linux/drivers/pci/msi.c

754 lines
18 KiB
C
Raw Normal View History

/*
* File: msi.c
* Purpose: PCI Message Signaled Interrupt (MSI)
*
* Copyright (C) 2003-2004 Intel
* Copyright (C) Tom Long Nguyen (tom.l.nguyen@intel.com)
*/
#include <linux/err.h>
#include <linux/mm.h>
#include <linux/irq.h>
#include <linux/interrupt.h>
#include <linux/init.h>
#include <linux/ioport.h>
#include <linux/pci.h>
#include <linux/proc_fs.h>
#include <linux/msi.h>
#include <linux/smp.h>
#include <asm/errno.h>
#include <asm/io.h>
#include "pci.h"
#include "msi.h"
static int pci_msi_enable = 1;
/* Arch hooks */
int __attribute__ ((weak))
arch_msi_check_device(struct pci_dev *dev, int nvec, int type)
{
return 0;
}
int __attribute__ ((weak))
arch_setup_msi_irq(struct pci_dev *dev, struct msi_desc *entry)
{
return 0;
}
int __attribute__ ((weak))
arch_setup_msi_irqs(struct pci_dev *dev, int nvec, int type)
{
struct msi_desc *entry;
int ret;
list_for_each_entry(entry, &dev->msi_list, list) {
ret = arch_setup_msi_irq(dev, entry);
if (ret)
return ret;
}
return 0;
}
void __attribute__ ((weak)) arch_teardown_msi_irq(unsigned int irq)
{
return;
}
void __attribute__ ((weak))
arch_teardown_msi_irqs(struct pci_dev *dev)
{
struct msi_desc *entry;
list_for_each_entry(entry, &dev->msi_list, list) {
if (entry->irq != 0)
arch_teardown_msi_irq(entry->irq);
}
}
static void msi_set_enable(struct pci_dev *dev, int enable)
{
int pos;
u16 control;
pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
if (pos) {
pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
control &= ~PCI_MSI_FLAGS_ENABLE;
if (enable)
control |= PCI_MSI_FLAGS_ENABLE;
pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
}
}
static void msix_set_enable(struct pci_dev *dev, int enable)
{
int pos;
u16 control;
pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
if (pos) {
pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
control &= ~PCI_MSIX_FLAGS_ENABLE;
if (enable)
control |= PCI_MSIX_FLAGS_ENABLE;
pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
}
}
static void msix_flush_writes(unsigned int irq)
{
struct msi_desc *entry;
entry = get_irq_msi(irq);
BUG_ON(!entry || !entry->dev);
switch (entry->msi_attrib.type) {
case PCI_CAP_ID_MSI:
/* nothing to do */
break;
case PCI_CAP_ID_MSIX:
{
int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
readl(entry->mask_base + offset);
break;
}
default:
BUG();
break;
}
}
static void msi_set_mask_bits(unsigned int irq, u32 mask, u32 flag)
{
struct msi_desc *entry;
entry = get_irq_msi(irq);
BUG_ON(!entry || !entry->dev);
switch (entry->msi_attrib.type) {
case PCI_CAP_ID_MSI:
if (entry->msi_attrib.maskbit) {
int pos;
u32 mask_bits;
pos = (long)entry->mask_base;
pci_read_config_dword(entry->dev, pos, &mask_bits);
mask_bits &= ~(mask);
mask_bits |= flag & mask;
pci_write_config_dword(entry->dev, pos, mask_bits);
} else {
msi_set_enable(entry->dev, !flag);
}
break;
case PCI_CAP_ID_MSIX:
{
int offset = entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE +
PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET;
writel(flag, entry->mask_base + offset);
[PATCH] msi: synchronously mask and unmask msi-x irqs. This is a simplified and actually more comprehensive form of a bug fix from Mitch Williams <mitch.a.williams@intel.com>. When we mask or unmask a msi-x irqs the writes may be posted because we are writing to memory mapped region. This means the mask and unmask don't happen immediately but at some unspecified time in the future. Which is out of sync with how the mask/unmask logic work for ioapic irqs. The practical result is that we get very subtle and hard to track down irq migration bugs. This patch performs a read flush after writes to the MSI-X table for mask and unmask operations. Since the SMP affinity is set while the interrupt is masked, and since it's unmasked immediately after, no additional flushes are required in the various affinity setting routines. The testing by Mitch Williams on his especially problematic system should still be valid as I have only simplified the code, not changed the functionality. We currently have 7 drivers: cciss, mthca, cxgb3, forceth, s2io, pcie/portdrv_core, and qla2xxx in 2.6.21 that are affected by this problem when the hardware they driver is plugged into the right slot. Given the difficulty of reproducing this bug and tracing it down to anything that even remotely resembles a cause, even if people are being affected we aren't likely to see many meaningful bug reports, and the people who see this bug aren't likely to be able to reproduce this bug in a timely fashion. So it is best to get this problem fixed as soon as we can so people don't have problems. Then if people do have a kernel message stating "No irq for vector" we will know it is yet another novel cause that needs a complete new investigation. Cc: Greg KH <greg@kroah.com> Cc: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Acked-by: Mitch Williams <mitch.a.williams@intel.com> Acked-by: "Siddha, Suresh B" <suresh.b.siddha@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-04-03 07:41:49 +00:00
readl(entry->mask_base + offset);
break;
}
default:
BUG();
break;
}
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
entry->msi_attrib.masked = !!flag;
}
void read_msi_msg(unsigned int irq, struct msi_msg *msg)
{
struct msi_desc *entry = get_irq_msi(irq);
switch(entry->msi_attrib.type) {
case PCI_CAP_ID_MSI:
{
struct pci_dev *dev = entry->dev;
int pos = entry->msi_attrib.pos;
u16 data;
pci_read_config_dword(dev, msi_lower_address_reg(pos),
&msg->address_lo);
if (entry->msi_attrib.is_64) {
pci_read_config_dword(dev, msi_upper_address_reg(pos),
&msg->address_hi);
pci_read_config_word(dev, msi_data_reg(pos, 1), &data);
} else {
msg->address_hi = 0;
pci_read_config_word(dev, msi_data_reg(pos, 0), &data);
}
msg->data = data;
break;
}
case PCI_CAP_ID_MSIX:
{
void __iomem *base;
base = entry->mask_base +
entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
msg->address_lo = readl(base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
msg->address_hi = readl(base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
msg->data = readl(base + PCI_MSIX_ENTRY_DATA_OFFSET);
break;
}
default:
BUG();
}
}
void write_msi_msg(unsigned int irq, struct msi_msg *msg)
{
struct msi_desc *entry = get_irq_msi(irq);
switch (entry->msi_attrib.type) {
case PCI_CAP_ID_MSI:
{
struct pci_dev *dev = entry->dev;
int pos = entry->msi_attrib.pos;
pci_write_config_dword(dev, msi_lower_address_reg(pos),
msg->address_lo);
if (entry->msi_attrib.is_64) {
pci_write_config_dword(dev, msi_upper_address_reg(pos),
msg->address_hi);
pci_write_config_word(dev, msi_data_reg(pos, 1),
msg->data);
} else {
pci_write_config_word(dev, msi_data_reg(pos, 0),
msg->data);
}
break;
}
case PCI_CAP_ID_MSIX:
{
void __iomem *base;
base = entry->mask_base +
entry->msi_attrib.entry_nr * PCI_MSIX_ENTRY_SIZE;
writel(msg->address_lo,
base + PCI_MSIX_ENTRY_LOWER_ADDR_OFFSET);
writel(msg->address_hi,
base + PCI_MSIX_ENTRY_UPPER_ADDR_OFFSET);
writel(msg->data, base + PCI_MSIX_ENTRY_DATA_OFFSET);
break;
}
default:
BUG();
}
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
entry->msg = *msg;
}
void mask_msi_irq(unsigned int irq)
{
msi_set_mask_bits(irq, 1, 1);
msix_flush_writes(irq);
}
void unmask_msi_irq(unsigned int irq)
{
msi_set_mask_bits(irq, 1, 0);
msix_flush_writes(irq);
}
static int msi_free_irqs(struct pci_dev* dev);
static struct msi_desc* alloc_msi_entry(void)
{
struct msi_desc *entry;
entry = kzalloc(sizeof(struct msi_desc), GFP_KERNEL);
if (!entry)
return NULL;
INIT_LIST_HEAD(&entry->list);
entry->irq = 0;
entry->dev = NULL;
return entry;
}
static void pci_intx_for_msi(struct pci_dev *dev, int enable)
{
if (!(dev->dev_flags & PCI_DEV_FLAGS_MSI_INTX_DISABLE_BUG))
pci_intx(dev, enable);
}
static void __pci_restore_msi_state(struct pci_dev *dev)
{
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
int pos;
u16 control;
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
struct msi_desc *entry;
if (!dev->msi_enabled)
return;
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
entry = get_irq_msi(dev->irq);
pos = entry->msi_attrib.pos;
pci_intx_for_msi(dev, 0);
msi_set_enable(dev, 0);
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
write_msi_msg(dev->irq, &entry->msg);
if (entry->msi_attrib.maskbit)
msi_set_mask_bits(dev->irq, entry->msi_attrib.maskbits_mask,
entry->msi_attrib.masked);
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
pci_read_config_word(dev, pos + PCI_MSI_FLAGS, &control);
control &= ~(PCI_MSI_FLAGS_QSIZE | PCI_MSI_FLAGS_ENABLE);
if (entry->msi_attrib.maskbit || !entry->msi_attrib.masked)
control |= PCI_MSI_FLAGS_ENABLE;
pci_write_config_word(dev, pos + PCI_MSI_FLAGS, control);
}
static void __pci_restore_msix_state(struct pci_dev *dev)
{
int pos;
struct msi_desc *entry;
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
u16 control;
if (!dev->msix_enabled)
return;
/* route the table */
pci_intx_for_msi(dev, 0);
msix_set_enable(dev, 0);
list_for_each_entry(entry, &dev->msi_list, list) {
write_msi_msg(entry->irq, &entry->msg);
msi_set_mask_bits(entry->irq, 1, entry->msi_attrib.masked);
}
BUG_ON(list_empty(&dev->msi_list));
entry = list_entry(dev->msi_list.next, struct msi_desc, list);
pos = entry->msi_attrib.pos;
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
pci_read_config_word(dev, pos + PCI_MSIX_FLAGS, &control);
control &= ~PCI_MSIX_FLAGS_MASKALL;
control |= PCI_MSIX_FLAGS_ENABLE;
pci_write_config_word(dev, pos + PCI_MSIX_FLAGS, control);
}
void pci_restore_msi_state(struct pci_dev *dev)
{
__pci_restore_msi_state(dev);
__pci_restore_msix_state(dev);
}
EXPORT_SYMBOL_GPL(pci_restore_msi_state);
/**
* msi_capability_init - configure device's MSI capability structure
* @dev: pointer to the pci_dev data structure of MSI device function
*
* Setup the MSI capability structure of device function with a single
* MSI irq, regardless of device function is capable of handling
* multiple messages. A return of zero indicates the successful setup
* of an entry zero with the new MSI irq or non-zero for otherwise.
**/
static int msi_capability_init(struct pci_dev *dev)
{
struct msi_desc *entry;
int pos, ret;
u16 control;
msi_set_enable(dev, 0); /* Ensure msi is disabled as I set it up */
pos = pci_find_capability(dev, PCI_CAP_ID_MSI);
pci_read_config_word(dev, msi_control_reg(pos), &control);
/* MSI Entry Initialization */
entry = alloc_msi_entry();
if (!entry)
return -ENOMEM;
entry->msi_attrib.type = PCI_CAP_ID_MSI;
entry->msi_attrib.is_64 = is_64bit_address(control);
entry->msi_attrib.entry_nr = 0;
entry->msi_attrib.maskbit = is_mask_bit_support(control);
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
entry->msi_attrib.masked = 1;
entry->msi_attrib.default_irq = dev->irq; /* Save IOAPIC IRQ */
entry->msi_attrib.pos = pos;
if (is_mask_bit_support(control)) {
entry->mask_base = (void __iomem *)(long)msi_mask_bits_reg(pos,
is_64bit_address(control));
}
entry->dev = dev;
if (entry->msi_attrib.maskbit) {
unsigned int maskbits, temp;
/* All MSIs are unmasked by default, Mask them all */
pci_read_config_dword(dev,
msi_mask_bits_reg(pos, is_64bit_address(control)),
&maskbits);
temp = (1 << multi_msi_capable(control));
temp = ((temp - 1) & ~temp);
maskbits |= temp;
pci_write_config_dword(dev,
msi_mask_bits_reg(pos, is_64bit_address(control)),
maskbits);
entry->msi_attrib.maskbits_mask = temp;
}
list_add_tail(&entry->list, &dev->msi_list);
/* Configure MSI capability structure */
ret = arch_setup_msi_irqs(dev, 1, PCI_CAP_ID_MSI);
if (ret) {
msi_free_irqs(dev);
return ret;
}
/* Set MSI enabled bits */
pci_intx_for_msi(dev, 0);
msi_set_enable(dev, 1);
dev->msi_enabled = 1;
dev->irq = entry->irq;
return 0;
}
/**
* msix_capability_init - configure device's MSI-X capability
* @dev: pointer to the pci_dev data structure of MSI-X device function
* @entries: pointer to an array of struct msix_entry entries
* @nvec: number of @entries
*
* Setup the MSI-X capability structure of device function with a
* single MSI-X irq. A return of zero indicates the successful setup of
* requested MSI-X entries with allocated irqs or non-zero for otherwise.
**/
static int msix_capability_init(struct pci_dev *dev,
struct msix_entry *entries, int nvec)
{
struct msi_desc *entry;
int pos, i, j, nr_entries, ret;
unsigned long phys_addr;
u32 table_offset;
u16 control;
u8 bir;
void __iomem *base;
msix_set_enable(dev, 0);/* Ensure msix is disabled as I set it up */
pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
/* Request & Map MSI-X table region */
pci_read_config_word(dev, msi_control_reg(pos), &control);
nr_entries = multi_msix_capable(control);
pci_read_config_dword(dev, msix_table_offset_reg(pos), &table_offset);
bir = (u8)(table_offset & PCI_MSIX_FLAGS_BIRMASK);
table_offset &= ~PCI_MSIX_FLAGS_BIRMASK;
phys_addr = pci_resource_start (dev, bir) + table_offset;
base = ioremap_nocache(phys_addr, nr_entries * PCI_MSIX_ENTRY_SIZE);
if (base == NULL)
return -ENOMEM;
/* MSI-X Table Initialization */
for (i = 0; i < nvec; i++) {
entry = alloc_msi_entry();
if (!entry)
break;
j = entries[i].entry;
entry->msi_attrib.type = PCI_CAP_ID_MSIX;
entry->msi_attrib.is_64 = 1;
entry->msi_attrib.entry_nr = j;
entry->msi_attrib.maskbit = 1;
[PATCH] msi: Safer state caching. There are two ways pci_save_state and pci_restore_state are used. As helper functions during suspend/resume, and as helper functions around a hardware reset event. When used as helper functions around a hardware reset event there is no reason to believe the calls will be paired, nor is there a good reason to believe that if we restore the msi state from before the reset that it will match the current msi state. Since arch code may change the msi message without going through the driver, drivers currently do not have enough information to even know when to call pci_save_state to ensure they will have msi state in sync with the other kernel irq reception data structures. It turns out the solution is straight forward, cache the state in the existing msi data structures (not the magic pci saved things) and have the msi code update the cached state each time we write to the hardware. This means we never need to read the hardware to figure out what the hardware state should be. By modifying the caching in this manner we get to remove our save_state routines and only need to provide restore_state routines. The only fields that were at all tricky to regenerate were the msi and msi-x control registers and the way we regenerate them currently is a bit dependent upon assumptions on how we use the allow msi registers to be configured and used making the code a little bit brittle. If we ever change what cases we allow or how we configure the msi bits we can address the fragility then. Signed-off-by: Eric W. Biederman <ebiederm@xmission.com> Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de> Acked-by: Auke Kok <auke-jan.h.kok@intel.com> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-03-08 20:04:57 +00:00
entry->msi_attrib.masked = 1;
entry->msi_attrib.default_irq = dev->irq;
entry->msi_attrib.pos = pos;
entry->dev = dev;
entry->mask_base = base;
list_add_tail(&entry->list, &dev->msi_list);
}
ret = arch_setup_msi_irqs(dev, nvec, PCI_CAP_ID_MSIX);
if (ret) {
int avail = 0;
list_for_each_entry(entry, &dev->msi_list, list) {
if (entry->irq != 0) {
avail++;
}
}
msi_free_irqs(dev);
/* If we had some success report the number of irqs
* we succeeded in setting up.
*/
if (avail == 0)
avail = ret;
return avail;
}
i = 0;
list_for_each_entry(entry, &dev->msi_list, list) {
entries[i].vector = entry->irq;
set_irq_msi(entry->irq, entry);
i++;
}
/* Set MSI-X enabled bits */
pci_intx_for_msi(dev, 0);
msix_set_enable(dev, 1);
dev->msix_enabled = 1;
return 0;
}
/**
* pci_msi_check_device - check whether MSI may be enabled on a device
* @dev: pointer to the pci_dev data structure of MSI device function
* @nvec: how many MSIs have been requested ?
* @type: are we checking for MSI or MSI-X ?
*
* Look at global flags, the device itself, and its parent busses
* to determine if MSI/-X are supported for the device. If MSI/-X is
* supported return 0, else return an error code.
**/
static int pci_msi_check_device(struct pci_dev* dev, int nvec, int type)
{
struct pci_bus *bus;
int ret;
/* MSI must be globally enabled and supported by the device */
if (!pci_msi_enable || !dev || dev->no_msi)
return -EINVAL;
/*
* You can't ask to have 0 or less MSIs configured.
* a) it's stupid ..
* b) the list manipulation code assumes nvec >= 1.
*/
if (nvec < 1)
return -ERANGE;
/* Any bridge which does NOT route MSI transactions from it's
* secondary bus to it's primary bus must set NO_MSI flag on
* the secondary pci_bus.
* We expect only arch-specific PCI host bus controller driver
* or quirks for specific PCI bridges to be setting NO_MSI.
*/
for (bus = dev->bus; bus; bus = bus->parent)
if (bus->bus_flags & PCI_BUS_FLAGS_NO_MSI)
return -EINVAL;
ret = arch_msi_check_device(dev, nvec, type);
if (ret)
return ret;
if (!pci_find_capability(dev, type))
return -EINVAL;
return 0;
}
/**
* pci_enable_msi - configure device's MSI capability structure
* @dev: pointer to the pci_dev data structure of MSI device function
*
* Setup the MSI capability structure of device function with
* a single MSI irq upon its software driver call to request for
* MSI mode enabled on its hardware device function. A return of zero
* indicates the successful setup of an entry zero with the new MSI
* irq or non-zero for otherwise.
**/
int pci_enable_msi(struct pci_dev* dev)
{
int status;
status = pci_msi_check_device(dev, 1, PCI_CAP_ID_MSI);
if (status)
return status;
WARN_ON(!!dev->msi_enabled);
/* Check whether driver already requested for MSI-X irqs */
if (dev->msix_enabled) {
printk(KERN_INFO "PCI: %s: Can't enable MSI. "
"Device already has MSI-X enabled\n",
pci_name(dev));
return -EINVAL;
}
status = msi_capability_init(dev);
return status;
}
EXPORT_SYMBOL(pci_enable_msi);
void pci_msi_shutdown(struct pci_dev* dev)
{
struct msi_desc *entry;
if (!pci_msi_enable || !dev || !dev->msi_enabled)
return;
msi_set_enable(dev, 0);
pci_intx_for_msi(dev, 1);
dev->msi_enabled = 0;
BUG_ON(list_empty(&dev->msi_list));
entry = list_entry(dev->msi_list.next, struct msi_desc, list);
/* Return the the pci reset with msi irqs unmasked */
if (entry->msi_attrib.maskbit) {
u32 mask = entry->msi_attrib.maskbits_mask;
msi_set_mask_bits(dev->irq, mask, ~mask);
}
if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI)
return;
/* Restore dev->irq to its default pin-assertion irq */
dev->irq = entry->msi_attrib.default_irq;
}
void pci_disable_msi(struct pci_dev* dev)
{
struct msi_desc *entry;
if (!pci_msi_enable || !dev || !dev->msi_enabled)
return;
pci_msi_shutdown(dev);
entry = list_entry(dev->msi_list.next, struct msi_desc, list);
if (!entry->dev || entry->msi_attrib.type != PCI_CAP_ID_MSI)
return;
msi_free_irqs(dev);
}
EXPORT_SYMBOL(pci_disable_msi);
static int msi_free_irqs(struct pci_dev* dev)
{
struct msi_desc *entry, *tmp;
list_for_each_entry(entry, &dev->msi_list, list) {
if (entry->irq)
BUG_ON(irq_has_action(entry->irq));
}
arch_teardown_msi_irqs(dev);
list_for_each_entry_safe(entry, tmp, &dev->msi_list, list) {
if (entry->msi_attrib.type == PCI_CAP_ID_MSIX) {
writel(1, entry->mask_base + entry->msi_attrib.entry_nr
* PCI_MSIX_ENTRY_SIZE
+ PCI_MSIX_ENTRY_VECTOR_CTRL_OFFSET);
msi: mask the msix vector before we unmap it With these two lines in the reverse order the drives/block/ccis.c was oopsing in msi_free_irqs. Silly us calling writel on an area after we unmap it. BUG: unable to handle kernel paging request at virtual address f8b2200c printing eip: c01e9cc7 *pdpt = 0000000000003001 *pde = 0000000037e48067 *pte = 0000000000000000 Oops: 0002 [#1] SMP Modules linked in: cciss ipv6 parport_pc lp parport autofs4 i2c_dev i2c_core sunrpc loop dm_multipath button battery asus_acpi ac tg3 floppy sg dm_snapshot dm_zero dm_mirror ext3 jbd dm_mod ata_piix libata mptsas scsi_transport_sas mptspi scsi_transport_spi mptscsih mptbase sd_mod scsi_mod CPU: 1 EIP: 0060:[<c01e9cc7>] Not tainted VLI EFLAGS: 00010286 (2.6.22-rc2-gd2579053 #1) EIP is at msi_free_irqs+0x81/0xbe eax: f8b22000 ebx: f71f3180 ecx: f7fff280 edx: c1886eb8 esi: f7c4e800 edi: f7c4ec48 ebp: 00000002 esp: f5a0dec8 ds: 007b es: 007b fs: 00d8 gs: 0033 ss: 0068 Process rmmod (pid: 5286, ti=f5a0d000 task=c47d2550 task.ti=f5a0d000) Stack: 00000002 f8b72294 00000400 f8b69ca7 f8b6bc6c 00000002 00000000 00000000 00000000 00000000 00000000 f5a997f4 f8b69d61 f7c5a4b0 f7c4e848 f7c4e848 f7c4e800 f7c4e800 f8b72294 f7c4e848 f8b72294 c01e3cdf f7c4e848 c024c469 Call Trace: [<f8b69ca7>] cciss_shutdown+0xae/0xc3 [cciss] [<f8b69d61>] cciss_remove_one+0xa5/0x178 [cciss] [<c01e3cdf>] pci_device_remove+0x16/0x35 [<c024c469>] __device_release_driver+0x71/0x8e [<c024c56e>] driver_detach+0xa0/0xde [<c024bc5c>] bus_remove_driver+0x27/0x41 [<c01e3ef3>] pci_unregister_driver+0xb/0x13 [<f8b6a343>] cciss_cleanup+0xf/0x51 [cciss] [<c0139ced>] sys_delete_module+0x110/0x135 [<c0104c7a>] sysenter_past_esp+0x5f/0x85 Here's a patch that just reverses the 2 lines of code as Eric suggests. Please consider this for inclusion. Signed-off-by: Mike Miller <mike.miller@hp.com> Signed-off-by: Chase Maupin <chase.maupin@hp.com> Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com> Cc: Andi Kleen <ak@suse.de> Cc: Greg KH <greg@kroah.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2007-06-01 07:46:33 +00:00
if (list_is_last(&entry->list, &dev->msi_list))
iounmap(entry->mask_base);
}
list_del(&entry->list);
kfree(entry);
}
return 0;
}
/**
* pci_enable_msix - configure device's MSI-X capability structure
* @dev: pointer to the pci_dev data structure of MSI-X device function
* @entries: pointer to an array of MSI-X entries
* @nvec: number of MSI-X irqs requested for allocation by device driver
*
* Setup the MSI-X capability structure of device function with the number
* of requested irqs upon its software driver call to request for
* MSI-X mode enabled on its hardware device function. A return of zero
* indicates the successful configuration of MSI-X capability structure
* with new allocated MSI-X irqs. A return of < 0 indicates a failure.
* Or a return of > 0 indicates that driver request is exceeding the number
* of irqs available. Driver should use the returned value to re-send
* its request.
**/
int pci_enable_msix(struct pci_dev* dev, struct msix_entry *entries, int nvec)
{
int status, pos, nr_entries;
int i, j;
u16 control;
if (!entries)
return -EINVAL;
status = pci_msi_check_device(dev, nvec, PCI_CAP_ID_MSIX);
if (status)
return status;
pos = pci_find_capability(dev, PCI_CAP_ID_MSIX);
pci_read_config_word(dev, msi_control_reg(pos), &control);
nr_entries = multi_msix_capable(control);
if (nvec > nr_entries)
return -EINVAL;
/* Check for any invalid entries */
for (i = 0; i < nvec; i++) {
if (entries[i].entry >= nr_entries)
return -EINVAL; /* invalid entry */
for (j = i + 1; j < nvec; j++) {
if (entries[i].entry == entries[j].entry)
return -EINVAL; /* duplicate entry */
}
}
WARN_ON(!!dev->msix_enabled);
/* Check whether driver already requested for MSI irq */
if (dev->msi_enabled) {
printk(KERN_INFO "PCI: %s: Can't enable MSI-X. "
"Device already has an MSI irq assigned\n",
pci_name(dev));
return -EINVAL;
}
status = msix_capability_init(dev, entries, nvec);
return status;
}
EXPORT_SYMBOL(pci_enable_msix);
static void msix_free_all_irqs(struct pci_dev *dev)
{
msi_free_irqs(dev);
}
void pci_msix_shutdown(struct pci_dev* dev)
{
if (!pci_msi_enable || !dev || !dev->msix_enabled)
return;
msix_set_enable(dev, 0);
pci_intx_for_msi(dev, 1);
dev->msix_enabled = 0;
}
void pci_disable_msix(struct pci_dev* dev)
{
if (!pci_msi_enable || !dev || !dev->msix_enabled)
return;
pci_msix_shutdown(dev);
msix_free_all_irqs(dev);
}
EXPORT_SYMBOL(pci_disable_msix);
/**
* msi_remove_pci_irq_vectors - reclaim MSI(X) irqs to unused state
* @dev: pointer to the pci_dev data structure of MSI(X) device function
*
* Being called during hotplug remove, from which the device function
* is hot-removed. All previous assigned MSI/MSI-X irqs, if
* allocated for this device function, are reclaimed to unused state,
* which may be used later on.
**/
void msi_remove_pci_irq_vectors(struct pci_dev* dev)
{
if (!pci_msi_enable || !dev)
return;
if (dev->msi_enabled)
msi_free_irqs(dev);
if (dev->msix_enabled)
msix_free_all_irqs(dev);
}
void pci_no_msi(void)
{
pci_msi_enable = 0;
}
void pci_msi_init_pci_dev(struct pci_dev *dev)
{
INIT_LIST_HEAD(&dev->msi_list);
}