linux/drivers/pci/probe.c

1669 lines
42 KiB
C
Raw Normal View History

/*
* probe.c - PCI detection and setup code
*/
#include <linux/kernel.h>
#include <linux/delay.h>
#include <linux/init.h>
#include <linux/pci.h>
#include <linux/slab.h>
#include <linux/module.h>
#include <linux/cpumask.h>
#include <linux/pci-aspm.h>
#include "pci.h"
#define CARDBUS_LATENCY_TIMER 176 /* secondary latency timer */
#define CARDBUS_RESERVE_BUSNR 3
/* Ugh. Need to stop exporting this to modules. */
LIST_HEAD(pci_root_buses);
EXPORT_SYMBOL(pci_root_buses);
static int find_anything(struct device *dev, void *data)
{
return 1;
}
/*
* Some device drivers need know if pci is initiated.
* Basically, we think pci is not initiated when there
* is no device to be found on the pci_bus_type.
*/
int no_pci_devices(void)
{
struct device *dev;
int no_devices;
dev = bus_find_device(&pci_bus_type, NULL, NULL, find_anything);
no_devices = (dev == NULL);
put_device(dev);
return no_devices;
}
EXPORT_SYMBOL(no_pci_devices);
/*
* PCI Bus Class
*/
static void release_pcibus_dev(struct device *dev)
{
struct pci_bus *pci_bus = to_pci_bus(dev);
if (pci_bus->bridge)
put_device(pci_bus->bridge);
pci_bus_remove_resources(pci_bus);
pci_release_bus_of_node(pci_bus);
kfree(pci_bus);
}
static struct class pcibus_class = {
.name = "pci_bus",
.dev_release = &release_pcibus_dev,
.dev_attrs = pcibus_dev_attrs,
};
static int __init pcibus_class_init(void)
{
return class_register(&pcibus_class);
}
postcore_initcall(pcibus_class_init);
static u64 pci_size(u64 base, u64 maxbase, u64 mask)
{
u64 size = mask & maxbase; /* Find the significant bits */
if (!size)
return 0;
/* Get the lowest of them to find the decode size, and
from that the extent. */
size = (size & ~(size-1)) - 1;
/* base == maxbase can be valid only if the BAR has
already been programmed with all 1s. */
if (base == maxbase && ((base | size) & mask) != mask)
return 0;
return size;
}
static inline unsigned long decode_bar(struct pci_dev *dev, u32 bar)
{
u32 mem_type;
unsigned long flags;
if ((bar & PCI_BASE_ADDRESS_SPACE) == PCI_BASE_ADDRESS_SPACE_IO) {
flags = bar & ~PCI_BASE_ADDRESS_IO_MASK;
flags |= IORESOURCE_IO;
return flags;
}
flags = bar & ~PCI_BASE_ADDRESS_MEM_MASK;
flags |= IORESOURCE_MEM;
if (flags & PCI_BASE_ADDRESS_MEM_PREFETCH)
flags |= IORESOURCE_PREFETCH;
mem_type = bar & PCI_BASE_ADDRESS_MEM_TYPE_MASK;
switch (mem_type) {
case PCI_BASE_ADDRESS_MEM_TYPE_32:
break;
case PCI_BASE_ADDRESS_MEM_TYPE_1M:
dev_info(&dev->dev, "1M mem BAR treated as 32-bit BAR\n");
break;
case PCI_BASE_ADDRESS_MEM_TYPE_64:
flags |= IORESOURCE_MEM_64;
break;
default:
dev_warn(&dev->dev,
"mem unknown type %x treated as 32-bit BAR\n",
mem_type);
break;
}
return flags;
}
/**
* pci_read_base - read a PCI BAR
* @dev: the PCI device
* @type: type of the BAR
* @res: resource buffer to be filled in
* @pos: BAR position in the config space
*
* Returns 1 if the BAR is 64-bit, or 0 if 32-bit.
*/
int __pci_read_base(struct pci_dev *dev, enum pci_bar_type type,
struct resource *res, unsigned int pos)
{
u32 l, sz, mask;
u16 orig_cmd;
mask = type ? PCI_ROM_ADDRESS_MASK : ~0;
if (!dev->mmio_always_on) {
pci_read_config_word(dev, PCI_COMMAND, &orig_cmd);
pci_write_config_word(dev, PCI_COMMAND,
orig_cmd & ~(PCI_COMMAND_MEMORY | PCI_COMMAND_IO));
}
res->name = pci_name(dev);
pci_read_config_dword(dev, pos, &l);
pci_write_config_dword(dev, pos, l | mask);
pci_read_config_dword(dev, pos, &sz);
pci_write_config_dword(dev, pos, l);
if (!dev->mmio_always_on)
pci_write_config_word(dev, PCI_COMMAND, orig_cmd);
/*
* All bits set in sz means the device isn't working properly.
* If the BAR isn't implemented, all bits must be 0. If it's a
* memory BAR or a ROM, bit 0 must be clear; if it's an io BAR, bit
* 1 must be clear.
*/
if (!sz || sz == 0xffffffff)
goto fail;
/*
* I don't know how l can have all bits set. Copied from old code.
* Maybe it fixes a bug on some ancient platform.
*/
if (l == 0xffffffff)
l = 0;
if (type == pci_bar_unknown) {
res->flags = decode_bar(dev, l);
res->flags |= IORESOURCE_SIZEALIGN;
if (res->flags & IORESOURCE_IO) {
l &= PCI_BASE_ADDRESS_IO_MASK;
mask = PCI_BASE_ADDRESS_IO_MASK & (u32) IO_SPACE_LIMIT;
} else {
l &= PCI_BASE_ADDRESS_MEM_MASK;
mask = (u32)PCI_BASE_ADDRESS_MEM_MASK;
}
} else {
res->flags |= (l & IORESOURCE_ROM_ENABLE);
l &= PCI_ROM_ADDRESS_MASK;
mask = (u32)PCI_ROM_ADDRESS_MASK;
}
if (res->flags & IORESOURCE_MEM_64) {
u64 l64 = l;
u64 sz64 = sz;
u64 mask64 = mask | (u64)~0 << 32;
pci_read_config_dword(dev, pos + 4, &l);
pci_write_config_dword(dev, pos + 4, ~0);
pci_read_config_dword(dev, pos + 4, &sz);
pci_write_config_dword(dev, pos + 4, l);
l64 |= ((u64)l << 32);
sz64 |= ((u64)sz << 32);
sz64 = pci_size(l64, sz64, mask64);
if (!sz64)
goto fail;
if ((sizeof(resource_size_t) < 8) && (sz64 > 0x100000000ULL)) {
dev_err(&dev->dev, "reg %x: can't handle 64-bit BAR\n",
pos);
goto fail;
}
if ((sizeof(resource_size_t) < 8) && l) {
/* Address above 32-bit boundary; disable the BAR */
pci_write_config_dword(dev, pos, 0);
pci_write_config_dword(dev, pos + 4, 0);
res->start = 0;
res->end = sz64;
} else {
res->start = l64;
res->end = l64 + sz64;
dev_printk(KERN_DEBUG, &dev->dev, "reg %x: %pR\n",
pos, res);
}
} else {
sz = pci_size(l, sz, mask);
if (!sz)
goto fail;
res->start = l;
res->end = l + sz;
dev_printk(KERN_DEBUG, &dev->dev, "reg %x: %pR\n", pos, res);
}
out:
return (res->flags & IORESOURCE_MEM_64) ? 1 : 0;
fail:
res->flags = 0;
goto out;
}
static void pci_read_bases(struct pci_dev *dev, unsigned int howmany, int rom)
{
unsigned int pos, reg;
for (pos = 0; pos < howmany; pos++) {
struct resource *res = &dev->resource[pos];
reg = PCI_BASE_ADDRESS_0 + (pos << 2);
pos += __pci_read_base(dev, pci_bar_unknown, res, reg);
}
if (rom) {
struct resource *res = &dev->resource[PCI_ROM_RESOURCE];
dev->rom_base_reg = rom;
res->flags = IORESOURCE_MEM | IORESOURCE_PREFETCH |
IORESOURCE_READONLY | IORESOURCE_CACHEABLE |
IORESOURCE_SIZEALIGN;
__pci_read_base(dev, pci_bar_mem32, res, rom);
}
}
static void __devinit pci_read_bridge_io(struct pci_bus *child)
{
struct pci_dev *dev = child->self;
u8 io_base_lo, io_limit_lo;
unsigned long base, limit;
struct resource *res;
res = child->resource[0];
pci_read_config_byte(dev, PCI_IO_BASE, &io_base_lo);
pci_read_config_byte(dev, PCI_IO_LIMIT, &io_limit_lo);
base = (io_base_lo & PCI_IO_RANGE_MASK) << 8;
limit = (io_limit_lo & PCI_IO_RANGE_MASK) << 8;
if ((io_base_lo & PCI_IO_RANGE_TYPE_MASK) == PCI_IO_RANGE_TYPE_32) {
u16 io_base_hi, io_limit_hi;
pci_read_config_word(dev, PCI_IO_BASE_UPPER16, &io_base_hi);
pci_read_config_word(dev, PCI_IO_LIMIT_UPPER16, &io_limit_hi);
base |= (io_base_hi << 16);
limit |= (io_limit_hi << 16);
}
if (base && base <= limit) {
res->flags = (io_base_lo & PCI_IO_RANGE_TYPE_MASK) | IORESOURCE_IO;
if (!res->start)
res->start = base;
if (!res->end)
res->end = limit + 0xfff;
dev_printk(KERN_DEBUG, &dev->dev, " bridge window %pR\n", res);
}
}
static void __devinit pci_read_bridge_mmio(struct pci_bus *child)
{
struct pci_dev *dev = child->self;
u16 mem_base_lo, mem_limit_lo;
unsigned long base, limit;
struct resource *res;
res = child->resource[1];
pci_read_config_word(dev, PCI_MEMORY_BASE, &mem_base_lo);
pci_read_config_word(dev, PCI_MEMORY_LIMIT, &mem_limit_lo);
base = (mem_base_lo & PCI_MEMORY_RANGE_MASK) << 16;
limit = (mem_limit_lo & PCI_MEMORY_RANGE_MASK) << 16;
if (base && base <= limit) {
res->flags = (mem_base_lo & PCI_MEMORY_RANGE_TYPE_MASK) | IORESOURCE_MEM;
res->start = base;
res->end = limit + 0xfffff;
dev_printk(KERN_DEBUG, &dev->dev, " bridge window %pR\n", res);
}
}
static void __devinit pci_read_bridge_mmio_pref(struct pci_bus *child)
{
struct pci_dev *dev = child->self;
u16 mem_base_lo, mem_limit_lo;
unsigned long base, limit;
struct resource *res;
res = child->resource[2];
pci_read_config_word(dev, PCI_PREF_MEMORY_BASE, &mem_base_lo);
pci_read_config_word(dev, PCI_PREF_MEMORY_LIMIT, &mem_limit_lo);
base = (mem_base_lo & PCI_PREF_RANGE_MASK) << 16;
limit = (mem_limit_lo & PCI_PREF_RANGE_MASK) << 16;
if ((mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) == PCI_PREF_RANGE_TYPE_64) {
u32 mem_base_hi, mem_limit_hi;
pci_read_config_dword(dev, PCI_PREF_BASE_UPPER32, &mem_base_hi);
pci_read_config_dword(dev, PCI_PREF_LIMIT_UPPER32, &mem_limit_hi);
/*
* Some bridges set the base > limit by default, and some
* (broken) BIOSes do not initialize them. If we find
* this, just assume they are not being used.
*/
if (mem_base_hi <= mem_limit_hi) {
#if BITS_PER_LONG == 64
base |= ((long) mem_base_hi) << 32;
limit |= ((long) mem_limit_hi) << 32;
#else
if (mem_base_hi || mem_limit_hi) {
dev_err(&dev->dev, "can't handle 64-bit "
"address space for bridge\n");
return;
}
#endif
}
}
if (base && base <= limit) {
res->flags = (mem_base_lo & PCI_PREF_RANGE_TYPE_MASK) |
IORESOURCE_MEM | IORESOURCE_PREFETCH;
if (res->flags & PCI_PREF_RANGE_TYPE_64)
res->flags |= IORESOURCE_MEM_64;
res->start = base;
res->end = limit + 0xfffff;
dev_printk(KERN_DEBUG, &dev->dev, " bridge window %pR\n", res);
}
}
void __devinit pci_read_bridge_bases(struct pci_bus *child)
{
struct pci_dev *dev = child->self;
struct resource *res;
int i;
if (pci_is_root_bus(child)) /* It's a host bus, nothing to read */
return;
dev_info(&dev->dev, "PCI bridge to [bus %02x-%02x]%s\n",
child->secondary, child->subordinate,
dev->transparent ? " (subtractive decode)" : "");
pci_bus_remove_resources(child);
for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++)
child->resource[i] = &dev->resource[PCI_BRIDGE_RESOURCES+i];
pci_read_bridge_io(child);
pci_read_bridge_mmio(child);
pci_read_bridge_mmio_pref(child);
if (dev->transparent) {
pci_bus_for_each_resource(child->parent, res, i) {
if (res) {
pci_bus_add_resource(child, res,
PCI_SUBTRACTIVE_DECODE);
dev_printk(KERN_DEBUG, &dev->dev,
" bridge window %pR (subtractive decode)\n",
res);
}
}
}
}
pci: do not mark exported functions as __devinit Functions marked __devinit will be removed after kernel init. But being exported they are potentially called by a module much later. So the safer choice seems to be to keep the function even in the non CONFIG_HOTPLUG case. This silence the follwoing section mismatch warnings: WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_add_device from __ksymtab_gpl between '__ksymtab_pci_bus_add_device' (at offset 0x20) and '__ksymtab_pci_walk_bus' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_create_bus from __ksymtab_gpl between '__ksymtab_pci_create_bus' (at offset 0x40) and '__ksymtab_pci_stop_bus_device' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_max_busnr from __ksymtab_gpl between '__ksymtab_pci_bus_max_busnr' (at offset 0xc0) and '__ksymtab_pci_assign_resource_fixed' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_claim_resource from __ksymtab_gpl between '__ksymtab_pci_claim_resource' (at offset 0xe0) and '__ksymtab_pcie_port_bus_type' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_add_devices from __ksymtab between '__ksymtab_pci_bus_add_devices' (at offset 0x70) and '__ksymtab_pci_bus_alloc_resource' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_scan_bus_parented from __ksymtab between '__ksymtab_pci_scan_bus_parented' (at offset 0x90) and '__ksymtab_pci_root_buses' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_assign_resources from __ksymtab between '__ksymtab_pci_bus_assign_resources' (at offset 0x4d0) and '__ksymtab_pci_bus_size_bridges' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_size_bridges from __ksymtab between '__ksymtab_pci_bus_size_bridges' (at offset 0x4e0) and '__ksymtab_pci_setup_cardbus' Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2007-03-27 07:53:30 +02:00
static struct pci_bus * pci_alloc_bus(void)
{
struct pci_bus *b;
b = kzalloc(sizeof(*b), GFP_KERNEL);
if (b) {
INIT_LIST_HEAD(&b->node);
INIT_LIST_HEAD(&b->children);
INIT_LIST_HEAD(&b->devices);
PCI: introduce pci_slot Currently, /sys/bus/pci/slots/ only exposes hotplug attributes when a hotplug driver is loaded, but PCI slots have attributes such as address, speed, width, etc. that are not related to hotplug at all. Introduce pci_slot as the primary data structure and kobject model. Hotplug attributes described in hotplug_slot become a secondary structure associated with the pci_slot. This patch only creates the infrastructure that allows the separation of PCI slot attributes and hotplug attributes. In this patch, the PCI hotplug core remains the only user of this infrastructure, and thus, /sys/bus/pci/slots/ will still only become populated when a hotplug driver is loaded. A later patch in this series will add a second user of this new infrastructure and demonstrate splitting the task of exposing pci_slot attributes from hotplug_slot attributes. - Make pci_slot the primary sysfs entity. hotplug_slot becomes a subsidiary structure. o pci_create_slot() creates and registers a slot with the PCI core o pci_slot_add_hotplug() gives it hotplug capability - Change the prototype of pci_hp_register() to take the bus and slot number (on parent bus) as parameters. - Remove all the ->get_address methods since this functionality is now handled by pci_slot directly. [achiang@hp.com: rpaphp-correctly-pci_hp_register-for-empty-pci-slots] Tested-by: Badari Pulavarty <pbadari@us.ibm.com> Acked-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> [akpm@linux-foundation.org: build fix] [akpm@linux-foundation.org: make headers_check happy] [akpm@linux-foundation.org: nuther build fix] [akpm@linux-foundation.org: fix typo in #include] Signed-off-by: Alex Chiang <achiang@hp.com> Signed-off-by: Matthew Wilcox <matthew@wil.cx> Cc: Greg KH <greg@kroah.com> Cc: Kristen Carlson Accardi <kristen.c.accardi@intel.com> Cc: Len Brown <lenb@kernel.org> Acked-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2008-06-10 23:28:50 +02:00
INIT_LIST_HEAD(&b->slots);
INIT_LIST_HEAD(&b->resources);
b->max_bus_speed = PCI_SPEED_UNKNOWN;
b->cur_bus_speed = PCI_SPEED_UNKNOWN;
}
return b;
}
static unsigned char pcix_bus_speed[] = {
PCI_SPEED_UNKNOWN, /* 0 */
PCI_SPEED_66MHz_PCIX, /* 1 */
PCI_SPEED_100MHz_PCIX, /* 2 */
PCI_SPEED_133MHz_PCIX, /* 3 */
PCI_SPEED_UNKNOWN, /* 4 */
PCI_SPEED_66MHz_PCIX_ECC, /* 5 */
PCI_SPEED_100MHz_PCIX_ECC, /* 6 */
PCI_SPEED_133MHz_PCIX_ECC, /* 7 */
PCI_SPEED_UNKNOWN, /* 8 */
PCI_SPEED_66MHz_PCIX_266, /* 9 */
PCI_SPEED_100MHz_PCIX_266, /* A */
PCI_SPEED_133MHz_PCIX_266, /* B */
PCI_SPEED_UNKNOWN, /* C */
PCI_SPEED_66MHz_PCIX_533, /* D */
PCI_SPEED_100MHz_PCIX_533, /* E */
PCI_SPEED_133MHz_PCIX_533 /* F */
};
static unsigned char pcie_link_speed[] = {
PCI_SPEED_UNKNOWN, /* 0 */
PCIE_SPEED_2_5GT, /* 1 */
PCIE_SPEED_5_0GT, /* 2 */
PCIE_SPEED_8_0GT, /* 3 */
PCI_SPEED_UNKNOWN, /* 4 */
PCI_SPEED_UNKNOWN, /* 5 */
PCI_SPEED_UNKNOWN, /* 6 */
PCI_SPEED_UNKNOWN, /* 7 */
PCI_SPEED_UNKNOWN, /* 8 */
PCI_SPEED_UNKNOWN, /* 9 */
PCI_SPEED_UNKNOWN, /* A */
PCI_SPEED_UNKNOWN, /* B */
PCI_SPEED_UNKNOWN, /* C */
PCI_SPEED_UNKNOWN, /* D */
PCI_SPEED_UNKNOWN, /* E */
PCI_SPEED_UNKNOWN /* F */
};
void pcie_update_link_speed(struct pci_bus *bus, u16 linksta)
{
bus->cur_bus_speed = pcie_link_speed[linksta & 0xf];
}
EXPORT_SYMBOL_GPL(pcie_update_link_speed);
static unsigned char agp_speeds[] = {
AGP_UNKNOWN,
AGP_1X,
AGP_2X,
AGP_4X,
AGP_8X
};
static enum pci_bus_speed agp_speed(int agp3, int agpstat)
{
int index = 0;
if (agpstat & 4)
index = 3;
else if (agpstat & 2)
index = 2;
else if (agpstat & 1)
index = 1;
else
goto out;
if (agp3) {
index += 2;
if (index == 5)
index = 0;
}
out:
return agp_speeds[index];
}
static void pci_set_bus_speed(struct pci_bus *bus)
{
struct pci_dev *bridge = bus->self;
int pos;
pos = pci_find_capability(bridge, PCI_CAP_ID_AGP);
if (!pos)
pos = pci_find_capability(bridge, PCI_CAP_ID_AGP3);
if (pos) {
u32 agpstat, agpcmd;
pci_read_config_dword(bridge, pos + PCI_AGP_STATUS, &agpstat);
bus->max_bus_speed = agp_speed(agpstat & 8, agpstat & 7);
pci_read_config_dword(bridge, pos + PCI_AGP_COMMAND, &agpcmd);
bus->cur_bus_speed = agp_speed(agpstat & 8, agpcmd & 7);
}
pos = pci_find_capability(bridge, PCI_CAP_ID_PCIX);
if (pos) {
u16 status;
enum pci_bus_speed max;
pci_read_config_word(bridge, pos + 2, &status);
if (status & 0x8000) {
max = PCI_SPEED_133MHz_PCIX_533;
} else if (status & 0x4000) {
max = PCI_SPEED_133MHz_PCIX_266;
} else if (status & 0x0002) {
if (((status >> 12) & 0x3) == 2) {
max = PCI_SPEED_133MHz_PCIX_ECC;
} else {
max = PCI_SPEED_133MHz_PCIX;
}
} else {
max = PCI_SPEED_66MHz_PCIX;
}
bus->max_bus_speed = max;
bus->cur_bus_speed = pcix_bus_speed[(status >> 6) & 0xf];
return;
}
pos = pci_find_capability(bridge, PCI_CAP_ID_EXP);
if (pos) {
u32 linkcap;
u16 linksta;
pci_read_config_dword(bridge, pos + PCI_EXP_LNKCAP, &linkcap);
bus->max_bus_speed = pcie_link_speed[linkcap & 0xf];
pci_read_config_word(bridge, pos + PCI_EXP_LNKSTA, &linksta);
pcie_update_link_speed(bus, linksta);
}
}
static struct pci_bus *pci_alloc_child_bus(struct pci_bus *parent,
struct pci_dev *bridge, int busnr)
{
struct pci_bus *child;
int i;
/*
* Allocate a new bus, and inherit stuff from the parent..
*/
child = pci_alloc_bus();
if (!child)
return NULL;
child->parent = parent;
child->ops = parent->ops;
child->sysdata = parent->sysdata;
child->bus_flags = parent->bus_flags;
/* initialize some portions of the bus device, but don't register it
* now as the parent is not properly set up yet. This device will get
* registered later in pci_bus_add_devices()
*/
child->dev.class = &pcibus_class;
dev_set_name(&child->dev, "%04x:%02x", pci_domain_nr(child), busnr);
/*
* Set up the primary, secondary and subordinate
* bus numbers.
*/
child->number = child->secondary = busnr;
child->primary = parent->secondary;
child->subordinate = 0xff;
if (!bridge)
return child;
child->self = bridge;
child->bridge = get_device(&bridge->dev);
pci_set_bus_of_node(child);
pci_set_bus_speed(child);
/* Set up default resource pointers and names.. */
for (i = 0; i < PCI_BRIDGE_RESOURCE_NUM; i++) {
child->resource[i] = &bridge->resource[PCI_BRIDGE_RESOURCES+i];
child->resource[i]->name = child->name;
}
bridge->subordinate = child;
return child;
}
struct pci_bus *__ref pci_add_new_bus(struct pci_bus *parent, struct pci_dev *dev, int busnr)
{
struct pci_bus *child;
child = pci_alloc_child_bus(parent, dev, busnr);
if (child) {
down_write(&pci_bus_sem);
list_add_tail(&child->node, &parent->children);
up_write(&pci_bus_sem);
}
return child;
}
pci: do not mark exported functions as __devinit Functions marked __devinit will be removed after kernel init. But being exported they are potentially called by a module much later. So the safer choice seems to be to keep the function even in the non CONFIG_HOTPLUG case. This silence the follwoing section mismatch warnings: WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_add_device from __ksymtab_gpl between '__ksymtab_pci_bus_add_device' (at offset 0x20) and '__ksymtab_pci_walk_bus' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_create_bus from __ksymtab_gpl between '__ksymtab_pci_create_bus' (at offset 0x40) and '__ksymtab_pci_stop_bus_device' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_max_busnr from __ksymtab_gpl between '__ksymtab_pci_bus_max_busnr' (at offset 0xc0) and '__ksymtab_pci_assign_resource_fixed' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_claim_resource from __ksymtab_gpl between '__ksymtab_pci_claim_resource' (at offset 0xe0) and '__ksymtab_pcie_port_bus_type' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_add_devices from __ksymtab between '__ksymtab_pci_bus_add_devices' (at offset 0x70) and '__ksymtab_pci_bus_alloc_resource' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_scan_bus_parented from __ksymtab between '__ksymtab_pci_scan_bus_parented' (at offset 0x90) and '__ksymtab_pci_root_buses' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_assign_resources from __ksymtab between '__ksymtab_pci_bus_assign_resources' (at offset 0x4d0) and '__ksymtab_pci_bus_size_bridges' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_size_bridges from __ksymtab between '__ksymtab_pci_bus_size_bridges' (at offset 0x4e0) and '__ksymtab_pci_setup_cardbus' Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2007-03-27 07:53:30 +02:00
static void pci_fixup_parent_subordinate_busnr(struct pci_bus *child, int max)
{
struct pci_bus *parent = child->parent;
/* Attempts to fix that up are really dangerous unless
we're going to re-assign all bus numbers. */
if (!pcibios_assign_all_busses())
return;
while (parent->parent && parent->subordinate < max) {
parent->subordinate = max;
pci_write_config_byte(parent->self, PCI_SUBORDINATE_BUS, max);
parent = parent->parent;
}
}
/*
* If it's a bridge, configure it and scan the bus behind it.
* For CardBus bridges, we don't scan behind as the devices will
* be handled by the bridge driver itself.
*
* We need to process bridges in two passes -- first we scan those
* already configured by the BIOS and after we are done with all of
* them, we proceed to assigning numbers to the remaining buses in
* order to avoid overlaps between old and new bus numbers.
*/
int __devinit pci_scan_bridge(struct pci_bus *bus, struct pci_dev *dev, int max, int pass)
{
struct pci_bus *child;
int is_cardbus = (dev->hdr_type == PCI_HEADER_TYPE_CARDBUS);
u32 buses, i, j = 0;
u16 bctl;
u8 primary, secondary, subordinate;
int broken = 0;
pci_read_config_dword(dev, PCI_PRIMARY_BUS, &buses);
primary = buses & 0xFF;
secondary = (buses >> 8) & 0xFF;
subordinate = (buses >> 16) & 0xFF;
dev_dbg(&dev->dev, "scanning [bus %02x-%02x] behind bridge, pass %d\n",
secondary, subordinate, pass);
/* Check if setup is sensible at all */
if (!pass &&
(primary != bus->number || secondary <= bus->number)) {
dev_dbg(&dev->dev, "bus configuration invalid, reconfiguring\n");
broken = 1;
}
/* Disable MasterAbortMode during probing to avoid reporting
of bus errors (in some architectures) */
pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl);
pci_write_config_word(dev, PCI_BRIDGE_CONTROL,
bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT);
if ((secondary || subordinate) && !pcibios_assign_all_busses() &&
!is_cardbus && !broken) {
unsigned int cmax;
/*
* Bus already configured by firmware, process it in the first
* pass and just note the configuration.
*/
if (pass)
[PATCH] PCI: Avoid leaving MASTER_ABORT disabled permanently when returning from pci_scan_bridge. > On Mon, Feb 13, 2006 at 05:13:21PM -0800, David S. Miller wrote: > > > > In drivers/pci/probe.c:pci_scan_bridge(), if this is not the first > > pass (pass != 0) we don't restore the PCI_BRIDGE_CONTROL_REGISTER and > > thus leave PCI_BRIDGE_CTL_MASTER_ABORT off: > > > > int __devinit pci_scan_bridge(struct pci_bus *bus, struct pci_dev * dev, int max, int pass) > > { > > ... > > /* Disable MasterAbortMode during probing to avoid reporting > > of bus errors (in some architectures) */ > > pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl); > > pci_write_config_word(dev, PCI_BRIDGE_CONTROL, > > bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT); > > ... > > if ((buses & 0xffff00) && !pcibios_assign_all_busses() && !is_cardbus) { > > unsigned int cmax, busnr; > > /* > > * Bus already configured by firmware, process it in the first > > * pass and just note the configuration. > > */ > > if (pass) > > return max; > > ... > > } > > > > pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl); > > ... > > > > This doesn't seem intentional. Agreed, looks like an accident. The patch [1] originally came from Kip Walker (Broadcom back then) between 2.6.0-test3 and 2.6.0-test4. As I recall it was supposed to fix an issue with with PCI aborts being signalled by the PCI bridge of the Broadcom BCM1250 family of SOCs when probing behind pci_scan_bridge. It is undeseriable to disable PCI_BRIDGE_CTL_MASTER_ABORT in pci_{read,write)_config_* and the behaviour wasn't considered a bug in need of a workaround, so this was put in probe.c. I don't have an affected system at hand, so can't really test but I propose something like the below patch. [1] http://www.linux-mips.org/git?p=linux.git;a=commit;h=599457e0cb702a31a3247ea6a5d9c6c99c4cf195 [PCI] Avoid leaving MASTER_ABORT disabled permanently when returning from pci_scan_bridge. Signed-off-by: Ralf Baechle <ralf@linux-mips.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2006-02-14 17:23:57 +01:00
goto out;
/*
* If we already got to this bus through a different bridge,
* don't re-add it. This can happen with the i450NX chipset.
*
* However, we continue to descend down the hierarchy and
* scan remaining child buses.
*/
child = pci_find_bus(pci_domain_nr(bus), secondary);
if (!child) {
child = pci_add_new_bus(bus, dev, secondary);
if (!child)
goto out;
child->primary = primary;
child->subordinate = subordinate;
child->bridge_ctl = bctl;
}
cmax = pci_scan_child_bus(child);
if (cmax > max)
max = cmax;
if (child->subordinate > max)
max = child->subordinate;
} else {
/*
* We need to assign a number to this bus which we always
* do in the second pass.
*/
if (!pass) {
if (pcibios_assign_all_busses() || broken)
/* Temporarily disable forwarding of the
configuration cycles on all bridges in
this bus segment to avoid possible
conflicts in the second pass between two
bridges programmed with overlapping
bus ranges. */
pci_write_config_dword(dev, PCI_PRIMARY_BUS,
buses & ~0xffffff);
[PATCH] PCI: Avoid leaving MASTER_ABORT disabled permanently when returning from pci_scan_bridge. > On Mon, Feb 13, 2006 at 05:13:21PM -0800, David S. Miller wrote: > > > > In drivers/pci/probe.c:pci_scan_bridge(), if this is not the first > > pass (pass != 0) we don't restore the PCI_BRIDGE_CONTROL_REGISTER and > > thus leave PCI_BRIDGE_CTL_MASTER_ABORT off: > > > > int __devinit pci_scan_bridge(struct pci_bus *bus, struct pci_dev * dev, int max, int pass) > > { > > ... > > /* Disable MasterAbortMode during probing to avoid reporting > > of bus errors (in some architectures) */ > > pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl); > > pci_write_config_word(dev, PCI_BRIDGE_CONTROL, > > bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT); > > ... > > if ((buses & 0xffff00) && !pcibios_assign_all_busses() && !is_cardbus) { > > unsigned int cmax, busnr; > > /* > > * Bus already configured by firmware, process it in the first > > * pass and just note the configuration. > > */ > > if (pass) > > return max; > > ... > > } > > > > pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl); > > ... > > > > This doesn't seem intentional. Agreed, looks like an accident. The patch [1] originally came from Kip Walker (Broadcom back then) between 2.6.0-test3 and 2.6.0-test4. As I recall it was supposed to fix an issue with with PCI aborts being signalled by the PCI bridge of the Broadcom BCM1250 family of SOCs when probing behind pci_scan_bridge. It is undeseriable to disable PCI_BRIDGE_CTL_MASTER_ABORT in pci_{read,write)_config_* and the behaviour wasn't considered a bug in need of a workaround, so this was put in probe.c. I don't have an affected system at hand, so can't really test but I propose something like the below patch. [1] http://www.linux-mips.org/git?p=linux.git;a=commit;h=599457e0cb702a31a3247ea6a5d9c6c99c4cf195 [PCI] Avoid leaving MASTER_ABORT disabled permanently when returning from pci_scan_bridge. Signed-off-by: Ralf Baechle <ralf@linux-mips.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2006-02-14 17:23:57 +01:00
goto out;
}
/* Clear errors */
pci_write_config_word(dev, PCI_STATUS, 0xffff);
/* Prevent assigning a bus number that already exists.
* This can happen when a bridge is hot-plugged, so in
* this case we only re-scan this bus. */
child = pci_find_bus(pci_domain_nr(bus), max+1);
if (!child) {
child = pci_add_new_bus(bus, dev, ++max);
if (!child)
goto out;
}
buses = (buses & 0xff000000)
| ((unsigned int)(child->primary) << 0)
| ((unsigned int)(child->secondary) << 8)
| ((unsigned int)(child->subordinate) << 16);
/*
* yenta.c forces a secondary latency timer of 176.
* Copy that behaviour here.
*/
if (is_cardbus) {
buses &= ~0xff000000;
buses |= CARDBUS_LATENCY_TIMER << 24;
}
/*
* We need to blast all three values with a single write.
*/
pci_write_config_dword(dev, PCI_PRIMARY_BUS, buses);
if (!is_cardbus) {
child->bridge_ctl = bctl;
/*
* Adjust subordinate busnr in parent buses.
* We do this before scanning for children because
* some devices may not be detected if the bios
* was lazy.
*/
pci_fixup_parent_subordinate_busnr(child, max);
/* Now we can scan all subordinate buses... */
max = pci_scan_child_bus(child);
/*
* now fix it up again since we have found
* the real value of max.
*/
pci_fixup_parent_subordinate_busnr(child, max);
} else {
/*
* For CardBus bridges, we leave 4 bus numbers
* as cards with a PCI-to-PCI bridge can be
* inserted later.
*/
for (i=0; i<CARDBUS_RESERVE_BUSNR; i++) {
struct pci_bus *parent = bus;
if (pci_find_bus(pci_domain_nr(bus),
max+i+1))
break;
while (parent->parent) {
if ((!pcibios_assign_all_busses()) &&
(parent->subordinate > max) &&
(parent->subordinate <= max+i)) {
j = 1;
}
parent = parent->parent;
}
if (j) {
/*
* Often, there are two cardbus bridges
* -- try to leave one valid bus number
* for each one.
*/
i /= 2;
break;
}
}
max += i;
pci_fixup_parent_subordinate_busnr(child, max);
}
/*
* Set the subordinate bus number to its real value.
*/
child->subordinate = max;
pci_write_config_byte(dev, PCI_SUBORDINATE_BUS, max);
}
sprintf(child->name,
(is_cardbus ? "PCI CardBus %04x:%02x" : "PCI Bus %04x:%02x"),
pci_domain_nr(bus), child->number);
PCI: lets kill the 'PCI hidden behind bridge' message Adrian Bunk wrote: > Alois Nešpor wrote >> PCI: Bus #0b (-#0e) is hidden behind transparent bridge #0a (-#0b) (try 'pci=assign-busses') >> Please report the result to linux-kernel to fix this permanently" >> >> dmesg: >> "Yenta: Raising subordinate bus# of parent bus (#0a) from #0b to #0e" >> without pci=assign-busses and nothing with pci=assign-busses. > > Bernhard? Ok, lets kill the message. As Alois Nešpor also saw, that's fixed up by Yenta, so PCI does not have to warn about it. PCI could still warn about it if is_cardbus is 0 in that instance of pci_scan_bridge(), but so far I have not seen a report where this would have been the case so I think we can spare the kernel of that check (removes ~300 lines of asm) unless debugging is done. History: The whole check was added in the days before we had the fixup for this in Yenta and pci=assign-busses was the only way to get CardBus cards detected on many (not all) of the machines which give this warning. In theory, there could be cases when this warning would be triggered and it's not cardbus, then the warning should still apply, but I think this should only be the case when working on a completely broken PCI setup, but one may have already enabled the debug code in drivers/pci and the patched check would then trigger. I do not sign this off yet because it's completely untested so far, but everyone is free to test it (with the #ifdef DEBUG replaced by #if 1 and pr_debug( changed to printk(. We may also dump the whole check (remove everything within the #ifdef from the source) if that's perferred. On Alois Nešpor's machine this would then (only when debugging) this message: "PCI: Bus #0b (-#0e) is partially hidden behind transparent bridge #0a (-#0b)" "partially" should be in the message on his machine because #0b of #0b-#0e is reachable behind #0a-#0b, but not #0c-#0e. But that differentiation is now moot anyway because the fixup in Yenta takes care of it as far as I could see so far, which means that unless somebody is debugging a totally broken PCI setup, this message is not needed anymore, not even for debugging PCI. Ok, here the patch with the following changes: * Refined to say that the bus is only partially hidden when the parent bus numbers are not totally way off (outside of) the child bus range * remove the reference to pci=assign-busses and the plea to report it We could add a pure source code-only comment to keep a reference to pci=assign-busses the in case when this is triggered by someone who is debugging the cause of this message and looking the way to solve it. From: Bernhard Kaindl <bk@suse.de> Cc: stable <stable@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2007-07-30 20:35:13 +02:00
/* Has only triggered on CardBus, fixup is in yenta_socket */
while (bus->parent) {
if ((child->subordinate > bus->subordinate) ||
(child->number > bus->subordinate) ||
(child->number < bus->number) ||
(child->subordinate < bus->number)) {
dev_info(&child->dev, "[bus %02x-%02x] %s "
"hidden behind%s bridge %s [bus %02x-%02x]\n",
PCI: lets kill the 'PCI hidden behind bridge' message Adrian Bunk wrote: > Alois Nešpor wrote >> PCI: Bus #0b (-#0e) is hidden behind transparent bridge #0a (-#0b) (try 'pci=assign-busses') >> Please report the result to linux-kernel to fix this permanently" >> >> dmesg: >> "Yenta: Raising subordinate bus# of parent bus (#0a) from #0b to #0e" >> without pci=assign-busses and nothing with pci=assign-busses. > > Bernhard? Ok, lets kill the message. As Alois Nešpor also saw, that's fixed up by Yenta, so PCI does not have to warn about it. PCI could still warn about it if is_cardbus is 0 in that instance of pci_scan_bridge(), but so far I have not seen a report where this would have been the case so I think we can spare the kernel of that check (removes ~300 lines of asm) unless debugging is done. History: The whole check was added in the days before we had the fixup for this in Yenta and pci=assign-busses was the only way to get CardBus cards detected on many (not all) of the machines which give this warning. In theory, there could be cases when this warning would be triggered and it's not cardbus, then the warning should still apply, but I think this should only be the case when working on a completely broken PCI setup, but one may have already enabled the debug code in drivers/pci and the patched check would then trigger. I do not sign this off yet because it's completely untested so far, but everyone is free to test it (with the #ifdef DEBUG replaced by #if 1 and pr_debug( changed to printk(. We may also dump the whole check (remove everything within the #ifdef from the source) if that's perferred. On Alois Nešpor's machine this would then (only when debugging) this message: "PCI: Bus #0b (-#0e) is partially hidden behind transparent bridge #0a (-#0b)" "partially" should be in the message on his machine because #0b of #0b-#0e is reachable behind #0a-#0b, but not #0c-#0e. But that differentiation is now moot anyway because the fixup in Yenta takes care of it as far as I could see so far, which means that unless somebody is debugging a totally broken PCI setup, this message is not needed anymore, not even for debugging PCI. Ok, here the patch with the following changes: * Refined to say that the bus is only partially hidden when the parent bus numbers are not totally way off (outside of) the child bus range * remove the reference to pci=assign-busses and the plea to report it We could add a pure source code-only comment to keep a reference to pci=assign-busses the in case when this is triggered by someone who is debugging the cause of this message and looking the way to solve it. From: Bernhard Kaindl <bk@suse.de> Cc: stable <stable@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2007-07-30 20:35:13 +02:00
child->number, child->subordinate,
(bus->number > child->subordinate &&
bus->subordinate < child->number) ?
"wholly" : "partially",
bus->self->transparent ? " transparent" : "",
dev_name(&bus->dev),
PCI: lets kill the 'PCI hidden behind bridge' message Adrian Bunk wrote: > Alois Nešpor wrote >> PCI: Bus #0b (-#0e) is hidden behind transparent bridge #0a (-#0b) (try 'pci=assign-busses') >> Please report the result to linux-kernel to fix this permanently" >> >> dmesg: >> "Yenta: Raising subordinate bus# of parent bus (#0a) from #0b to #0e" >> without pci=assign-busses and nothing with pci=assign-busses. > > Bernhard? Ok, lets kill the message. As Alois Nešpor also saw, that's fixed up by Yenta, so PCI does not have to warn about it. PCI could still warn about it if is_cardbus is 0 in that instance of pci_scan_bridge(), but so far I have not seen a report where this would have been the case so I think we can spare the kernel of that check (removes ~300 lines of asm) unless debugging is done. History: The whole check was added in the days before we had the fixup for this in Yenta and pci=assign-busses was the only way to get CardBus cards detected on many (not all) of the machines which give this warning. In theory, there could be cases when this warning would be triggered and it's not cardbus, then the warning should still apply, but I think this should only be the case when working on a completely broken PCI setup, but one may have already enabled the debug code in drivers/pci and the patched check would then trigger. I do not sign this off yet because it's completely untested so far, but everyone is free to test it (with the #ifdef DEBUG replaced by #if 1 and pr_debug( changed to printk(. We may also dump the whole check (remove everything within the #ifdef from the source) if that's perferred. On Alois Nešpor's machine this would then (only when debugging) this message: "PCI: Bus #0b (-#0e) is partially hidden behind transparent bridge #0a (-#0b)" "partially" should be in the message on his machine because #0b of #0b-#0e is reachable behind #0a-#0b, but not #0c-#0e. But that differentiation is now moot anyway because the fixup in Yenta takes care of it as far as I could see so far, which means that unless somebody is debugging a totally broken PCI setup, this message is not needed anymore, not even for debugging PCI. Ok, here the patch with the following changes: * Refined to say that the bus is only partially hidden when the parent bus numbers are not totally way off (outside of) the child bus range * remove the reference to pci=assign-busses and the plea to report it We could add a pure source code-only comment to keep a reference to pci=assign-busses the in case when this is triggered by someone who is debugging the cause of this message and looking the way to solve it. From: Bernhard Kaindl <bk@suse.de> Cc: stable <stable@kernel.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2007-07-30 20:35:13 +02:00
bus->number, bus->subordinate);
}
bus = bus->parent;
}
[PATCH] PCI: Avoid leaving MASTER_ABORT disabled permanently when returning from pci_scan_bridge. > On Mon, Feb 13, 2006 at 05:13:21PM -0800, David S. Miller wrote: > > > > In drivers/pci/probe.c:pci_scan_bridge(), if this is not the first > > pass (pass != 0) we don't restore the PCI_BRIDGE_CONTROL_REGISTER and > > thus leave PCI_BRIDGE_CTL_MASTER_ABORT off: > > > > int __devinit pci_scan_bridge(struct pci_bus *bus, struct pci_dev * dev, int max, int pass) > > { > > ... > > /* Disable MasterAbortMode during probing to avoid reporting > > of bus errors (in some architectures) */ > > pci_read_config_word(dev, PCI_BRIDGE_CONTROL, &bctl); > > pci_write_config_word(dev, PCI_BRIDGE_CONTROL, > > bctl & ~PCI_BRIDGE_CTL_MASTER_ABORT); > > ... > > if ((buses & 0xffff00) && !pcibios_assign_all_busses() && !is_cardbus) { > > unsigned int cmax, busnr; > > /* > > * Bus already configured by firmware, process it in the first > > * pass and just note the configuration. > > */ > > if (pass) > > return max; > > ... > > } > > > > pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl); > > ... > > > > This doesn't seem intentional. Agreed, looks like an accident. The patch [1] originally came from Kip Walker (Broadcom back then) between 2.6.0-test3 and 2.6.0-test4. As I recall it was supposed to fix an issue with with PCI aborts being signalled by the PCI bridge of the Broadcom BCM1250 family of SOCs when probing behind pci_scan_bridge. It is undeseriable to disable PCI_BRIDGE_CTL_MASTER_ABORT in pci_{read,write)_config_* and the behaviour wasn't considered a bug in need of a workaround, so this was put in probe.c. I don't have an affected system at hand, so can't really test but I propose something like the below patch. [1] http://www.linux-mips.org/git?p=linux.git;a=commit;h=599457e0cb702a31a3247ea6a5d9c6c99c4cf195 [PCI] Avoid leaving MASTER_ABORT disabled permanently when returning from pci_scan_bridge. Signed-off-by: Ralf Baechle <ralf@linux-mips.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2006-02-14 17:23:57 +01:00
out:
pci_write_config_word(dev, PCI_BRIDGE_CONTROL, bctl);
return max;
}
/*
* Read interrupt line and base address registers.
* The architecture-dependent code can tweak these, of course.
*/
static void pci_read_irq(struct pci_dev *dev)
{
unsigned char irq;
pci_read_config_byte(dev, PCI_INTERRUPT_PIN, &irq);
dev->pin = irq;
if (irq)
pci_read_config_byte(dev, PCI_INTERRUPT_LINE, &irq);
dev->irq = irq;
}
void set_pcie_port_type(struct pci_dev *pdev)
{
int pos;
u16 reg16;
pos = pci_find_capability(pdev, PCI_CAP_ID_EXP);
if (!pos)
return;
pdev->is_pcie = 1;
pdev->pcie_cap = pos;
pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, &reg16);
pdev->pcie_type = (reg16 & PCI_EXP_FLAGS_TYPE) >> 4;
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
pci_read_config_word(pdev, pos + PCI_EXP_DEVCAP, &reg16);
pdev->pcie_mpss = reg16 & PCI_EXP_DEVCAP_PAYLOAD;
}
void set_pcie_hotplug_bridge(struct pci_dev *pdev)
{
int pos;
u16 reg16;
u32 reg32;
pos = pci_pcie_cap(pdev);
if (!pos)
return;
pci_read_config_word(pdev, pos + PCI_EXP_FLAGS, &reg16);
if (!(reg16 & PCI_EXP_FLAGS_SLOT))
return;
pci_read_config_dword(pdev, pos + PCI_EXP_SLTCAP, &reg32);
if (reg32 & PCI_EXP_SLTCAP_HPC)
pdev->is_hotplug_bridge = 1;
}
#define LEGACY_IO_RESOURCE (IORESOURCE_IO | IORESOURCE_PCI_FIXED)
/**
* pci_setup_device - fill in class and map information of a device
* @dev: the device structure to fill
*
* Initialize the device structure with information about the device's
* vendor,class,memory and IO-space addresses,IRQ lines etc.
* Called at initialisation of the PCI subsystem and by CardBus services.
* Returns 0 on success and negative if unknown type of device (not normal,
* bridge or CardBus).
*/
int pci_setup_device(struct pci_dev *dev)
{
u32 class;
u8 hdr_type;
struct pci_slot *slot;
int pos = 0;
if (pci_read_config_byte(dev, PCI_HEADER_TYPE, &hdr_type))
return -EIO;
dev->sysdata = dev->bus->sysdata;
dev->dev.parent = dev->bus->bridge;
dev->dev.bus = &pci_bus_type;
dev->hdr_type = hdr_type & 0x7f;
dev->multifunction = !!(hdr_type & 0x80);
dev->error_state = pci_channel_io_normal;
set_pcie_port_type(dev);
list_for_each_entry(slot, &dev->bus->slots, list)
if (PCI_SLOT(dev->devfn) == slot->number)
dev->slot = slot;
/* Assume 32-bit PCI; let 64-bit PCI cards (which are far rarer)
set this higher, assuming the system even supports it. */
dev->dma_mask = 0xffffffff;
dev_set_name(&dev->dev, "%04x:%02x:%02x.%d", pci_domain_nr(dev->bus),
dev->bus->number, PCI_SLOT(dev->devfn),
PCI_FUNC(dev->devfn));
pci_read_config_dword(dev, PCI_CLASS_REVISION, &class);
dev->revision = class & 0xff;
class >>= 8; /* upper 3 bytes */
dev->class = class;
class >>= 8;
dev_printk(KERN_DEBUG, &dev->dev, "[%04x:%04x] type %d class %#08x\n",
dev->vendor, dev->device, dev->hdr_type, class);
/* need to have dev->class ready */
dev->cfg_size = pci_cfg_space_size(dev);
/* "Unknown power state" */
dev->current_state = PCI_UNKNOWN;
/* Early fixups, before probing the BARs */
pci_fixup_device(pci_fixup_early, dev);
/* device class may be changed after fixup */
class = dev->class >> 8;
switch (dev->hdr_type) { /* header type */
case PCI_HEADER_TYPE_NORMAL: /* standard header */
if (class == PCI_CLASS_BRIDGE_PCI)
goto bad;
pci_read_irq(dev);
pci_read_bases(dev, 6, PCI_ROM_ADDRESS);
pci_read_config_word(dev, PCI_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
pci_read_config_word(dev, PCI_SUBSYSTEM_ID, &dev->subsystem_device);
/*
* Do the ugly legacy mode stuff here rather than broken chip
* quirk code. Legacy mode ATA controllers have fixed
* addresses. These are not always echoed in BAR0-3, and
* BAR0-3 in a few cases contain junk!
*/
if (class == PCI_CLASS_STORAGE_IDE) {
u8 progif;
pci_read_config_byte(dev, PCI_CLASS_PROG, &progif);
if ((progif & 1) == 0) {
dev->resource[0].start = 0x1F0;
dev->resource[0].end = 0x1F7;
dev->resource[0].flags = LEGACY_IO_RESOURCE;
dev->resource[1].start = 0x3F6;
dev->resource[1].end = 0x3F6;
dev->resource[1].flags = LEGACY_IO_RESOURCE;
}
if ((progif & 4) == 0) {
dev->resource[2].start = 0x170;
dev->resource[2].end = 0x177;
dev->resource[2].flags = LEGACY_IO_RESOURCE;
dev->resource[3].start = 0x376;
dev->resource[3].end = 0x376;
dev->resource[3].flags = LEGACY_IO_RESOURCE;
}
}
break;
case PCI_HEADER_TYPE_BRIDGE: /* bridge header */
if (class != PCI_CLASS_BRIDGE_PCI)
goto bad;
/* The PCI-to-PCI bridge spec requires that subtractive
decoding (i.e. transparent) bridge must have programming
interface code of 0x01. */
pci_read_irq(dev);
dev->transparent = ((dev->class & 0xff) == 1);
pci_read_bases(dev, 2, PCI_ROM_ADDRESS1);
set_pcie_hotplug_bridge(dev);
pos = pci_find_capability(dev, PCI_CAP_ID_SSVID);
if (pos) {
pci_read_config_word(dev, pos + PCI_SSVID_VENDOR_ID, &dev->subsystem_vendor);
pci_read_config_word(dev, pos + PCI_SSVID_DEVICE_ID, &dev->subsystem_device);
}
break;
case PCI_HEADER_TYPE_CARDBUS: /* CardBus bridge header */
if (class != PCI_CLASS_BRIDGE_CARDBUS)
goto bad;
pci_read_irq(dev);
pci_read_bases(dev, 1, 0);
pci_read_config_word(dev, PCI_CB_SUBSYSTEM_VENDOR_ID, &dev->subsystem_vendor);
pci_read_config_word(dev, PCI_CB_SUBSYSTEM_ID, &dev->subsystem_device);
break;
default: /* unknown header */
dev_err(&dev->dev, "unknown header type %02x, "
"ignoring device\n", dev->hdr_type);
return -EIO;
bad:
dev_err(&dev->dev, "ignoring class %02x (doesn't match header "
"type %02x)\n", class, dev->hdr_type);
dev->class = PCI_CLASS_NOT_DEFINED;
}
/* We found a fine healthy device, go go go... */
return 0;
}
static void pci_release_capabilities(struct pci_dev *dev)
{
pci_vpd_release(dev);
pci_iov_release(dev);
}
/**
* pci_release_dev - free a pci device structure when all users of it are finished.
* @dev: device that's been disconnected
*
* Will be called only by the device core when all users of this pci device are
* done.
*/
static void pci_release_dev(struct device *dev)
{
struct pci_dev *pci_dev;
pci_dev = to_pci_dev(dev);
pci_release_capabilities(pci_dev);
pci_release_of_node(pci_dev);
kfree(pci_dev);
}
/**
* pci_cfg_space_size - get the configuration space size of the PCI device.
* @dev: PCI device
*
* Regular PCI devices have 256 bytes, but PCI-X 2 and PCI Express devices
* have 4096 bytes. Even if the device is capable, that doesn't mean we can
* access it. Maybe we don't have a way to generate extended config space
* accesses, or the device is behind a reverse Express bridge. So we try
* reading the dword at 0x100 which must either be 0 or a valid extended
* capability header.
*/
int pci_cfg_space_size_ext(struct pci_dev *dev)
{
u32 status;
int pos = PCI_CFG_SPACE_SIZE;
if (pci_read_config_dword(dev, pos, &status) != PCIBIOS_SUCCESSFUL)
goto fail;
if (status == 0xffffffff)
goto fail;
return PCI_CFG_SPACE_EXP_SIZE;
fail:
return PCI_CFG_SPACE_SIZE;
}
int pci_cfg_space_size(struct pci_dev *dev)
{
int pos;
u32 status;
u16 class;
class = dev->class >> 8;
if (class == PCI_CLASS_BRIDGE_HOST)
return pci_cfg_space_size_ext(dev);
pos = pci_pcie_cap(dev);
if (!pos) {
pos = pci_find_capability(dev, PCI_CAP_ID_PCIX);
if (!pos)
goto fail;
pci_read_config_dword(dev, pos + PCI_X_STATUS, &status);
if (!(status & (PCI_X_STATUS_266MHZ | PCI_X_STATUS_533MHZ)))
goto fail;
}
return pci_cfg_space_size_ext(dev);
fail:
return PCI_CFG_SPACE_SIZE;
}
static void pci_release_bus_bridge_dev(struct device *dev)
{
kfree(dev);
}
struct pci_dev *alloc_pci_dev(void)
{
struct pci_dev *dev;
dev = kzalloc(sizeof(struct pci_dev), GFP_KERNEL);
if (!dev)
return NULL;
INIT_LIST_HEAD(&dev->bus_list);
return dev;
}
EXPORT_SYMBOL(alloc_pci_dev);
/*
* Read the config data for a PCI device, sanity-check it
* and fill in the dev structure...
*/
static struct pci_dev *pci_scan_device(struct pci_bus *bus, int devfn)
{
struct pci_dev *dev;
u32 l;
int delay = 1;
if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, &l))
return NULL;
/* some broken boards return 0 or ~0 if a slot is empty: */
if (l == 0xffffffff || l == 0x00000000 ||
l == 0x0000ffff || l == 0xffff0000)
return NULL;
/* Configuration request Retry Status */
while (l == 0xffff0001) {
msleep(delay);
delay *= 2;
if (pci_bus_read_config_dword(bus, devfn, PCI_VENDOR_ID, &l))
return NULL;
/* Card hasn't responded in 60 seconds? Must be stuck. */
if (delay > 60 * 1000) {
printk(KERN_WARNING "pci %04x:%02x:%02x.%d: not "
"responding\n", pci_domain_nr(bus),
bus->number, PCI_SLOT(devfn),
PCI_FUNC(devfn));
return NULL;
}
}
dev = alloc_pci_dev();
if (!dev)
return NULL;
dev->bus = bus;
dev->devfn = devfn;
dev->vendor = l & 0xffff;
dev->device = (l >> 16) & 0xffff;
pci_set_of_node(dev);
if (pci_setup_device(dev)) {
kfree(dev);
return NULL;
}
return dev;
}
static void pci_init_capabilities(struct pci_dev *dev)
{
/* MSI/MSI-X list */
pci_msi_init_pci_dev(dev);
/* Buffers for saving PCIe and PCI-X capabilities */
pci_allocate_cap_save_buffers(dev);
/* Power Management */
pci_pm_init(dev);
platform_pci_wakeup_init(dev);
/* Vital Product Data */
pci_vpd_pci22_init(dev);
/* Alternative Routing-ID Forwarding */
pci_enable_ari(dev);
/* Single Root I/O Virtualization */
pci_iov_init(dev);
/* Enable ACS P2P upstream forwarding */
pci_enable_acs(dev);
}
pci: do not mark exported functions as __devinit Functions marked __devinit will be removed after kernel init. But being exported they are potentially called by a module much later. So the safer choice seems to be to keep the function even in the non CONFIG_HOTPLUG case. This silence the follwoing section mismatch warnings: WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_add_device from __ksymtab_gpl between '__ksymtab_pci_bus_add_device' (at offset 0x20) and '__ksymtab_pci_walk_bus' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_create_bus from __ksymtab_gpl between '__ksymtab_pci_create_bus' (at offset 0x40) and '__ksymtab_pci_stop_bus_device' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_max_busnr from __ksymtab_gpl between '__ksymtab_pci_bus_max_busnr' (at offset 0xc0) and '__ksymtab_pci_assign_resource_fixed' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_claim_resource from __ksymtab_gpl between '__ksymtab_pci_claim_resource' (at offset 0xe0) and '__ksymtab_pcie_port_bus_type' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_add_devices from __ksymtab between '__ksymtab_pci_bus_add_devices' (at offset 0x70) and '__ksymtab_pci_bus_alloc_resource' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_scan_bus_parented from __ksymtab between '__ksymtab_pci_scan_bus_parented' (at offset 0x90) and '__ksymtab_pci_root_buses' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_assign_resources from __ksymtab between '__ksymtab_pci_bus_assign_resources' (at offset 0x4d0) and '__ksymtab_pci_bus_size_bridges' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_size_bridges from __ksymtab between '__ksymtab_pci_bus_size_bridges' (at offset 0x4e0) and '__ksymtab_pci_setup_cardbus' Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2007-03-27 07:53:30 +02:00
void pci_device_add(struct pci_dev *dev, struct pci_bus *bus)
{
device_initialize(&dev->dev);
dev->dev.release = pci_release_dev;
pci_dev_get(dev);
dev->dev.dma_mask = &dev->dma_mask;
dev->dev.dma_parms = &dev->dma_parms;
dev->dev.coherent_dma_mask = 0xffffffffull;
pci_set_dma_max_seg_size(dev, 65536);
pci_set_dma_seg_boundary(dev, 0xffffffff);
/* Fix up broken headers */
pci_fixup_device(pci_fixup_header, dev);
/* Clear the state_saved flag. */
dev->state_saved = false;
/* Initialize various capabilities */
pci_init_capabilities(dev);
PCI ACPI: Rework PCI handling of wake-up * Introduce function acpi_pm_device_sleep_wake() for enabling and disabling the system wake-up capability of devices that are power manageable by ACPI. * Introduce function acpi_bus_can_wakeup() allowing other (dependent) subsystems to check if ACPI is able to enable the system wake-up capability of given device. * Introduce callback .sleep_wake() in struct pci_platform_pm_ops and for the ACPI PCI 'driver' make it use acpi_pm_device_sleep_wake(). * Introduce callback .can_wakeup() in struct pci_platform_pm_ops and for the ACPI 'driver' make it use acpi_bus_can_wakeup(). * Move the PME# handlig code out of pci_enable_wake() and split it into two functions, pci_pme_capable() and pci_pme_active(), allowing the caller to check if given device is capable of generating PME# from given power state and to enable/disable the device's PME# functionality, respectively. * Modify pci_enable_wake() to use the new ACPI callbacks and the new PME#-related functions. * Drop the generic .platform_enable_wakeup() callback that is not used any more. * Introduce device_set_wakeup_capable() that will set the power.can_wakeup flag of given device. * Rework PCI device PM initialization so that, if given device is capable of generating wake-up events, either natively through the PME# mechanism, or with the help of the platform, its power.can_wakeup flag is set and its power.should_wakeup flag is unset as appropriate. * Make ACPI set the power.can_wakeup flag for devices found to be wake-up capable by it. * Make the ACPI wake-up code enable/disable GPEs for devices that have the wakeup.flags.prepared flag set (which means that their wake-up power has been enabled). Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2008-07-07 03:34:48 +02:00
/*
* Add the device to our list of discovered devices
* and the bus list for fixup functions, etc.
*/
down_write(&pci_bus_sem);
list_add_tail(&dev->bus_list, &bus->devices);
up_write(&pci_bus_sem);
}
struct pci_dev *__ref pci_scan_single_device(struct pci_bus *bus, int devfn)
{
struct pci_dev *dev;
dev = pci_get_slot(bus, devfn);
if (dev) {
pci_dev_put(dev);
return dev;
}
dev = pci_scan_device(bus, devfn);
if (!dev)
return NULL;
pci_device_add(dev, bus);
return dev;
}
EXPORT_SYMBOL(pci_scan_single_device);
static unsigned next_ari_fn(struct pci_dev *dev, unsigned fn)
{
u16 cap;
unsigned pos, next_fn;
if (!dev)
return 0;
pos = pci_find_ext_capability(dev, PCI_EXT_CAP_ID_ARI);
if (!pos)
return 0;
pci_read_config_word(dev, pos + 4, &cap);
next_fn = cap >> 8;
if (next_fn <= fn)
return 0;
return next_fn;
}
static unsigned next_trad_fn(struct pci_dev *dev, unsigned fn)
{
return (fn + 1) % 8;
}
static unsigned no_next_fn(struct pci_dev *dev, unsigned fn)
{
return 0;
}
static int only_one_child(struct pci_bus *bus)
{
struct pci_dev *parent = bus->self;
if (!parent || !pci_is_pcie(parent))
return 0;
if (parent->pcie_type == PCI_EXP_TYPE_ROOT_PORT ||
parent->pcie_type == PCI_EXP_TYPE_DOWNSTREAM)
return 1;
return 0;
}
/**
* pci_scan_slot - scan a PCI slot on a bus for devices.
* @bus: PCI bus to scan
* @devfn: slot number to scan (must have zero function.)
*
* Scan a PCI slot on the specified PCI bus for devices, adding
* discovered devices to the @bus->devices list. New devices
* will not have is_added set.
*
* Returns the number of new devices found.
*/
pci: do not mark exported functions as __devinit Functions marked __devinit will be removed after kernel init. But being exported they are potentially called by a module much later. So the safer choice seems to be to keep the function even in the non CONFIG_HOTPLUG case. This silence the follwoing section mismatch warnings: WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_add_device from __ksymtab_gpl between '__ksymtab_pci_bus_add_device' (at offset 0x20) and '__ksymtab_pci_walk_bus' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_create_bus from __ksymtab_gpl between '__ksymtab_pci_create_bus' (at offset 0x40) and '__ksymtab_pci_stop_bus_device' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_max_busnr from __ksymtab_gpl between '__ksymtab_pci_bus_max_busnr' (at offset 0xc0) and '__ksymtab_pci_assign_resource_fixed' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_claim_resource from __ksymtab_gpl between '__ksymtab_pci_claim_resource' (at offset 0xe0) and '__ksymtab_pcie_port_bus_type' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_add_devices from __ksymtab between '__ksymtab_pci_bus_add_devices' (at offset 0x70) and '__ksymtab_pci_bus_alloc_resource' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_scan_bus_parented from __ksymtab between '__ksymtab_pci_scan_bus_parented' (at offset 0x90) and '__ksymtab_pci_root_buses' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_assign_resources from __ksymtab between '__ksymtab_pci_bus_assign_resources' (at offset 0x4d0) and '__ksymtab_pci_bus_size_bridges' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_size_bridges from __ksymtab between '__ksymtab_pci_bus_size_bridges' (at offset 0x4e0) and '__ksymtab_pci_setup_cardbus' Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2007-03-27 07:53:30 +02:00
int pci_scan_slot(struct pci_bus *bus, int devfn)
{
unsigned fn, nr = 0;
struct pci_dev *dev;
unsigned (*next_fn)(struct pci_dev *, unsigned) = no_next_fn;
if (only_one_child(bus) && (devfn > 0))
return 0; /* Already scanned the entire slot */
dev = pci_scan_single_device(bus, devfn);
if (!dev)
return 0;
if (!dev->is_added)
nr++;
if (pci_ari_enabled(bus))
next_fn = next_ari_fn;
else if (dev->multifunction)
next_fn = next_trad_fn;
for (fn = next_fn(dev, 0); fn > 0; fn = next_fn(dev, fn)) {
dev = pci_scan_single_device(bus, devfn + fn);
if (dev) {
if (!dev->is_added)
nr++;
dev->multifunction = 1;
}
}
/* only one slot has pcie device */
if (bus->self && nr)
pcie_aspm_init_link_state(bus->self);
return nr;
}
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
static int pcie_find_smpss(struct pci_dev *dev, void *data)
{
u8 *smpss = data;
if (!pci_is_pcie(dev))
return 0;
/* For PCIE hotplug enabled slots not connected directly to a
* PCI-E root port, there can be problems when hotplugging
* devices. This is due to the possibility of hotplugging a
* device into the fabric with a smaller MPS that the devices
* currently running have configured. Modifying the MPS on the
* running devices could cause a fatal bus error due to an
* incoming frame being larger than the newly configured MPS.
* To work around this, the MPS for the entire fabric must be
* set to the minimum size. Any devices hotplugged into this
* fabric will have the minimum MPS set. If the PCI hotplug
* slot is directly connected to the root port and there are not
* other devices on the fabric (which seems to be the most
* common case), then this is not an issue and MPS discovery
* will occur as normal.
*/
if (dev->is_hotplug_bridge && (!list_is_singular(&dev->bus->devices) ||
(dev->bus->self &&
dev->bus->self->pcie_type != PCI_EXP_TYPE_ROOT_PORT)))
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
*smpss = 0;
if (*smpss > dev->pcie_mpss)
*smpss = dev->pcie_mpss;
return 0;
}
static void pcie_write_mps(struct pci_dev *dev, int mps)
{
int rc;
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
if (pcie_bus_config == PCIE_BUS_PERFORMANCE) {
mps = 128 << dev->pcie_mpss;
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
if (dev->pcie_type != PCI_EXP_TYPE_ROOT_PORT && dev->bus->self)
/* For "Performance", the assumption is made that
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
* downstream communication will never be larger than
* the MRRS. So, the MPS only needs to be configured
* for the upstream communication. This being the case,
* walk from the top down and set the MPS of the child
* to that of the parent bus.
*
* Configure the device MPS with the smaller of the
* device MPSS or the bridge MPS (which is assumed to be
* properly configured at this point to the largest
* allowable MPS based on its parent bus).
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
*/
mps = min(mps, pcie_get_mps(dev->bus->self));
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
}
rc = pcie_set_mps(dev, mps);
if (rc)
dev_err(&dev->dev, "Failed attempting to set the MPS\n");
}
static void pcie_write_mrrs(struct pci_dev *dev)
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
{
int rc, mrrs;
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
/* In the "safe" case, do not configure the MRRS. There appear to be
* issues with setting MRRS to 0 on a number of devices.
*/
if (pcie_bus_config != PCIE_BUS_PERFORMANCE)
return;
/* For Max performance, the MRRS must be set to the largest supported
* value. However, it cannot be configured larger than the MPS the
* device or the bus can support. This should already be properly
* configured by a prior call to pcie_write_mps.
*/
mrrs = pcie_get_mps(dev);
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
/* MRRS is a R/W register. Invalid values can be written, but a
* subsequent read will verify if the value is acceptable or not.
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
* If the MRRS value provided is not acceptable (e.g., too large),
* shrink the value until it is acceptable to the HW.
*/
while (mrrs != pcie_get_readrq(dev) && mrrs >= 128) {
rc = pcie_set_readrq(dev, mrrs);
if (!rc)
break;
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
dev_warn(&dev->dev, "Failed attempting to set the MRRS\n");
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
mrrs /= 2;
}
if (mrrs < 128)
dev_err(&dev->dev, "MRRS was unable to be configured with a "
"safe value. If problems are experienced, try running "
"with pci=pcie_bus_safe.\n");
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
}
static int pcie_bus_configure_set(struct pci_dev *dev, void *data)
{
int mps, orig_mps;
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
if (!pci_is_pcie(dev))
return 0;
mps = 128 << *(u8 *)data;
orig_mps = pcie_get_mps(dev);
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
pcie_write_mps(dev, mps);
pcie_write_mrrs(dev);
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
dev_info(&dev->dev, "PCI-E Max Payload Size set to %4d/%4d (was %4d), "
"Max Read Rq %4d\n", pcie_get_mps(dev), 128 << dev->pcie_mpss,
orig_mps, pcie_get_readrq(dev));
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
return 0;
}
/* pcie_bus_configure_settings requires that pci_walk_bus work in a top-down,
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
* parents then children fashion. If this changes, then this code will not
* work as designed.
*/
void pcie_bus_configure_settings(struct pci_bus *bus, u8 mpss)
{
u8 smpss;
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
if (!pci_is_pcie(bus->self))
return;
if (pcie_bus_config == PCIE_BUS_TUNE_OFF)
return;
/* FIXME - Peer to peer DMA is possible, though the endpoint would need
* to be aware to the MPS of the destination. To work around this,
* simply force the MPS of the entire system to the smallest possible.
*/
if (pcie_bus_config == PCIE_BUS_PEER2PEER)
smpss = 0;
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
if (pcie_bus_config == PCIE_BUS_SAFE) {
smpss = mpss;
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
pcie_find_smpss(bus->self, &smpss);
pci_walk_bus(bus, pcie_find_smpss, &smpss);
}
pcie_bus_configure_set(bus->self, &smpss);
pci_walk_bus(bus, pcie_bus_configure_set, &smpss);
}
EXPORT_SYMBOL_GPL(pcie_bus_configure_settings);
PCI: Set PCI-E Max Payload Size on fabric On a given PCI-E fabric, each device, bridge, and root port can have a different PCI-E maximum payload size. There is a sizable performance boost for having the largest possible maximum payload size on each PCI-E device. However, if improperly configured, fatal bus errors can occur. Thus, it is important to ensure that PCI-E payloads sends by a device are never larger than the MPS setting of all devices on the way to the destination. This can be achieved two ways: - A conservative approach is to use the smallest common denominator of the entire tree below a root complex for every device on that fabric. This means for example that having a 128 bytes MPS USB controller on one leg of a switch will dramatically reduce performances of a video card or 10GE adapter on another leg of that same switch. It also means that any hierarchy supporting hotplug slots (including expresscard or thunderbolt I suppose, dbl check that) will have to be entirely clamped to 128 bytes since we cannot predict what will be plugged into those slots, and we cannot change the MPS on a "live" system. - A more optimal way is possible, if it falls within a couple of constraints: * The top-level host bridge will never generate packets larger than the smallest TLP (or if it can be controlled independently from its MPS at least) * The device will never generate packets larger than MPS (which can be configured via MRRS) * No support of direct PCI-E <-> PCI-E transfers between devices without some additional code to specifically deal with that case Then we can use an approach that basically ignores downstream requests and focuses exclusively on upstream requests. In that case, all we need to care about is that a device MPS is no larger than its parent MPS, which allows us to keep all switches/bridges to the max MPS supported by their parent and eventually the PHB. In this case, your USB controller would no longer "starve" your 10GE Ethernet and your hotplug slots won't affect your global MPS. Additionally, the hotplugged devices themselves can be configured to a larger MPS up to the value configured in the hotplug bridge. To choose between the two available options, two PCI kernel boot args have been added to the PCI calls. "pcie_bus_safe" will provide the former behavior, while "pcie_bus_perf" will perform the latter behavior. By default, the latter behavior is used. NOTE: due to the location of the enablement, each arch will need to add calls to this function. This patch only enables x86. This patch includes a number of changes recommended by Benjamin Herrenschmidt. Tested-by: Jordan_Hargrave@dell.com Signed-off-by: Jon Mason <mason@myri.com> Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
2011-07-20 22:20:54 +02:00
unsigned int __devinit pci_scan_child_bus(struct pci_bus *bus)
{
unsigned int devfn, pass, max = bus->secondary;
struct pci_dev *dev;
dev_dbg(&bus->dev, "scanning bus\n");
/* Go find them, Rover! */
for (devfn = 0; devfn < 0x100; devfn += 8)
pci_scan_slot(bus, devfn);
/* Reserve buses for SR-IOV capability. */
max += pci_iov_bus_range(bus);
/*
* After performing arch-dependent fixup of the bus, look behind
* all PCI-to-PCI bridges on this bus.
*/
if (!bus->is_added) {
dev_dbg(&bus->dev, "fixups for bus\n");
pcibios_fixup_bus(bus);
if (pci_is_root_bus(bus))
bus->is_added = 1;
}
for (pass=0; pass < 2; pass++)
list_for_each_entry(dev, &bus->devices, bus_list) {
if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE ||
dev->hdr_type == PCI_HEADER_TYPE_CARDBUS)
max = pci_scan_bridge(bus, dev, max, pass);
}
/*
* We've scanned the bus and so we know all about what's on
* the other side of any bridges that may be on this bus plus
* any devices.
*
* Return how far we've got finding sub-buses.
*/
dev_dbg(&bus->dev, "bus scan returning with max=%02x\n", max);
return max;
}
pci: do not mark exported functions as __devinit Functions marked __devinit will be removed after kernel init. But being exported they are potentially called by a module much later. So the safer choice seems to be to keep the function even in the non CONFIG_HOTPLUG case. This silence the follwoing section mismatch warnings: WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_add_device from __ksymtab_gpl between '__ksymtab_pci_bus_add_device' (at offset 0x20) and '__ksymtab_pci_walk_bus' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_create_bus from __ksymtab_gpl between '__ksymtab_pci_create_bus' (at offset 0x40) and '__ksymtab_pci_stop_bus_device' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_max_busnr from __ksymtab_gpl between '__ksymtab_pci_bus_max_busnr' (at offset 0xc0) and '__ksymtab_pci_assign_resource_fixed' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_claim_resource from __ksymtab_gpl between '__ksymtab_pci_claim_resource' (at offset 0xe0) and '__ksymtab_pcie_port_bus_type' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_add_devices from __ksymtab between '__ksymtab_pci_bus_add_devices' (at offset 0x70) and '__ksymtab_pci_bus_alloc_resource' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_scan_bus_parented from __ksymtab between '__ksymtab_pci_scan_bus_parented' (at offset 0x90) and '__ksymtab_pci_root_buses' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_assign_resources from __ksymtab between '__ksymtab_pci_bus_assign_resources' (at offset 0x4d0) and '__ksymtab_pci_bus_size_bridges' WARNING: drivers/built-in.o - Section mismatch: reference to .init.text:pci_bus_size_bridges from __ksymtab between '__ksymtab_pci_bus_size_bridges' (at offset 0x4e0) and '__ksymtab_pci_setup_cardbus' Signed-off-by: Sam Ravnborg <sam@ravnborg.org> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2007-03-27 07:53:30 +02:00
struct pci_bus * pci_create_bus(struct device *parent,
int bus, struct pci_ops *ops, void *sysdata)
{
int error;
struct pci_bus *b, *b2;
struct device *dev;
b = pci_alloc_bus();
if (!b)
return NULL;
dev = kzalloc(sizeof(*dev), GFP_KERNEL);
if (!dev){
kfree(b);
return NULL;
}
b->sysdata = sysdata;
b->ops = ops;
b2 = pci_find_bus(pci_domain_nr(b), bus);
if (b2) {
/* If we already got to this bus through a different bridge, ignore it */
dev_dbg(&b2->dev, "bus already known\n");
goto err_out;
}
down_write(&pci_bus_sem);
list_add_tail(&b->node, &pci_root_buses);
up_write(&pci_bus_sem);
dev->parent = parent;
dev->release = pci_release_bus_bridge_dev;
dev_set_name(dev, "pci%04x:%02x", pci_domain_nr(b), bus);
error = device_register(dev);
if (error)
goto dev_reg_err;
b->bridge = get_device(dev);
device_enable_async_suspend(b->bridge);
pci_set_bus_of_node(b);
if (!parent)
set_dev_node(b->bridge, pcibus_to_node(b));
b->dev.class = &pcibus_class;
b->dev.parent = b->bridge;
dev_set_name(&b->dev, "%04x:%02x", pci_domain_nr(b), bus);
error = device_register(&b->dev);
if (error)
goto class_dev_reg_err;
/* Create legacy_io and legacy_mem files for this bus */
pci_create_legacy_files(b);
b->number = b->secondary = bus;
b->resource[0] = &ioport_resource;
b->resource[1] = &iomem_resource;
return b;
class_dev_reg_err:
device_unregister(dev);
dev_reg_err:
down_write(&pci_bus_sem);
list_del(&b->node);
up_write(&pci_bus_sem);
err_out:
kfree(dev);
kfree(b);
return NULL;
}
struct pci_bus * __devinit pci_scan_bus_parented(struct device *parent,
int bus, struct pci_ops *ops, void *sysdata)
{
struct pci_bus *b;
b = pci_create_bus(parent, bus, ops, sysdata);
if (b)
b->subordinate = pci_scan_child_bus(b);
return b;
}
EXPORT_SYMBOL(pci_scan_bus_parented);
#ifdef CONFIG_HOTPLUG
/**
* pci_rescan_bus - scan a PCI bus for devices.
* @bus: PCI bus to scan
*
* Scan a PCI bus and child buses for new devices, adds them,
* and enables them.
*
* Returns the max number of subordinate bus discovered.
*/
unsigned int __ref pci_rescan_bus(struct pci_bus *bus)
{
unsigned int max;
struct pci_dev *dev;
max = pci_scan_child_bus(bus);
down_read(&pci_bus_sem);
list_for_each_entry(dev, &bus->devices, bus_list)
if (dev->hdr_type == PCI_HEADER_TYPE_BRIDGE ||
dev->hdr_type == PCI_HEADER_TYPE_CARDBUS)
if (dev->subordinate)
pci_bus_size_bridges(dev->subordinate);
up_read(&pci_bus_sem);
pci_bus_assign_resources(bus);
pci_enable_bridges(bus);
pci_bus_add_devices(bus);
return max;
}
EXPORT_SYMBOL_GPL(pci_rescan_bus);
EXPORT_SYMBOL(pci_add_new_bus);
EXPORT_SYMBOL(pci_scan_slot);
EXPORT_SYMBOL(pci_scan_bridge);
EXPORT_SYMBOL_GPL(pci_scan_child_bus);
#endif
PCI: optionally sort device lists breadth-first Problem: New Dell PowerEdge servers have 2 embedded ethernet ports, which are labeled NIC1 and NIC2 on the chassis, in the BIOS setup screens, and in the printed documentation. Assuming no other add-in ethernet ports in the system, Linux 2.4 kernels name these eth0 and eth1 respectively. Many people have come to expect this naming. Linux 2.6 kernels name these eth1 and eth0 respectively (backwards from expectations). I also have reports that various Sun and HP servers have similar behavior. Root cause: Linux 2.4 kernels walk the pci_devices list, which happens to be sorted in breadth-first order (or pcbios_find_device order on i386, which most often is breadth-first also). 2.6 kernels have both the pci_devices list and the pci_bus_type.klist_devices list, the latter is what is walked at driver load time to match the pci_id tables; this klist happens to be in depth-first order. On systems where, for physical routing reasons, NIC1 appears on a lower bus number than NIC2, but NIC2's bridge is discovered first in the depth-first ordering, NIC2 will be discovered before NIC1. If the list were sorted breadth-first, NIC1 would be discovered before NIC2. A PowerEdge 1955 system has the following topology which easily exhibits the difference between depth-first and breadth-first device lists. -[0000:00]-+-00.0 Intel Corporation 5000P Chipset Memory Controller Hub +-02.0-[0000:03-08]--+-00.0-[0000:04-07]--+-00.0-[0000:05-06]----00.0-[0000:06]----00.0 Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (labeled NIC2, 2.4 kernel name eth1, 2.6 kernel name eth0) +-1c.0-[0000:01-02]----00.0-[0000:02]----00.0 Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (labeled NIC1, 2.4 kernel name eth0, 2.6 kernel name eth1) Other factors, such as device driver load order and the presence of PCI slots at various points in the bus hierarchy further complicate this problem; I'm not trying to solve those here, just restore the device order, and thus basic behavior, that 2.4 kernels had. Solution: The solution can come in multiple steps. Suggested fix #1: kernel Patch below optionally sorts the two device lists into breadth-first ordering to maintain compatibility with 2.4 kernels. It adds two new command line options: pci=bfsort pci=nobfsort to force the sort order, or not, as you wish. It also adds DMI checks for the specific Dell systems which exhibit "backwards" ordering, to make them "right". Suggested fix #2: udev rules from userland Many people also have the expectation that embedded NICs are always discovered before add-in NICs (which this patch does not try to do). Using the PCI IRQ Routing Table provided by system BIOS, it's easy to determine which PCI devices are embedded, or if add-in, which PCI slot they're in. I'm working on a tool that would allow udev to name ethernet devices in ascending embedded, slot 1 .. slot N order, subsort by PCI bus/dev/fn breadth-first. It'll be possible to use it independent of udev as well for those distributions that don't use udev in their installers. Suggested fix #3: system board routing rules One can constrain the system board layout to put NIC1 ahead of NIC2 regardless of breadth-first or depth-first discovery order. This adds a significant level of complexity to board routing, and may not be possible in all instances (witness the above systems from several major manufacturers). I don't want to encourage this particular train of thought too far, at the expense of not doing #1 or #2 above. Feedback appreciated. Patch tested on a Dell PowerEdge 1955 blade with 2.6.18. You'll also note I took some liberty and temporarily break the klist abstraction to simplify and speed up the sort algorithm. I think that's both safe and appropriate in this instance. Signed-off-by: Matt Domsch <Matt_Domsch@dell.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2006-09-29 22:23:23 +02:00
static int __init pci_sort_bf_cmp(const struct device *d_a, const struct device *d_b)
PCI: optionally sort device lists breadth-first Problem: New Dell PowerEdge servers have 2 embedded ethernet ports, which are labeled NIC1 and NIC2 on the chassis, in the BIOS setup screens, and in the printed documentation. Assuming no other add-in ethernet ports in the system, Linux 2.4 kernels name these eth0 and eth1 respectively. Many people have come to expect this naming. Linux 2.6 kernels name these eth1 and eth0 respectively (backwards from expectations). I also have reports that various Sun and HP servers have similar behavior. Root cause: Linux 2.4 kernels walk the pci_devices list, which happens to be sorted in breadth-first order (or pcbios_find_device order on i386, which most often is breadth-first also). 2.6 kernels have both the pci_devices list and the pci_bus_type.klist_devices list, the latter is what is walked at driver load time to match the pci_id tables; this klist happens to be in depth-first order. On systems where, for physical routing reasons, NIC1 appears on a lower bus number than NIC2, but NIC2's bridge is discovered first in the depth-first ordering, NIC2 will be discovered before NIC1. If the list were sorted breadth-first, NIC1 would be discovered before NIC2. A PowerEdge 1955 system has the following topology which easily exhibits the difference between depth-first and breadth-first device lists. -[0000:00]-+-00.0 Intel Corporation 5000P Chipset Memory Controller Hub +-02.0-[0000:03-08]--+-00.0-[0000:04-07]--+-00.0-[0000:05-06]----00.0-[0000:06]----00.0 Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (labeled NIC2, 2.4 kernel name eth1, 2.6 kernel name eth0) +-1c.0-[0000:01-02]----00.0-[0000:02]----00.0 Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (labeled NIC1, 2.4 kernel name eth0, 2.6 kernel name eth1) Other factors, such as device driver load order and the presence of PCI slots at various points in the bus hierarchy further complicate this problem; I'm not trying to solve those here, just restore the device order, and thus basic behavior, that 2.4 kernels had. Solution: The solution can come in multiple steps. Suggested fix #1: kernel Patch below optionally sorts the two device lists into breadth-first ordering to maintain compatibility with 2.4 kernels. It adds two new command line options: pci=bfsort pci=nobfsort to force the sort order, or not, as you wish. It also adds DMI checks for the specific Dell systems which exhibit "backwards" ordering, to make them "right". Suggested fix #2: udev rules from userland Many people also have the expectation that embedded NICs are always discovered before add-in NICs (which this patch does not try to do). Using the PCI IRQ Routing Table provided by system BIOS, it's easy to determine which PCI devices are embedded, or if add-in, which PCI slot they're in. I'm working on a tool that would allow udev to name ethernet devices in ascending embedded, slot 1 .. slot N order, subsort by PCI bus/dev/fn breadth-first. It'll be possible to use it independent of udev as well for those distributions that don't use udev in their installers. Suggested fix #3: system board routing rules One can constrain the system board layout to put NIC1 ahead of NIC2 regardless of breadth-first or depth-first discovery order. This adds a significant level of complexity to board routing, and may not be possible in all instances (witness the above systems from several major manufacturers). I don't want to encourage this particular train of thought too far, at the expense of not doing #1 or #2 above. Feedback appreciated. Patch tested on a Dell PowerEdge 1955 blade with 2.6.18. You'll also note I took some liberty and temporarily break the klist abstraction to simplify and speed up the sort algorithm. I think that's both safe and appropriate in this instance. Signed-off-by: Matt Domsch <Matt_Domsch@dell.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2006-09-29 22:23:23 +02:00
{
const struct pci_dev *a = to_pci_dev(d_a);
const struct pci_dev *b = to_pci_dev(d_b);
PCI: optionally sort device lists breadth-first Problem: New Dell PowerEdge servers have 2 embedded ethernet ports, which are labeled NIC1 and NIC2 on the chassis, in the BIOS setup screens, and in the printed documentation. Assuming no other add-in ethernet ports in the system, Linux 2.4 kernels name these eth0 and eth1 respectively. Many people have come to expect this naming. Linux 2.6 kernels name these eth1 and eth0 respectively (backwards from expectations). I also have reports that various Sun and HP servers have similar behavior. Root cause: Linux 2.4 kernels walk the pci_devices list, which happens to be sorted in breadth-first order (or pcbios_find_device order on i386, which most often is breadth-first also). 2.6 kernels have both the pci_devices list and the pci_bus_type.klist_devices list, the latter is what is walked at driver load time to match the pci_id tables; this klist happens to be in depth-first order. On systems where, for physical routing reasons, NIC1 appears on a lower bus number than NIC2, but NIC2's bridge is discovered first in the depth-first ordering, NIC2 will be discovered before NIC1. If the list were sorted breadth-first, NIC1 would be discovered before NIC2. A PowerEdge 1955 system has the following topology which easily exhibits the difference between depth-first and breadth-first device lists. -[0000:00]-+-00.0 Intel Corporation 5000P Chipset Memory Controller Hub +-02.0-[0000:03-08]--+-00.0-[0000:04-07]--+-00.0-[0000:05-06]----00.0-[0000:06]----00.0 Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (labeled NIC2, 2.4 kernel name eth1, 2.6 kernel name eth0) +-1c.0-[0000:01-02]----00.0-[0000:02]----00.0 Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (labeled NIC1, 2.4 kernel name eth0, 2.6 kernel name eth1) Other factors, such as device driver load order and the presence of PCI slots at various points in the bus hierarchy further complicate this problem; I'm not trying to solve those here, just restore the device order, and thus basic behavior, that 2.4 kernels had. Solution: The solution can come in multiple steps. Suggested fix #1: kernel Patch below optionally sorts the two device lists into breadth-first ordering to maintain compatibility with 2.4 kernels. It adds two new command line options: pci=bfsort pci=nobfsort to force the sort order, or not, as you wish. It also adds DMI checks for the specific Dell systems which exhibit "backwards" ordering, to make them "right". Suggested fix #2: udev rules from userland Many people also have the expectation that embedded NICs are always discovered before add-in NICs (which this patch does not try to do). Using the PCI IRQ Routing Table provided by system BIOS, it's easy to determine which PCI devices are embedded, or if add-in, which PCI slot they're in. I'm working on a tool that would allow udev to name ethernet devices in ascending embedded, slot 1 .. slot N order, subsort by PCI bus/dev/fn breadth-first. It'll be possible to use it independent of udev as well for those distributions that don't use udev in their installers. Suggested fix #3: system board routing rules One can constrain the system board layout to put NIC1 ahead of NIC2 regardless of breadth-first or depth-first discovery order. This adds a significant level of complexity to board routing, and may not be possible in all instances (witness the above systems from several major manufacturers). I don't want to encourage this particular train of thought too far, at the expense of not doing #1 or #2 above. Feedback appreciated. Patch tested on a Dell PowerEdge 1955 blade with 2.6.18. You'll also note I took some liberty and temporarily break the klist abstraction to simplify and speed up the sort algorithm. I think that's both safe and appropriate in this instance. Signed-off-by: Matt Domsch <Matt_Domsch@dell.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2006-09-29 22:23:23 +02:00
if (pci_domain_nr(a->bus) < pci_domain_nr(b->bus)) return -1;
else if (pci_domain_nr(a->bus) > pci_domain_nr(b->bus)) return 1;
if (a->bus->number < b->bus->number) return -1;
else if (a->bus->number > b->bus->number) return 1;
if (a->devfn < b->devfn) return -1;
else if (a->devfn > b->devfn) return 1;
return 0;
}
void __init pci_sort_breadthfirst(void)
PCI: optionally sort device lists breadth-first Problem: New Dell PowerEdge servers have 2 embedded ethernet ports, which are labeled NIC1 and NIC2 on the chassis, in the BIOS setup screens, and in the printed documentation. Assuming no other add-in ethernet ports in the system, Linux 2.4 kernels name these eth0 and eth1 respectively. Many people have come to expect this naming. Linux 2.6 kernels name these eth1 and eth0 respectively (backwards from expectations). I also have reports that various Sun and HP servers have similar behavior. Root cause: Linux 2.4 kernels walk the pci_devices list, which happens to be sorted in breadth-first order (or pcbios_find_device order on i386, which most often is breadth-first also). 2.6 kernels have both the pci_devices list and the pci_bus_type.klist_devices list, the latter is what is walked at driver load time to match the pci_id tables; this klist happens to be in depth-first order. On systems where, for physical routing reasons, NIC1 appears on a lower bus number than NIC2, but NIC2's bridge is discovered first in the depth-first ordering, NIC2 will be discovered before NIC1. If the list were sorted breadth-first, NIC1 would be discovered before NIC2. A PowerEdge 1955 system has the following topology which easily exhibits the difference between depth-first and breadth-first device lists. -[0000:00]-+-00.0 Intel Corporation 5000P Chipset Memory Controller Hub +-02.0-[0000:03-08]--+-00.0-[0000:04-07]--+-00.0-[0000:05-06]----00.0-[0000:06]----00.0 Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (labeled NIC2, 2.4 kernel name eth1, 2.6 kernel name eth0) +-1c.0-[0000:01-02]----00.0-[0000:02]----00.0 Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (labeled NIC1, 2.4 kernel name eth0, 2.6 kernel name eth1) Other factors, such as device driver load order and the presence of PCI slots at various points in the bus hierarchy further complicate this problem; I'm not trying to solve those here, just restore the device order, and thus basic behavior, that 2.4 kernels had. Solution: The solution can come in multiple steps. Suggested fix #1: kernel Patch below optionally sorts the two device lists into breadth-first ordering to maintain compatibility with 2.4 kernels. It adds two new command line options: pci=bfsort pci=nobfsort to force the sort order, or not, as you wish. It also adds DMI checks for the specific Dell systems which exhibit "backwards" ordering, to make them "right". Suggested fix #2: udev rules from userland Many people also have the expectation that embedded NICs are always discovered before add-in NICs (which this patch does not try to do). Using the PCI IRQ Routing Table provided by system BIOS, it's easy to determine which PCI devices are embedded, or if add-in, which PCI slot they're in. I'm working on a tool that would allow udev to name ethernet devices in ascending embedded, slot 1 .. slot N order, subsort by PCI bus/dev/fn breadth-first. It'll be possible to use it independent of udev as well for those distributions that don't use udev in their installers. Suggested fix #3: system board routing rules One can constrain the system board layout to put NIC1 ahead of NIC2 regardless of breadth-first or depth-first discovery order. This adds a significant level of complexity to board routing, and may not be possible in all instances (witness the above systems from several major manufacturers). I don't want to encourage this particular train of thought too far, at the expense of not doing #1 or #2 above. Feedback appreciated. Patch tested on a Dell PowerEdge 1955 blade with 2.6.18. You'll also note I took some liberty and temporarily break the klist abstraction to simplify and speed up the sort algorithm. I think that's both safe and appropriate in this instance. Signed-off-by: Matt Domsch <Matt_Domsch@dell.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2006-09-29 22:23:23 +02:00
{
bus_sort_breadthfirst(&pci_bus_type, &pci_sort_bf_cmp);
PCI: optionally sort device lists breadth-first Problem: New Dell PowerEdge servers have 2 embedded ethernet ports, which are labeled NIC1 and NIC2 on the chassis, in the BIOS setup screens, and in the printed documentation. Assuming no other add-in ethernet ports in the system, Linux 2.4 kernels name these eth0 and eth1 respectively. Many people have come to expect this naming. Linux 2.6 kernels name these eth1 and eth0 respectively (backwards from expectations). I also have reports that various Sun and HP servers have similar behavior. Root cause: Linux 2.4 kernels walk the pci_devices list, which happens to be sorted in breadth-first order (or pcbios_find_device order on i386, which most often is breadth-first also). 2.6 kernels have both the pci_devices list and the pci_bus_type.klist_devices list, the latter is what is walked at driver load time to match the pci_id tables; this klist happens to be in depth-first order. On systems where, for physical routing reasons, NIC1 appears on a lower bus number than NIC2, but NIC2's bridge is discovered first in the depth-first ordering, NIC2 will be discovered before NIC1. If the list were sorted breadth-first, NIC1 would be discovered before NIC2. A PowerEdge 1955 system has the following topology which easily exhibits the difference between depth-first and breadth-first device lists. -[0000:00]-+-00.0 Intel Corporation 5000P Chipset Memory Controller Hub +-02.0-[0000:03-08]--+-00.0-[0000:04-07]--+-00.0-[0000:05-06]----00.0-[0000:06]----00.0 Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (labeled NIC2, 2.4 kernel name eth1, 2.6 kernel name eth0) +-1c.0-[0000:01-02]----00.0-[0000:02]----00.0 Broadcom Corporation NetXtreme II BCM5708S Gigabit Ethernet (labeled NIC1, 2.4 kernel name eth0, 2.6 kernel name eth1) Other factors, such as device driver load order and the presence of PCI slots at various points in the bus hierarchy further complicate this problem; I'm not trying to solve those here, just restore the device order, and thus basic behavior, that 2.4 kernels had. Solution: The solution can come in multiple steps. Suggested fix #1: kernel Patch below optionally sorts the two device lists into breadth-first ordering to maintain compatibility with 2.4 kernels. It adds two new command line options: pci=bfsort pci=nobfsort to force the sort order, or not, as you wish. It also adds DMI checks for the specific Dell systems which exhibit "backwards" ordering, to make them "right". Suggested fix #2: udev rules from userland Many people also have the expectation that embedded NICs are always discovered before add-in NICs (which this patch does not try to do). Using the PCI IRQ Routing Table provided by system BIOS, it's easy to determine which PCI devices are embedded, or if add-in, which PCI slot they're in. I'm working on a tool that would allow udev to name ethernet devices in ascending embedded, slot 1 .. slot N order, subsort by PCI bus/dev/fn breadth-first. It'll be possible to use it independent of udev as well for those distributions that don't use udev in their installers. Suggested fix #3: system board routing rules One can constrain the system board layout to put NIC1 ahead of NIC2 regardless of breadth-first or depth-first discovery order. This adds a significant level of complexity to board routing, and may not be possible in all instances (witness the above systems from several major manufacturers). I don't want to encourage this particular train of thought too far, at the expense of not doing #1 or #2 above. Feedback appreciated. Patch tested on a Dell PowerEdge 1955 blade with 2.6.18. You'll also note I took some liberty and temporarily break the klist abstraction to simplify and speed up the sort algorithm. I think that's both safe and appropriate in this instance. Signed-off-by: Matt Domsch <Matt_Domsch@dell.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
2006-09-29 22:23:23 +02:00
}