{
	kmem_free(pc, sizeof(struct pci_chipset_tag));
}

int
pci_chipset_tag_create(pci_chipset_tag_t opc, const uint64_t present,
    const struct pci_overrides *ov, void *ctx, pci_chipset_tag_t *pcp)
{
	uint64_t bit, bits, nbits;
	pci_chipset_tag_t pc;
	const void *fp;

	if (ov == NULL || present == 0)
		return EINVAL;

	pc = kmem_alloc(sizeof(struct pci_chipset_tag), KM_SLEEP);

	if (pc == NULL)
		return ENOMEM;

	pc->pc_super = opc;

	for (bits = present; bits != 0; bits = nbits) {
		nbits = bits & (bits - 1);
		bit = nbits ^ bits;
		if ((fp = bit_to_function_pointer(ov, bit)) == NULL) {
#ifdef DEBUG
			printf("%s: missing bit %" PRIx64 "\n", __func__, bit);
#endif
			goto einval;
		}
	}

	pc->pc_ov = ov;
	pc->pc_present = present;
	pc->pc_ctx = ctx;

	*pcp = pc;

	return 0;
einval:
	kmem_free(pc, sizeof(struct pci_chipset_tag));
	return EINVAL;
}

static void
x86_genfb_set_mapreg(void *opaque, int index, int r, int g, int b)
{
	outb(IO_VGA + VGA_DAC_ADDRW, index);
	outb(IO_VGA + VGA_DAC_PALETTE, (uint8_t)r >> 2);
	outb(IO_VGA + VGA_DAC_PALETTE, (uint8_t)g >> 2);
	outb(IO_VGA + VGA_DAC_PALETTE, (uint8_t)b >> 2);
}

static bool
x86_genfb_setmode(struct genfb_softc *sc, int newmode)
{
#if NGENFB > 0
# if NACPICA > 0 && defined(VGA_POST)
	static int curmode = WSDISPLAYIO_MODE_EMUL;
# endif

	switch (newmode) {
	case WSDISPLAYIO_MODE_EMUL:
		x86_genfb_mtrr_init(sc->sc_fboffset,
		    sc->sc_height * sc->sc_stride);
# if NACPICA > 0 && defined(VGA_POST)
		if (curmode != newmode) {
			if (vga_posth != NULL && acpi_md_vesa_modenum != 0) {
				vga_post_set_vbe(vga_posth,
				    acpi_md_vesa_modenum);
			}
		}
# endif
		break;
	}

# if NACPICA > 0 && defined(VGA_POST)
	curmode = newmode;
# endif
#endif
	return true;
}

static bool
x86_genfb_suspend(device_t dev, const pmf_qual_t *qual)
{
	return true;
}

static bool
x86_genfb_resume(device_t dev, const pmf_qual_t *qual)
{
#if NGENFB > 0
	struct pci_genfb_softc *psc = device_private(dev);

#if NACPICA > 0 && defined(VGA_POST)
	if (vga_posth != NULL && acpi_md_vbios_reset == 2) {
		vga_post_call(vga_posth);
		if (acpi_md_vesa_modenum != 0)
			vga_post_set_vbe(vga_posth, acpi_md_vesa_modenum);
	}
#endif
	genfb_restore_palette(&psc->sc_gen);
#endif

	return true;
}

device_t
device_pci_register(device_t dev, void *aux)
{
	static bool found_console = false;

	device_pci_props_register(dev, aux);

	/*
	 * Handle network interfaces here, the attachment information is
	 * not available driver-independently later.
	 *
	 * For disks, there is nothing useful available at attach time.
	 */
	if (device_class(dev) == DV_IFNET) {
		struct btinfo_netif *bin = lookup_bootinfo(BTINFO_NETIF);
		if (bin == NULL)
			return NULL;

		/*
		 * We don't check the driver name against the device name
		 * passed by the boot ROM.  The ROM should stay usable if
		 * the driver becomes obsolete.  The physical attachment
		 * information (checked below) must be sufficient to
		 * identify the device.
		 */
		if (bin->bus == BI_BUS_PCI &&
		    device_is_a(device_parent(dev), "pci")) {
			struct pci_attach_args *paa = aux;
			int b, d, f;

			/*
			 * Calculate BIOS representation of:
			 *
			 *	<bus,device,function>
			 *
			 * and compare.
			 */
			pci_decompose_tag(paa->pa_pc, paa->pa_tag, &b, &d, &f);
			if (bin->addr.tag == ((b << 8) | (d << 3) | f))
				return dev;
		}
	}
	if (device_parent(dev) && device_is_a(device_parent(dev), "pci") &&
	    found_console == false) {
		struct btinfo_framebuffer *fbinfo;
		struct pci_attach_args *pa = aux;
		prop_dictionary_t dict;

		if (PCI_CLASS(pa->pa_class) == PCI_CLASS_DISPLAY) {
#if NWSDISPLAY > 0 && NGENFB > 0
			extern struct vcons_screen x86_genfb_console_screen;
			struct rasops_info *ri;

			ri = &x86_genfb_console_screen.scr_ri;
#endif

			fbinfo = lookup_bootinfo(BTINFO_FRAMEBUFFER);
			dict = device_properties(dev);
			/*
			 * framebuffer drivers other than genfb can work
			 * without the address property
			 */
			if (fbinfo != NULL) {
				if (fbinfo->physaddr != 0) {
				prop_dictionary_set_uint32(dict, "width",
				    fbinfo->width);
				prop_dictionary_set_uint32(dict, "height",
				    fbinfo->height);
				prop_dictionary_set_uint8(dict, "depth",
				    fbinfo->depth);
				prop_dictionary_set_uint16(dict, "linebytes",
				    fbinfo->stride);

				prop_dictionary_set_uint64(dict, "address",
				    fbinfo->physaddr);
#if NWSDISPLAY > 0 && NGENFB > 0
				if (ri->ri_bits != NULL) {
					prop_dictionary_set_uint64(dict,
					    "virtual_address",
					    (vaddr_t)ri->ri_origbits);
				}
#endif
				}
#if notyet
				prop_dictionary_set_bool(dict, "splash",
				    fbinfo->flags & BI_FB_SPLASH ?
				     true : false);
#endif
				if (fbinfo->depth == 8) {
					gfb_cb.gcc_cookie = NULL;
					gfb_cb.gcc_set_mapreg =
					    x86_genfb_set_mapreg;
					prop_dictionary_set_uint64(dict,
					    "cmap_callback",
					    (uint64_t)(uintptr_t)&gfb_cb);
				}
				if (fbinfo->physaddr != 0) {
					mode_cb.gmc_setmode = x86_genfb_setmode;
					prop_dictionary_set_uint64(dict,
					    "mode_callback",
					    (uint64_t)(uintptr_t)&mode_cb);
				}

#if NWSDISPLAY > 0 && NGENFB > 0
				if (device_is_a(dev, "genfb")) {
					x86_genfb_set_console_dev(dev);
#ifdef DDB
					db_trap_callback =
					    x86_genfb_ddb_trap_callback;
#endif
				}
#endif
			}
#if 1 && NWSDISPLAY > 0 && NGENFB > 0
			/* XXX */
			if (device_is_a(dev, "genfb")) {
				prop_dictionary_set_bool(dict, "is_console",
				    genfb_is_console());
			} else
#endif
			prop_dictionary_set_bool(dict, "is_console", true);

			prop_dictionary_set_bool(dict, "clear-screen", false);
#if NWSDISPLAY > 0 && NGENFB > 0
			prop_dictionary_set_uint16(dict, "cursor-row",
			    x86_genfb_console_screen.scr_ri.ri_crow);
#endif
#if notyet
			prop_dictionary_set_bool(dict, "splash",
			    fbinfo->flags & BI_FB_SPLASH ? true : false);
#endif
			pmf_cb.gpc_suspend = x86_genfb_suspend;
			pmf_cb.gpc_resume = x86_genfb_resume;
			prop_dictionary_set_uint64(dict,
			    "pmf_callback", (uint64_t)(uintptr_t)&pmf_cb);
#ifdef VGA_POST
			vga_posth = vga_post_init(pa->pa_bus, pa->pa_device,
			    pa->pa_function);
#endif
			found_console = true;
			return NULL;
		}
	}
	return NULL;
}

#ifndef PUC_CNBUS
#define PUC_CNBUS 0
#endif

#if NCOM > 0
int
cpu_puc_cnprobe(struct consdev *cn, struct pci_attach_args *pa)
{
	pci_mode_detect();
	pa->pa_iot = x86_bus_space_io;
	pa->pa_memt = x86_bus_space_mem;
	pa->pa_pc = 0;
	pa->pa_tag = pci_make_tag(0, PUC_CNBUS, pci_bus_maxdevs(NULL, 0) - 1,
				  0);

	return 0;
}
#endif