linux_dsm_epyc7002/arch/powerpc/platforms/pseries/eeh_cache.c
Gavin Shan d50a7d4c6f powerpc/eeh: Replace pci_dn with eeh_dev for EEH address cache
With original EEH implementation, struct pci_dn is used while building
PCI I/O address cache, which helps on searching the corresponding
PCI device according to the given physical I/O address. Besides, pci_dn
is associated with the corresponding PCI device while building its
I/O cache.

The patch replaces struct pci_dn with struct eeh_dev so that EEH address
cache won't depend on struct pci_dn. That will help EEH to become an
independent module in future. Besides, the binding of eeh_dev and PCI
device is done while building PCI device I/O cache.

Signed-off-by: Gavin Shan <shangw@linux.vnet.ibm.com>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
2012-03-09 11:39:42 +11:00

319 lines
8.8 KiB
C

/*
* PCI address cache; allows the lookup of PCI devices based on I/O address
*
* Copyright IBM Corporation 2004
* Copyright Linas Vepstas <linas@austin.ibm.com> 2004
*
* This program is free software; you can redistribute it and/or modify
* it under the terms of the GNU General Public License as published by
* the Free Software Foundation; either version 2 of the License, or
* (at your option) any later version.
*
* This program is distributed in the hope that it will be useful,
* but WITHOUT ANY WARRANTY; without even the implied warranty of
* MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
* GNU General Public License for more details.
*
* You should have received a copy of the GNU General Public License
* along with this program; if not, write to the Free Software
* Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
*/
#include <linux/list.h>
#include <linux/pci.h>
#include <linux/rbtree.h>
#include <linux/slab.h>
#include <linux/spinlock.h>
#include <linux/atomic.h>
#include <asm/pci-bridge.h>
#include <asm/ppc-pci.h>
/**
* The pci address cache subsystem. This subsystem places
* PCI device address resources into a red-black tree, sorted
* according to the address range, so that given only an i/o
* address, the corresponding PCI device can be **quickly**
* found. It is safe to perform an address lookup in an interrupt
* context; this ability is an important feature.
*
* Currently, the only customer of this code is the EEH subsystem;
* thus, this code has been somewhat tailored to suit EEH better.
* In particular, the cache does *not* hold the addresses of devices
* for which EEH is not enabled.
*
* (Implementation Note: The RB tree seems to be better/faster
* than any hash algo I could think of for this problem, even
* with the penalty of slow pointer chases for d-cache misses).
*/
struct pci_io_addr_range {
struct rb_node rb_node;
unsigned long addr_lo;
unsigned long addr_hi;
struct pci_dev *pcidev;
unsigned int flags;
};
static struct pci_io_addr_cache {
struct rb_root rb_root;
spinlock_t piar_lock;
} pci_io_addr_cache_root;
static inline struct pci_dev *__pci_addr_cache_get_device(unsigned long addr)
{
struct rb_node *n = pci_io_addr_cache_root.rb_root.rb_node;
while (n) {
struct pci_io_addr_range *piar;
piar = rb_entry(n, struct pci_io_addr_range, rb_node);
if (addr < piar->addr_lo) {
n = n->rb_left;
} else {
if (addr > piar->addr_hi) {
n = n->rb_right;
} else {
pci_dev_get(piar->pcidev);
return piar->pcidev;
}
}
}
return NULL;
}
/**
* pci_addr_cache_get_device - Get device, given only address
* @addr: mmio (PIO) phys address or i/o port number
*
* Given an mmio phys address, or a port number, find a pci device
* that implements this address. Be sure to pci_dev_put the device
* when finished. I/O port numbers are assumed to be offset
* from zero (that is, they do *not* have pci_io_addr added in).
* It is safe to call this function within an interrupt.
*/
struct pci_dev *pci_addr_cache_get_device(unsigned long addr)
{
struct pci_dev *dev;
unsigned long flags;
spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
dev = __pci_addr_cache_get_device(addr);
spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
return dev;
}
#ifdef DEBUG
/*
* Handy-dandy debug print routine, does nothing more
* than print out the contents of our addr cache.
*/
static void pci_addr_cache_print(struct pci_io_addr_cache *cache)
{
struct rb_node *n;
int cnt = 0;
n = rb_first(&cache->rb_root);
while (n) {
struct pci_io_addr_range *piar;
piar = rb_entry(n, struct pci_io_addr_range, rb_node);
printk(KERN_DEBUG "PCI: %s addr range %d [%lx-%lx]: %s\n",
(piar->flags & IORESOURCE_IO) ? "i/o" : "mem", cnt,
piar->addr_lo, piar->addr_hi, pci_name(piar->pcidev));
cnt++;
n = rb_next(n);
}
}
#endif
/* Insert address range into the rb tree. */
static struct pci_io_addr_range *
pci_addr_cache_insert(struct pci_dev *dev, unsigned long alo,
unsigned long ahi, unsigned int flags)
{
struct rb_node **p = &pci_io_addr_cache_root.rb_root.rb_node;
struct rb_node *parent = NULL;
struct pci_io_addr_range *piar;
/* Walk tree, find a place to insert into tree */
while (*p) {
parent = *p;
piar = rb_entry(parent, struct pci_io_addr_range, rb_node);
if (ahi < piar->addr_lo) {
p = &parent->rb_left;
} else if (alo > piar->addr_hi) {
p = &parent->rb_right;
} else {
if (dev != piar->pcidev ||
alo != piar->addr_lo || ahi != piar->addr_hi) {
printk(KERN_WARNING "PIAR: overlapping address range\n");
}
return piar;
}
}
piar = kmalloc(sizeof(struct pci_io_addr_range), GFP_ATOMIC);
if (!piar)
return NULL;
pci_dev_get(dev);
piar->addr_lo = alo;
piar->addr_hi = ahi;
piar->pcidev = dev;
piar->flags = flags;
#ifdef DEBUG
printk(KERN_DEBUG "PIAR: insert range=[%lx:%lx] dev=%s\n",
alo, ahi, pci_name(dev));
#endif
rb_link_node(&piar->rb_node, parent, p);
rb_insert_color(&piar->rb_node, &pci_io_addr_cache_root.rb_root);
return piar;
}
static void __pci_addr_cache_insert_device(struct pci_dev *dev)
{
struct device_node *dn;
struct eeh_dev *edev;
int i;
dn = pci_device_to_OF_node(dev);
if (!dn) {
printk(KERN_WARNING "PCI: no pci dn found for dev=%s\n", pci_name(dev));
return;
}
edev = of_node_to_eeh_dev(dn);
if (!edev) {
pr_warning("PCI: no EEH dev found for dn=%s\n",
dn->full_name);
return;
}
/* Skip any devices for which EEH is not enabled. */
if (!(edev->mode & EEH_MODE_SUPPORTED) ||
edev->mode & EEH_MODE_NOCHECK) {
#ifdef DEBUG
pr_info("PCI: skip building address cache for=%s - %s\n",
pci_name(dev), dn->full_name);
#endif
return;
}
/* Walk resources on this device, poke them into the tree */
for (i = 0; i < DEVICE_COUNT_RESOURCE; i++) {
unsigned long start = pci_resource_start(dev,i);
unsigned long end = pci_resource_end(dev,i);
unsigned int flags = pci_resource_flags(dev,i);
/* We are interested only bus addresses, not dma or other stuff */
if (0 == (flags & (IORESOURCE_IO | IORESOURCE_MEM)))
continue;
if (start == 0 || ~start == 0 || end == 0 || ~end == 0)
continue;
pci_addr_cache_insert(dev, start, end, flags);
}
}
/**
* pci_addr_cache_insert_device - Add a device to the address cache
* @dev: PCI device whose I/O addresses we are interested in.
*
* In order to support the fast lookup of devices based on addresses,
* we maintain a cache of devices that can be quickly searched.
* This routine adds a device to that cache.
*/
void pci_addr_cache_insert_device(struct pci_dev *dev)
{
unsigned long flags;
/* Ignore PCI bridges */
if ((dev->class >> 16) == PCI_BASE_CLASS_BRIDGE)
return;
spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
__pci_addr_cache_insert_device(dev);
spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
}
static inline void __pci_addr_cache_remove_device(struct pci_dev *dev)
{
struct rb_node *n;
restart:
n = rb_first(&pci_io_addr_cache_root.rb_root);
while (n) {
struct pci_io_addr_range *piar;
piar = rb_entry(n, struct pci_io_addr_range, rb_node);
if (piar->pcidev == dev) {
rb_erase(n, &pci_io_addr_cache_root.rb_root);
pci_dev_put(piar->pcidev);
kfree(piar);
goto restart;
}
n = rb_next(n);
}
}
/**
* pci_addr_cache_remove_device - remove pci device from addr cache
* @dev: device to remove
*
* Remove a device from the addr-cache tree.
* This is potentially expensive, since it will walk
* the tree multiple times (once per resource).
* But so what; device removal doesn't need to be that fast.
*/
void pci_addr_cache_remove_device(struct pci_dev *dev)
{
unsigned long flags;
spin_lock_irqsave(&pci_io_addr_cache_root.piar_lock, flags);
__pci_addr_cache_remove_device(dev);
spin_unlock_irqrestore(&pci_io_addr_cache_root.piar_lock, flags);
}
/**
* pci_addr_cache_build - Build a cache of I/O addresses
*
* Build a cache of pci i/o addresses. This cache will be used to
* find the pci device that corresponds to a given address.
* This routine scans all pci busses to build the cache.
* Must be run late in boot process, after the pci controllers
* have been scanned for devices (after all device resources are known).
*/
void __init pci_addr_cache_build(void)
{
struct device_node *dn;
struct eeh_dev *edev;
struct pci_dev *dev = NULL;
spin_lock_init(&pci_io_addr_cache_root.piar_lock);
for_each_pci_dev(dev) {
pci_addr_cache_insert_device(dev);
dn = pci_device_to_OF_node(dev);
if (!dn)
continue;
edev = of_node_to_eeh_dev(dn);
if (!edev)
continue;
pci_dev_get(dev); /* matching put is in eeh_remove_device() */
dev->dev.archdata.edev = edev;
edev->pdev = dev;
eeh_sysfs_add_device(dev);
}
#ifdef DEBUG
/* Verify tree built up above, echo back the list of addrs. */
pci_addr_cache_print(&pci_io_addr_cache_root);
#endif
}