OpenCloudOS-Kernel/drivers/pci/hotplug/pciehp_ctrl.c

565 lines
14 KiB
C
Raw Normal View History

/*
* PCI Express Hot Plug Controller Driver
*
* Copyright (C) 1995,2001 Compaq Computer Corporation
* Copyright (C) 2001 Greg Kroah-Hartman (greg@kroah.com)
* Copyright (C) 2001 IBM Corp.
* Copyright (C) 2003-2004 Intel Corporation
*
* All rights reserved.
*
* 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, GOOD TITLE or
* NON INFRINGEMENT. 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., 675 Mass Ave, Cambridge, MA 02139, USA.
*
* Send feedback to <greg@kroah.com>, <kristen.c.accardi@intel.com>
*
*/
#include <linux/module.h>
#include <linux/kernel.h>
#include <linux/types.h>
include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h percpu.h is included by sched.h and module.h and thus ends up being included when building most .c files. percpu.h includes slab.h which in turn includes gfp.h making everything defined by the two files universally available and complicating inclusion dependencies. percpu.h -> slab.h dependency is about to be removed. Prepare for this change by updating users of gfp and slab facilities include those headers directly instead of assuming availability. As this conversion needs to touch large number of source files, the following script is used as the basis of conversion. http://userweb.kernel.org/~tj/misc/slabh-sweep.py The script does the followings. * Scan files for gfp and slab usages and update includes such that only the necessary includes are there. ie. if only gfp is used, gfp.h, if slab is used, slab.h. * When the script inserts a new include, it looks at the include blocks and try to put the new include such that its order conforms to its surrounding. It's put in the include block which contains core kernel includes, in the same order that the rest are ordered - alphabetical, Christmas tree, rev-Xmas-tree or at the end if there doesn't seem to be any matching order. * If the script can't find a place to put a new include (mostly because the file doesn't have fitting include block), it prints out an error message indicating which .h file needs to be added to the file. The conversion was done in the following steps. 1. The initial automatic conversion of all .c files updated slightly over 4000 files, deleting around 700 includes and adding ~480 gfp.h and ~3000 slab.h inclusions. The script emitted errors for ~400 files. 2. Each error was manually checked. Some didn't need the inclusion, some needed manual addition while adding it to implementation .h or embedding .c file was more appropriate for others. This step added inclusions to around 150 files. 3. The script was run again and the output was compared to the edits from #2 to make sure no file was left behind. 4. Several build tests were done and a couple of problems were fixed. e.g. lib/decompress_*.c used malloc/free() wrappers around slab APIs requiring slab.h to be added manually. 5. The script was run on all .h files but without automatically editing them as sprinkling gfp.h and slab.h inclusions around .h files could easily lead to inclusion dependency hell. Most gfp.h inclusion directives were ignored as stuff from gfp.h was usually wildly available and often used in preprocessor macros. Each slab.h inclusion directive was examined and added manually as necessary. 6. percpu.h was updated not to include slab.h. 7. Build test were done on the following configurations and failures were fixed. CONFIG_GCOV_KERNEL was turned off for all tests (as my distributed build env didn't work with gcov compiles) and a few more options had to be turned off depending on archs to make things build (like ipr on powerpc/64 which failed due to missing writeq). * x86 and x86_64 UP and SMP allmodconfig and a custom test config. * powerpc and powerpc64 SMP allmodconfig * sparc and sparc64 SMP allmodconfig * ia64 SMP allmodconfig * s390 SMP allmodconfig * alpha SMP allmodconfig * um on x86_64 SMP allmodconfig 8. percpu.h modifications were reverted so that it could be applied as a separate patch and serve as bisection point. Given the fact that I had only a couple of failures from tests on step 6, I'm fairly confident about the coverage of this conversion patch. If there is a breakage, it's likely to be something in one of the arch headers which should be easily discoverable easily on most builds of the specific arch. Signed-off-by: Tejun Heo <tj@kernel.org> Guess-its-ok-by: Christoph Lameter <cl@linux-foundation.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
2010-03-24 16:04:11 +08:00
#include <linux/slab.h>
#include <linux/pci.h>
#include "../pci.h"
#include "pciehp.h"
static void interrupt_event_handler(struct work_struct *work);
void pciehp_queue_interrupt_event(struct slot *p_slot, u32 event_type)
{
struct event_info *info;
info = kmalloc(sizeof(*info), GFP_ATOMIC);
if (!info) {
ctrl_err(p_slot->ctrl, "dropped event %d (ENOMEM)\n", event_type);
return;
}
INIT_WORK(&info->work, interrupt_event_handler);
info->event_type = event_type;
info->p_slot = p_slot;
PCI: pciehp: Use per-slot workqueues to avoid deadlock When we have a hotplug-capable PCIe port with a second hotplug-capable PCIe port below it, removing the device below the upstream port causes a deadlock. The deadlock happens because we use the pciehp_wq workqueue to run pciehp_power_thread(), which uses pciehp_disable_slot() to remove devices below the upstream port. When we remove the downstream PCIe port, we call pciehp_remove(), the pciehp driver's .remove() method. That calls flush_workqueue(pciehp_wq), which deadlocks because the pciehp_power_thread() work item is still running. This patch avoids the deadlock by creating a workqueue for every PCIe port and removing the single shared workqueue. Here's the call path that leads to the deadlock: pciehp_queue_pushbutton_work queue_work(pciehp_wq) # queue pciehp_power_thread ... pciehp_power_thread pciehp_disable_slot remove_board pciehp_unconfigure_device pci_stop_and_remove_bus_device ... pciehp_remove # pciehp driver .remove method pciehp_release_ctrl pcie_cleanup_slot flush_workqueue(pciehp_wq) This is fairly urgent because it can be caused by simply unplugging a Thunderbolt adapter, as reported by Daniel below. [bhelgaas: changelog] Reference: http://lkml.kernel.org/r/CAMVG2ssiRgcTD1bej2tkUUfsWmpL5eNtPcNif9va2-Gzb2u8nQ@mail.gmail.com Reported-and-tested-by: Daniel J Blueman <daniel@quora.org> Reviewed-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com> Signed-off-by: Yijing Wang <wangyijing@huawei.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> CC: stable@vger.kernel.org
2013-01-11 10:15:54 +08:00
queue_work(p_slot->wq, &info->work);
}
/* The following routines constitute the bulk of the
hotplug controller logic
*/
static void set_slot_off(struct controller *ctrl, struct slot *pslot)
{
/* turn off slot, turn on Amber LED, turn off Green LED if supported*/
if (POWER_CTRL(ctrl)) {
pciehp_power_off_slot(pslot);
/*
* After turning power off, we must wait for at least 1 second
* before taking any action that relies on power having been
* removed from the slot/adapter.
*/
msleep(1000);
}
pciehp_green_led_off(pslot);
pciehp_set_attention_status(pslot, 1);
}
/**
* board_added - Called after a board has been added to the system.
* @p_slot: &slot where board is added
*
* Turns power on for the board.
* Configures board.
*/
static int board_added(struct slot *p_slot)
{
int retval = 0;
struct controller *ctrl = p_slot->ctrl;
struct pci_bus *parent = ctrl->pcie->port->subordinate;
if (POWER_CTRL(ctrl)) {
/* Power on slot */
retval = pciehp_power_on_slot(p_slot);
if (retval)
return retval;
}
pciehp_green_led_blink(p_slot);
/* Check link training status */
retval = pciehp_check_link_status(ctrl);
if (retval) {
ctrl_err(ctrl, "Failed to check link status\n");
goto err_exit;
}
/* Check for a power fault */
if (ctrl->power_fault_detected || pciehp_query_power_fault(p_slot)) {
ctrl_err(ctrl, "Power fault on slot %s\n", slot_name(p_slot));
retval = -EIO;
goto err_exit;
}
retval = pciehp_configure_device(p_slot);
if (retval) {
ctrl_err(ctrl, "Cannot add device at %04x:%02x:00\n",
pci_domain_nr(parent), parent->number);
if (retval != -EEXIST)
goto err_exit;
}
pciehp_green_led_on(p_slot);
return 0;
err_exit:
set_slot_off(ctrl, p_slot);
return retval;
}
/**
* remove_board - Turns off slot and LEDs
* @p_slot: slot where board is being removed
*/
static int remove_board(struct slot *p_slot)
{
int retval;
struct controller *ctrl = p_slot->ctrl;
retval = pciehp_unconfigure_device(p_slot);
if (retval)
return retval;
if (POWER_CTRL(ctrl)) {
pciehp_power_off_slot(p_slot);
/*
* After turning power off, we must wait for at least 1 second
* before taking any action that relies on power having been
* removed from the slot/adapter.
*/
msleep(1000);
}
/* turn off Green LED */
pciehp_green_led_off(p_slot);
return 0;
}
struct power_work_info {
struct slot *p_slot;
struct work_struct work;
PCI: pciehp: Ensure very fast hotplug events are also processed Today, this is how all the hotplug and unplug events work: Hotplug / Removal needs to be done => Set slot->state (protected by slot->lock) to either POWERON_STATE (for enabling) or POWEROFF_STATE (for disabling). => Submit the work item for pciehp_power_thread() to slot->wq. Problem: There is a problem if the hotplug events can happen fast enough that they do not give SW enough time to add or remove the new devices. => Assume: Event for unplug comes (e.g. surprise removal). But before the pciehp_power_thread() work item was executed, the card was replaced by another card, causing surprise hotplug event. => What goes wrong: => The hot-removal event sets slot->state to POWEROFF_STATE, and schedules the pciehp_power_thread(). => The hot-add event sets slot->state to POWERON_STATE, and schedules the pciehp_power_thread(). => Now the pciehp_power_thread() is scheduled twice, and on both occasions it will find POWERON_STATE and will try to add the devices on the slot, and will fail complaining that the devices already exist. => Why this is a problem: If the device was replaced between the hot removal and hot-add, then we should unload the old driver and reload the new one. This does not happen today. The kernel or the driver is not even aware that the device was replaced. The problem is that the pciehp_power_thread() only looks at the slot->state which would only contain the *latest* state - not the actual event (add / remove) that was the intent of the IRQ handler who submitted the work. What this patch does: => Hotplug events pass on an actual request (for addition or removal) to pciehp_power_thread() which is local to that work item submission. => pciehp_power_thread() does not need to look at slote->state and hence no locks needed in that. => Essentially this results in all the hotplug and unplug events "replayed" by pciehp_power_thread(). Signed-off-by: Rajat Jain <rajatxjain@gmail.com> Signed-off-by: Rajat Jain <rajatjain@juniper.net> Signed-off-by: Guenter Roeck <groeck@juniper.net> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
2014-02-05 10:30:56 +08:00
unsigned int req;
#define DISABLE_REQ 0
#define ENABLE_REQ 1
};
/**
* pciehp_power_thread - handle pushbutton events
* @work: &struct work_struct describing work to be done
*
* Scheduled procedure to handle blocking stuff for the pushbuttons.
* Handles all pending events and exits.
*/
static void pciehp_power_thread(struct work_struct *work)
{
struct power_work_info *info =
container_of(work, struct power_work_info, work);
struct slot *p_slot = info->p_slot;
int ret;
PCI: pciehp: Ensure very fast hotplug events are also processed Today, this is how all the hotplug and unplug events work: Hotplug / Removal needs to be done => Set slot->state (protected by slot->lock) to either POWERON_STATE (for enabling) or POWEROFF_STATE (for disabling). => Submit the work item for pciehp_power_thread() to slot->wq. Problem: There is a problem if the hotplug events can happen fast enough that they do not give SW enough time to add or remove the new devices. => Assume: Event for unplug comes (e.g. surprise removal). But before the pciehp_power_thread() work item was executed, the card was replaced by another card, causing surprise hotplug event. => What goes wrong: => The hot-removal event sets slot->state to POWEROFF_STATE, and schedules the pciehp_power_thread(). => The hot-add event sets slot->state to POWERON_STATE, and schedules the pciehp_power_thread(). => Now the pciehp_power_thread() is scheduled twice, and on both occasions it will find POWERON_STATE and will try to add the devices on the slot, and will fail complaining that the devices already exist. => Why this is a problem: If the device was replaced between the hot removal and hot-add, then we should unload the old driver and reload the new one. This does not happen today. The kernel or the driver is not even aware that the device was replaced. The problem is that the pciehp_power_thread() only looks at the slot->state which would only contain the *latest* state - not the actual event (add / remove) that was the intent of the IRQ handler who submitted the work. What this patch does: => Hotplug events pass on an actual request (for addition or removal) to pciehp_power_thread() which is local to that work item submission. => pciehp_power_thread() does not need to look at slote->state and hence no locks needed in that. => Essentially this results in all the hotplug and unplug events "replayed" by pciehp_power_thread(). Signed-off-by: Rajat Jain <rajatxjain@gmail.com> Signed-off-by: Rajat Jain <rajatjain@juniper.net> Signed-off-by: Guenter Roeck <groeck@juniper.net> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
2014-02-05 10:30:56 +08:00
switch (info->req) {
case DISABLE_REQ:
mutex_lock(&p_slot->hotplug_lock);
pciehp_disable_slot(p_slot);
mutex_unlock(&p_slot->hotplug_lock);
mutex_lock(&p_slot->lock);
p_slot->state = STATIC_STATE;
mutex_unlock(&p_slot->lock);
PCI: pciehp: Ensure very fast hotplug events are also processed Today, this is how all the hotplug and unplug events work: Hotplug / Removal needs to be done => Set slot->state (protected by slot->lock) to either POWERON_STATE (for enabling) or POWEROFF_STATE (for disabling). => Submit the work item for pciehp_power_thread() to slot->wq. Problem: There is a problem if the hotplug events can happen fast enough that they do not give SW enough time to add or remove the new devices. => Assume: Event for unplug comes (e.g. surprise removal). But before the pciehp_power_thread() work item was executed, the card was replaced by another card, causing surprise hotplug event. => What goes wrong: => The hot-removal event sets slot->state to POWEROFF_STATE, and schedules the pciehp_power_thread(). => The hot-add event sets slot->state to POWERON_STATE, and schedules the pciehp_power_thread(). => Now the pciehp_power_thread() is scheduled twice, and on both occasions it will find POWERON_STATE and will try to add the devices on the slot, and will fail complaining that the devices already exist. => Why this is a problem: If the device was replaced between the hot removal and hot-add, then we should unload the old driver and reload the new one. This does not happen today. The kernel or the driver is not even aware that the device was replaced. The problem is that the pciehp_power_thread() only looks at the slot->state which would only contain the *latest* state - not the actual event (add / remove) that was the intent of the IRQ handler who submitted the work. What this patch does: => Hotplug events pass on an actual request (for addition or removal) to pciehp_power_thread() which is local to that work item submission. => pciehp_power_thread() does not need to look at slote->state and hence no locks needed in that. => Essentially this results in all the hotplug and unplug events "replayed" by pciehp_power_thread(). Signed-off-by: Rajat Jain <rajatxjain@gmail.com> Signed-off-by: Rajat Jain <rajatjain@juniper.net> Signed-off-by: Guenter Roeck <groeck@juniper.net> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
2014-02-05 10:30:56 +08:00
break;
case ENABLE_REQ:
mutex_lock(&p_slot->hotplug_lock);
ret = pciehp_enable_slot(p_slot);
mutex_unlock(&p_slot->hotplug_lock);
if (ret)
pciehp_green_led_off(p_slot);
mutex_lock(&p_slot->lock);
p_slot->state = STATIC_STATE;
PCI: pciehp: Ensure very fast hotplug events are also processed Today, this is how all the hotplug and unplug events work: Hotplug / Removal needs to be done => Set slot->state (protected by slot->lock) to either POWERON_STATE (for enabling) or POWEROFF_STATE (for disabling). => Submit the work item for pciehp_power_thread() to slot->wq. Problem: There is a problem if the hotplug events can happen fast enough that they do not give SW enough time to add or remove the new devices. => Assume: Event for unplug comes (e.g. surprise removal). But before the pciehp_power_thread() work item was executed, the card was replaced by another card, causing surprise hotplug event. => What goes wrong: => The hot-removal event sets slot->state to POWEROFF_STATE, and schedules the pciehp_power_thread(). => The hot-add event sets slot->state to POWERON_STATE, and schedules the pciehp_power_thread(). => Now the pciehp_power_thread() is scheduled twice, and on both occasions it will find POWERON_STATE and will try to add the devices on the slot, and will fail complaining that the devices already exist. => Why this is a problem: If the device was replaced between the hot removal and hot-add, then we should unload the old driver and reload the new one. This does not happen today. The kernel or the driver is not even aware that the device was replaced. The problem is that the pciehp_power_thread() only looks at the slot->state which would only contain the *latest* state - not the actual event (add / remove) that was the intent of the IRQ handler who submitted the work. What this patch does: => Hotplug events pass on an actual request (for addition or removal) to pciehp_power_thread() which is local to that work item submission. => pciehp_power_thread() does not need to look at slote->state and hence no locks needed in that. => Essentially this results in all the hotplug and unplug events "replayed" by pciehp_power_thread(). Signed-off-by: Rajat Jain <rajatxjain@gmail.com> Signed-off-by: Rajat Jain <rajatjain@juniper.net> Signed-off-by: Guenter Roeck <groeck@juniper.net> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
2014-02-05 10:30:56 +08:00
mutex_unlock(&p_slot->lock);
break;
default:
break;
}
kfree(info);
}
void pciehp_queue_pushbutton_work(struct work_struct *work)
{
struct slot *p_slot = container_of(work, struct slot, work.work);
struct power_work_info *info;
info = kmalloc(sizeof(*info), GFP_KERNEL);
if (!info) {
ctrl_err(p_slot->ctrl, "%s: Cannot allocate memory\n",
__func__);
return;
}
info->p_slot = p_slot;
INIT_WORK(&info->work, pciehp_power_thread);
mutex_lock(&p_slot->lock);
switch (p_slot->state) {
case BLINKINGOFF_STATE:
p_slot->state = POWEROFF_STATE;
PCI: pciehp: Ensure very fast hotplug events are also processed Today, this is how all the hotplug and unplug events work: Hotplug / Removal needs to be done => Set slot->state (protected by slot->lock) to either POWERON_STATE (for enabling) or POWEROFF_STATE (for disabling). => Submit the work item for pciehp_power_thread() to slot->wq. Problem: There is a problem if the hotplug events can happen fast enough that they do not give SW enough time to add or remove the new devices. => Assume: Event for unplug comes (e.g. surprise removal). But before the pciehp_power_thread() work item was executed, the card was replaced by another card, causing surprise hotplug event. => What goes wrong: => The hot-removal event sets slot->state to POWEROFF_STATE, and schedules the pciehp_power_thread(). => The hot-add event sets slot->state to POWERON_STATE, and schedules the pciehp_power_thread(). => Now the pciehp_power_thread() is scheduled twice, and on both occasions it will find POWERON_STATE and will try to add the devices on the slot, and will fail complaining that the devices already exist. => Why this is a problem: If the device was replaced between the hot removal and hot-add, then we should unload the old driver and reload the new one. This does not happen today. The kernel or the driver is not even aware that the device was replaced. The problem is that the pciehp_power_thread() only looks at the slot->state which would only contain the *latest* state - not the actual event (add / remove) that was the intent of the IRQ handler who submitted the work. What this patch does: => Hotplug events pass on an actual request (for addition or removal) to pciehp_power_thread() which is local to that work item submission. => pciehp_power_thread() does not need to look at slote->state and hence no locks needed in that. => Essentially this results in all the hotplug and unplug events "replayed" by pciehp_power_thread(). Signed-off-by: Rajat Jain <rajatxjain@gmail.com> Signed-off-by: Rajat Jain <rajatjain@juniper.net> Signed-off-by: Guenter Roeck <groeck@juniper.net> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
2014-02-05 10:30:56 +08:00
info->req = DISABLE_REQ;
break;
case BLINKINGON_STATE:
p_slot->state = POWERON_STATE;
PCI: pciehp: Ensure very fast hotplug events are also processed Today, this is how all the hotplug and unplug events work: Hotplug / Removal needs to be done => Set slot->state (protected by slot->lock) to either POWERON_STATE (for enabling) or POWEROFF_STATE (for disabling). => Submit the work item for pciehp_power_thread() to slot->wq. Problem: There is a problem if the hotplug events can happen fast enough that they do not give SW enough time to add or remove the new devices. => Assume: Event for unplug comes (e.g. surprise removal). But before the pciehp_power_thread() work item was executed, the card was replaced by another card, causing surprise hotplug event. => What goes wrong: => The hot-removal event sets slot->state to POWEROFF_STATE, and schedules the pciehp_power_thread(). => The hot-add event sets slot->state to POWERON_STATE, and schedules the pciehp_power_thread(). => Now the pciehp_power_thread() is scheduled twice, and on both occasions it will find POWERON_STATE and will try to add the devices on the slot, and will fail complaining that the devices already exist. => Why this is a problem: If the device was replaced between the hot removal and hot-add, then we should unload the old driver and reload the new one. This does not happen today. The kernel or the driver is not even aware that the device was replaced. The problem is that the pciehp_power_thread() only looks at the slot->state which would only contain the *latest* state - not the actual event (add / remove) that was the intent of the IRQ handler who submitted the work. What this patch does: => Hotplug events pass on an actual request (for addition or removal) to pciehp_power_thread() which is local to that work item submission. => pciehp_power_thread() does not need to look at slote->state and hence no locks needed in that. => Essentially this results in all the hotplug and unplug events "replayed" by pciehp_power_thread(). Signed-off-by: Rajat Jain <rajatxjain@gmail.com> Signed-off-by: Rajat Jain <rajatjain@juniper.net> Signed-off-by: Guenter Roeck <groeck@juniper.net> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
2014-02-05 10:30:56 +08:00
info->req = ENABLE_REQ;
break;
default:
kfree(info);
goto out;
}
PCI: pciehp: Use per-slot workqueues to avoid deadlock When we have a hotplug-capable PCIe port with a second hotplug-capable PCIe port below it, removing the device below the upstream port causes a deadlock. The deadlock happens because we use the pciehp_wq workqueue to run pciehp_power_thread(), which uses pciehp_disable_slot() to remove devices below the upstream port. When we remove the downstream PCIe port, we call pciehp_remove(), the pciehp driver's .remove() method. That calls flush_workqueue(pciehp_wq), which deadlocks because the pciehp_power_thread() work item is still running. This patch avoids the deadlock by creating a workqueue for every PCIe port and removing the single shared workqueue. Here's the call path that leads to the deadlock: pciehp_queue_pushbutton_work queue_work(pciehp_wq) # queue pciehp_power_thread ... pciehp_power_thread pciehp_disable_slot remove_board pciehp_unconfigure_device pci_stop_and_remove_bus_device ... pciehp_remove # pciehp driver .remove method pciehp_release_ctrl pcie_cleanup_slot flush_workqueue(pciehp_wq) This is fairly urgent because it can be caused by simply unplugging a Thunderbolt adapter, as reported by Daniel below. [bhelgaas: changelog] Reference: http://lkml.kernel.org/r/CAMVG2ssiRgcTD1bej2tkUUfsWmpL5eNtPcNif9va2-Gzb2u8nQ@mail.gmail.com Reported-and-tested-by: Daniel J Blueman <daniel@quora.org> Reviewed-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com> Signed-off-by: Yijing Wang <wangyijing@huawei.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> CC: stable@vger.kernel.org
2013-01-11 10:15:54 +08:00
queue_work(p_slot->wq, &info->work);
out:
mutex_unlock(&p_slot->lock);
}
/*
* Note: This function must be called with slot->lock held
*/
static void handle_button_press_event(struct slot *p_slot)
{
struct controller *ctrl = p_slot->ctrl;
u8 getstatus;
switch (p_slot->state) {
case STATIC_STATE:
pciehp_get_power_status(p_slot, &getstatus);
if (getstatus) {
p_slot->state = BLINKINGOFF_STATE;
ctrl_info(ctrl, "PCI slot #%s - powering off due to button press\n",
slot_name(p_slot));
} else {
p_slot->state = BLINKINGON_STATE;
ctrl_info(ctrl, "PCI slot #%s - powering on due to button press\n",
slot_name(p_slot));
}
/* blink green LED and turn off amber */
pciehp_green_led_blink(p_slot);
pciehp_set_attention_status(p_slot, 0);
PCI: pciehp: Use per-slot workqueues to avoid deadlock When we have a hotplug-capable PCIe port with a second hotplug-capable PCIe port below it, removing the device below the upstream port causes a deadlock. The deadlock happens because we use the pciehp_wq workqueue to run pciehp_power_thread(), which uses pciehp_disable_slot() to remove devices below the upstream port. When we remove the downstream PCIe port, we call pciehp_remove(), the pciehp driver's .remove() method. That calls flush_workqueue(pciehp_wq), which deadlocks because the pciehp_power_thread() work item is still running. This patch avoids the deadlock by creating a workqueue for every PCIe port and removing the single shared workqueue. Here's the call path that leads to the deadlock: pciehp_queue_pushbutton_work queue_work(pciehp_wq) # queue pciehp_power_thread ... pciehp_power_thread pciehp_disable_slot remove_board pciehp_unconfigure_device pci_stop_and_remove_bus_device ... pciehp_remove # pciehp driver .remove method pciehp_release_ctrl pcie_cleanup_slot flush_workqueue(pciehp_wq) This is fairly urgent because it can be caused by simply unplugging a Thunderbolt adapter, as reported by Daniel below. [bhelgaas: changelog] Reference: http://lkml.kernel.org/r/CAMVG2ssiRgcTD1bej2tkUUfsWmpL5eNtPcNif9va2-Gzb2u8nQ@mail.gmail.com Reported-and-tested-by: Daniel J Blueman <daniel@quora.org> Reviewed-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com> Signed-off-by: Yijing Wang <wangyijing@huawei.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> CC: stable@vger.kernel.org
2013-01-11 10:15:54 +08:00
queue_delayed_work(p_slot->wq, &p_slot->work, 5*HZ);
break;
case BLINKINGOFF_STATE:
case BLINKINGON_STATE:
/*
* Cancel if we are still blinking; this means that we
* press the attention again before the 5 sec. limit
* expires to cancel hot-add or hot-remove
*/
ctrl_info(ctrl, "Button cancel on Slot(%s)\n", slot_name(p_slot));
cancel_delayed_work(&p_slot->work);
if (p_slot->state == BLINKINGOFF_STATE)
pciehp_green_led_on(p_slot);
else
pciehp_green_led_off(p_slot);
pciehp_set_attention_status(p_slot, 0);
ctrl_info(ctrl, "PCI slot #%s - action canceled due to button press\n",
slot_name(p_slot));
p_slot->state = STATIC_STATE;
break;
case POWEROFF_STATE:
case POWERON_STATE:
/*
* Ignore if the slot is on power-on or power-off state;
* this means that the previous attention button action
* to hot-add or hot-remove is undergoing
*/
ctrl_info(ctrl, "Button ignore on Slot(%s)\n", slot_name(p_slot));
break;
default:
ctrl_warn(ctrl, "ignoring invalid state %#x\n", p_slot->state);
break;
}
}
/*
* Note: This function must be called with slot->lock held
*/
static void handle_surprise_event(struct slot *p_slot)
{
u8 getstatus;
struct power_work_info *info;
info = kmalloc(sizeof(*info), GFP_KERNEL);
if (!info) {
ctrl_err(p_slot->ctrl, "%s: Cannot allocate memory\n",
__func__);
return;
}
info->p_slot = p_slot;
INIT_WORK(&info->work, pciehp_power_thread);
pciehp_get_adapter_status(p_slot, &getstatus);
PCI: pciehp: Ensure very fast hotplug events are also processed Today, this is how all the hotplug and unplug events work: Hotplug / Removal needs to be done => Set slot->state (protected by slot->lock) to either POWERON_STATE (for enabling) or POWEROFF_STATE (for disabling). => Submit the work item for pciehp_power_thread() to slot->wq. Problem: There is a problem if the hotplug events can happen fast enough that they do not give SW enough time to add or remove the new devices. => Assume: Event for unplug comes (e.g. surprise removal). But before the pciehp_power_thread() work item was executed, the card was replaced by another card, causing surprise hotplug event. => What goes wrong: => The hot-removal event sets slot->state to POWEROFF_STATE, and schedules the pciehp_power_thread(). => The hot-add event sets slot->state to POWERON_STATE, and schedules the pciehp_power_thread(). => Now the pciehp_power_thread() is scheduled twice, and on both occasions it will find POWERON_STATE and will try to add the devices on the slot, and will fail complaining that the devices already exist. => Why this is a problem: If the device was replaced between the hot removal and hot-add, then we should unload the old driver and reload the new one. This does not happen today. The kernel or the driver is not even aware that the device was replaced. The problem is that the pciehp_power_thread() only looks at the slot->state which would only contain the *latest* state - not the actual event (add / remove) that was the intent of the IRQ handler who submitted the work. What this patch does: => Hotplug events pass on an actual request (for addition or removal) to pciehp_power_thread() which is local to that work item submission. => pciehp_power_thread() does not need to look at slote->state and hence no locks needed in that. => Essentially this results in all the hotplug and unplug events "replayed" by pciehp_power_thread(). Signed-off-by: Rajat Jain <rajatxjain@gmail.com> Signed-off-by: Rajat Jain <rajatjain@juniper.net> Signed-off-by: Guenter Roeck <groeck@juniper.net> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
2014-02-05 10:30:56 +08:00
if (!getstatus) {
p_slot->state = POWEROFF_STATE;
PCI: pciehp: Ensure very fast hotplug events are also processed Today, this is how all the hotplug and unplug events work: Hotplug / Removal needs to be done => Set slot->state (protected by slot->lock) to either POWERON_STATE (for enabling) or POWEROFF_STATE (for disabling). => Submit the work item for pciehp_power_thread() to slot->wq. Problem: There is a problem if the hotplug events can happen fast enough that they do not give SW enough time to add or remove the new devices. => Assume: Event for unplug comes (e.g. surprise removal). But before the pciehp_power_thread() work item was executed, the card was replaced by another card, causing surprise hotplug event. => What goes wrong: => The hot-removal event sets slot->state to POWEROFF_STATE, and schedules the pciehp_power_thread(). => The hot-add event sets slot->state to POWERON_STATE, and schedules the pciehp_power_thread(). => Now the pciehp_power_thread() is scheduled twice, and on both occasions it will find POWERON_STATE and will try to add the devices on the slot, and will fail complaining that the devices already exist. => Why this is a problem: If the device was replaced between the hot removal and hot-add, then we should unload the old driver and reload the new one. This does not happen today. The kernel or the driver is not even aware that the device was replaced. The problem is that the pciehp_power_thread() only looks at the slot->state which would only contain the *latest* state - not the actual event (add / remove) that was the intent of the IRQ handler who submitted the work. What this patch does: => Hotplug events pass on an actual request (for addition or removal) to pciehp_power_thread() which is local to that work item submission. => pciehp_power_thread() does not need to look at slote->state and hence no locks needed in that. => Essentially this results in all the hotplug and unplug events "replayed" by pciehp_power_thread(). Signed-off-by: Rajat Jain <rajatxjain@gmail.com> Signed-off-by: Rajat Jain <rajatjain@juniper.net> Signed-off-by: Guenter Roeck <groeck@juniper.net> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
2014-02-05 10:30:56 +08:00
info->req = DISABLE_REQ;
} else {
p_slot->state = POWERON_STATE;
PCI: pciehp: Ensure very fast hotplug events are also processed Today, this is how all the hotplug and unplug events work: Hotplug / Removal needs to be done => Set slot->state (protected by slot->lock) to either POWERON_STATE (for enabling) or POWEROFF_STATE (for disabling). => Submit the work item for pciehp_power_thread() to slot->wq. Problem: There is a problem if the hotplug events can happen fast enough that they do not give SW enough time to add or remove the new devices. => Assume: Event for unplug comes (e.g. surprise removal). But before the pciehp_power_thread() work item was executed, the card was replaced by another card, causing surprise hotplug event. => What goes wrong: => The hot-removal event sets slot->state to POWEROFF_STATE, and schedules the pciehp_power_thread(). => The hot-add event sets slot->state to POWERON_STATE, and schedules the pciehp_power_thread(). => Now the pciehp_power_thread() is scheduled twice, and on both occasions it will find POWERON_STATE and will try to add the devices on the slot, and will fail complaining that the devices already exist. => Why this is a problem: If the device was replaced between the hot removal and hot-add, then we should unload the old driver and reload the new one. This does not happen today. The kernel or the driver is not even aware that the device was replaced. The problem is that the pciehp_power_thread() only looks at the slot->state which would only contain the *latest* state - not the actual event (add / remove) that was the intent of the IRQ handler who submitted the work. What this patch does: => Hotplug events pass on an actual request (for addition or removal) to pciehp_power_thread() which is local to that work item submission. => pciehp_power_thread() does not need to look at slote->state and hence no locks needed in that. => Essentially this results in all the hotplug and unplug events "replayed" by pciehp_power_thread(). Signed-off-by: Rajat Jain <rajatxjain@gmail.com> Signed-off-by: Rajat Jain <rajatjain@juniper.net> Signed-off-by: Guenter Roeck <groeck@juniper.net> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
2014-02-05 10:30:56 +08:00
info->req = ENABLE_REQ;
}
PCI: pciehp: Use per-slot workqueues to avoid deadlock When we have a hotplug-capable PCIe port with a second hotplug-capable PCIe port below it, removing the device below the upstream port causes a deadlock. The deadlock happens because we use the pciehp_wq workqueue to run pciehp_power_thread(), which uses pciehp_disable_slot() to remove devices below the upstream port. When we remove the downstream PCIe port, we call pciehp_remove(), the pciehp driver's .remove() method. That calls flush_workqueue(pciehp_wq), which deadlocks because the pciehp_power_thread() work item is still running. This patch avoids the deadlock by creating a workqueue for every PCIe port and removing the single shared workqueue. Here's the call path that leads to the deadlock: pciehp_queue_pushbutton_work queue_work(pciehp_wq) # queue pciehp_power_thread ... pciehp_power_thread pciehp_disable_slot remove_board pciehp_unconfigure_device pci_stop_and_remove_bus_device ... pciehp_remove # pciehp driver .remove method pciehp_release_ctrl pcie_cleanup_slot flush_workqueue(pciehp_wq) This is fairly urgent because it can be caused by simply unplugging a Thunderbolt adapter, as reported by Daniel below. [bhelgaas: changelog] Reference: http://lkml.kernel.org/r/CAMVG2ssiRgcTD1bej2tkUUfsWmpL5eNtPcNif9va2-Gzb2u8nQ@mail.gmail.com Reported-and-tested-by: Daniel J Blueman <daniel@quora.org> Reviewed-by: Kenji Kaneshige <kaneshige.kenji@jp.fujitsu.com> Signed-off-by: Yijing Wang <wangyijing@huawei.com> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com> CC: stable@vger.kernel.org
2013-01-11 10:15:54 +08:00
queue_work(p_slot->wq, &info->work);
}
/*
* Note: This function must be called with slot->lock held
*/
static void handle_link_event(struct slot *p_slot, u32 event)
{
struct controller *ctrl = p_slot->ctrl;
struct power_work_info *info;
info = kmalloc(sizeof(*info), GFP_KERNEL);
if (!info) {
ctrl_err(p_slot->ctrl, "%s: Cannot allocate memory\n",
__func__);
return;
}
info->p_slot = p_slot;
PCI: pciehp: Ensure very fast hotplug events are also processed Today, this is how all the hotplug and unplug events work: Hotplug / Removal needs to be done => Set slot->state (protected by slot->lock) to either POWERON_STATE (for enabling) or POWEROFF_STATE (for disabling). => Submit the work item for pciehp_power_thread() to slot->wq. Problem: There is a problem if the hotplug events can happen fast enough that they do not give SW enough time to add or remove the new devices. => Assume: Event for unplug comes (e.g. surprise removal). But before the pciehp_power_thread() work item was executed, the card was replaced by another card, causing surprise hotplug event. => What goes wrong: => The hot-removal event sets slot->state to POWEROFF_STATE, and schedules the pciehp_power_thread(). => The hot-add event sets slot->state to POWERON_STATE, and schedules the pciehp_power_thread(). => Now the pciehp_power_thread() is scheduled twice, and on both occasions it will find POWERON_STATE and will try to add the devices on the slot, and will fail complaining that the devices already exist. => Why this is a problem: If the device was replaced between the hot removal and hot-add, then we should unload the old driver and reload the new one. This does not happen today. The kernel or the driver is not even aware that the device was replaced. The problem is that the pciehp_power_thread() only looks at the slot->state which would only contain the *latest* state - not the actual event (add / remove) that was the intent of the IRQ handler who submitted the work. What this patch does: => Hotplug events pass on an actual request (for addition or removal) to pciehp_power_thread() which is local to that work item submission. => pciehp_power_thread() does not need to look at slote->state and hence no locks needed in that. => Essentially this results in all the hotplug and unplug events "replayed" by pciehp_power_thread(). Signed-off-by: Rajat Jain <rajatxjain@gmail.com> Signed-off-by: Rajat Jain <rajatjain@juniper.net> Signed-off-by: Guenter Roeck <groeck@juniper.net> Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
2014-02-05 10:30:56 +08:00
info->req = event == INT_LINK_UP ? ENABLE_REQ : DISABLE_REQ;
INIT_WORK(&info->work, pciehp_power_thread);
switch (p_slot->state) {
case BLINKINGON_STATE:
case BLINKINGOFF_STATE:
cancel_delayed_work(&p_slot->work);
/* Fall through */
case STATIC_STATE:
p_slot->state = event == INT_LINK_UP ?
POWERON_STATE : POWEROFF_STATE;
queue_work(p_slot->wq, &info->work);
break;
case POWERON_STATE:
if (event == INT_LINK_UP) {
ctrl_info(ctrl,
"Link Up event ignored on slot(%s): already powering on\n",
slot_name(p_slot));
kfree(info);
} else {
ctrl_info(ctrl,
"Link Down event queued on slot(%s): currently getting powered on\n",
slot_name(p_slot));
p_slot->state = POWEROFF_STATE;
queue_work(p_slot->wq, &info->work);
}
break;
case POWEROFF_STATE:
if (event == INT_LINK_UP) {
ctrl_info(ctrl,
"Link Up event queued on slot(%s): currently getting powered off\n",
slot_name(p_slot));
p_slot->state = POWERON_STATE;
queue_work(p_slot->wq, &info->work);
} else {
ctrl_info(ctrl,
"Link Down event ignored on slot(%s): already powering off\n",
slot_name(p_slot));
kfree(info);
}
break;
default:
ctrl_err(ctrl, "ignoring invalid state %#x on slot(%s)\n",
p_slot->state, slot_name(p_slot));
kfree(info);
break;
}
}
static void interrupt_event_handler(struct work_struct *work)
{
struct event_info *info = container_of(work, struct event_info, work);
struct slot *p_slot = info->p_slot;
struct controller *ctrl = p_slot->ctrl;
mutex_lock(&p_slot->lock);
switch (info->event_type) {
case INT_BUTTON_PRESS:
handle_button_press_event(p_slot);
break;
case INT_POWER_FAULT:
if (!POWER_CTRL(ctrl))
break;
pciehp_set_attention_status(p_slot, 1);
pciehp_green_led_off(p_slot);
break;
case INT_PRESENCE_ON:
handle_surprise_event(p_slot);
break;
case INT_PRESENCE_OFF:
/*
* Regardless of surprise capability, we need to
* definitely remove a card that has been pulled out!
*/
handle_surprise_event(p_slot);
break;
case INT_LINK_UP:
case INT_LINK_DOWN:
handle_link_event(p_slot, info->event_type);
break;
default:
break;
}
mutex_unlock(&p_slot->lock);
kfree(info);
}
/*
* Note: This function must be called with slot->hotplug_lock held
*/
int pciehp_enable_slot(struct slot *p_slot)
{
u8 getstatus = 0;
int rc;
struct controller *ctrl = p_slot->ctrl;
pciehp_get_adapter_status(p_slot, &getstatus);
if (!getstatus) {
ctrl_info(ctrl, "No adapter on slot(%s)\n", slot_name(p_slot));
return -ENODEV;
}
if (MRL_SENS(p_slot->ctrl)) {
pciehp_get_latch_status(p_slot, &getstatus);
if (getstatus) {
ctrl_info(ctrl, "Latch open on slot(%s)\n",
slot_name(p_slot));
return -ENODEV;
}
}
if (POWER_CTRL(p_slot->ctrl)) {
pciehp_get_power_status(p_slot, &getstatus);
if (getstatus) {
ctrl_info(ctrl, "Already enabled on slot(%s)\n",
slot_name(p_slot));
return -EINVAL;
}
}
pciehp_get_latch_status(p_slot, &getstatus);
rc = board_added(p_slot);
if (rc)
pciehp_get_latch_status(p_slot, &getstatus);
return rc;
}
/*
* Note: This function must be called with slot->hotplug_lock held
*/
int pciehp_disable_slot(struct slot *p_slot)
{
u8 getstatus = 0;
struct controller *ctrl = p_slot->ctrl;
if (!p_slot->ctrl)
return 1;
if (POWER_CTRL(p_slot->ctrl)) {
pciehp_get_power_status(p_slot, &getstatus);
if (!getstatus) {
ctrl_info(ctrl, "Already disabled on slot(%s)\n",
slot_name(p_slot));
return -EINVAL;
}
}
return remove_board(p_slot);
}
int pciehp_sysfs_enable_slot(struct slot *p_slot)
{
int retval = -ENODEV;
struct controller *ctrl = p_slot->ctrl;
mutex_lock(&p_slot->lock);
switch (p_slot->state) {
case BLINKINGON_STATE:
cancel_delayed_work(&p_slot->work);
case STATIC_STATE:
p_slot->state = POWERON_STATE;
mutex_unlock(&p_slot->lock);
mutex_lock(&p_slot->hotplug_lock);
retval = pciehp_enable_slot(p_slot);
mutex_unlock(&p_slot->hotplug_lock);
mutex_lock(&p_slot->lock);
p_slot->state = STATIC_STATE;
break;
case POWERON_STATE:
ctrl_info(ctrl, "Slot %s is already in powering on state\n",
slot_name(p_slot));
break;
case BLINKINGOFF_STATE:
case POWEROFF_STATE:
ctrl_info(ctrl, "Already enabled on slot %s\n",
slot_name(p_slot));
break;
default:
ctrl_err(ctrl, "invalid state %#x on slot %s\n",
p_slot->state, slot_name(p_slot));
break;
}
mutex_unlock(&p_slot->lock);
return retval;
}
int pciehp_sysfs_disable_slot(struct slot *p_slot)
{
int retval = -ENODEV;
struct controller *ctrl = p_slot->ctrl;
mutex_lock(&p_slot->lock);
switch (p_slot->state) {
case BLINKINGOFF_STATE:
cancel_delayed_work(&p_slot->work);
case STATIC_STATE:
p_slot->state = POWEROFF_STATE;
mutex_unlock(&p_slot->lock);
retval = pciehp_disable_slot(p_slot);
mutex_lock(&p_slot->lock);
p_slot->state = STATIC_STATE;
break;
case POWEROFF_STATE:
ctrl_info(ctrl, "Slot %s is already in powering off state\n",
slot_name(p_slot));
break;
case BLINKINGON_STATE:
case POWERON_STATE:
ctrl_info(ctrl, "Already disabled on slot %s\n",
slot_name(p_slot));
break;
default:
ctrl_err(ctrl, "invalid state %#x on slot %s\n",
p_slot->state, slot_name(p_slot));
break;
}
mutex_unlock(&p_slot->lock);
return retval;
}