kernel_optimize_test/drivers/macintosh/adb.c

914 lines
21 KiB
C
Raw Normal View History

/*
* Device driver for the Apple Desktop Bus
* and the /dev/adb device on macintoshes.
*
* Copyright (C) 1996 Paul Mackerras.
*
* Modified to declare controllers as structures, added
* client notification of bus reset and handles PowerBook
* sleep, by Benjamin Herrenschmidt.
*
* To do:
*
* - /sys/bus/adb to list the devices and infos
* - more /dev/adb to allow userland to receive the
* flow of auto-polling datas from a given device.
* - move bus probe to a kernel thread
*/
#include <linux/config.h>
#include <linux/types.h>
#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/slab.h>
#include <linux/module.h>
#include <linux/fs.h>
#include <linux/mm.h>
#include <linux/sched.h>
#include <linux/smp_lock.h>
#include <linux/adb.h>
#include <linux/cuda.h>
#include <linux/pmu.h>
#include <linux/notifier.h>
#include <linux/wait.h>
#include <linux/init.h>
#include <linux/delay.h>
#include <linux/spinlock.h>
#include <linux/completion.h>
#include <linux/device.h>
#include <linux/devfs_fs_kernel.h>
#include <asm/uaccess.h>
#include <asm/semaphore.h>
#ifdef CONFIG_PPC
#include <asm/prom.h>
#include <asm/machdep.h>
#endif
EXPORT_SYMBOL(adb_controller);
EXPORT_SYMBOL(adb_client_list);
extern struct adb_driver via_macii_driver;
extern struct adb_driver via_maciisi_driver;
extern struct adb_driver via_cuda_driver;
extern struct adb_driver adb_iop_driver;
extern struct adb_driver via_pmu_driver;
extern struct adb_driver macio_adb_driver;
static struct adb_driver *adb_driver_list[] = {
#ifdef CONFIG_ADB_MACII
&via_macii_driver,
#endif
#ifdef CONFIG_ADB_MACIISI
&via_maciisi_driver,
#endif
#ifdef CONFIG_ADB_CUDA
&via_cuda_driver,
#endif
#ifdef CONFIG_ADB_IOP
&adb_iop_driver,
#endif
#if defined(CONFIG_ADB_PMU) || defined(CONFIG_ADB_PMU68K)
&via_pmu_driver,
#endif
#ifdef CONFIG_ADB_MACIO
&macio_adb_driver,
#endif
NULL
};
static struct class *adb_dev_class;
struct adb_driver *adb_controller;
[PATCH] Notifier chain update: API changes The kernel's implementation of notifier chains is unsafe. There is no protection against entries being added to or removed from a chain while the chain is in use. The issues were discussed in this thread: http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2 We noticed that notifier chains in the kernel fall into two basic usage classes: "Blocking" chains are always called from a process context and the callout routines are allowed to sleep; "Atomic" chains can be called from an atomic context and the callout routines are not allowed to sleep. We decided to codify this distinction and make it part of the API. Therefore this set of patches introduces three new, parallel APIs: one for blocking notifiers, one for atomic notifiers, and one for "raw" notifiers (which is really just the old API under a new name). New kinds of data structures are used for the heads of the chains, and new routines are defined for registration, unregistration, and calling a chain. The three APIs are explained in include/linux/notifier.h and their implementation is in kernel/sys.c. With atomic and blocking chains, the implementation guarantees that the chain links will not be corrupted and that chain callers will not get messed up by entries being added or removed. For raw chains the implementation provides no guarantees at all; users of this API must provide their own protections. (The idea was that situations may come up where the assumptions of the atomic and blocking APIs are not appropriate, so it should be possible for users to handle these things in their own way.) There are some limitations, which should not be too hard to live with. For atomic/blocking chains, registration and unregistration must always be done in a process context since the chain is protected by a mutex/rwsem. Also, a callout routine for a non-raw chain must not try to register or unregister entries on its own chain. (This did happen in a couple of places and the code had to be changed to avoid it.) Since atomic chains may be called from within an NMI handler, they cannot use spinlocks for synchronization. Instead we use RCU. The overhead falls almost entirely in the unregister routine, which is okay since unregistration is much less frequent that calling a chain. Here is the list of chains that we adjusted and their classifications. None of them use the raw API, so for the moment it is only a placeholder. ATOMIC CHAINS ------------- arch/i386/kernel/traps.c: i386die_chain arch/ia64/kernel/traps.c: ia64die_chain arch/powerpc/kernel/traps.c: powerpc_die_chain arch/sparc64/kernel/traps.c: sparc64die_chain arch/x86_64/kernel/traps.c: die_chain drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list kernel/panic.c: panic_notifier_list kernel/profile.c: task_free_notifier net/bluetooth/hci_core.c: hci_notifier net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain net/ipv6/addrconf.c: inet6addr_chain net/netfilter/nf_conntrack_core.c: nf_conntrack_chain net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain net/netlink/af_netlink.c: netlink_chain BLOCKING CHAINS --------------- arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain arch/s390/kernel/process.c: idle_chain arch/x86_64/kernel/process.c idle_notifier drivers/base/memory.c: memory_chain drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list drivers/macintosh/adb.c: adb_client_list drivers/macintosh/via-pmu.c sleep_notifier_list drivers/macintosh/via-pmu68k.c sleep_notifier_list drivers/macintosh/windfarm_core.c wf_client_list drivers/usb/core/notify.c usb_notifier_list drivers/video/fbmem.c fb_notifier_list kernel/cpu.c cpu_chain kernel/module.c module_notify_list kernel/profile.c munmap_notifier kernel/profile.c task_exit_notifier kernel/sys.c reboot_notifier_list net/core/dev.c netdev_chain net/decnet/dn_dev.c: dnaddr_chain net/ipv4/devinet.c: inetaddr_chain It's possible that some of these classifications are wrong. If they are, please let us know or submit a patch to fix them. Note that any chain that gets called very frequently should be atomic, because the rwsem read-locking used for blocking chains is very likely to incur cache misses on SMP systems. (However, if the chain's callout routines may sleep then the chain cannot be atomic.) The patch set was written by Alan Stern and Chandra Seetharaman, incorporating material written by Keith Owens and suggestions from Paul McKenney and Andrew Morton. [jes@sgi.com: restructure the notifier chain initialization macros] Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com> Signed-off-by: Jes Sorensen <jes@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
BLOCKING_NOTIFIER_HEAD(adb_client_list);
static int adb_got_sleep;
static int adb_inited;
static pid_t adb_probe_task_pid;
static DECLARE_MUTEX(adb_probe_mutex);
static struct completion adb_probe_task_comp;
static int sleepy_trackpad;
static int autopoll_devs;
int __adb_probe_sync;
#ifdef CONFIG_PM
static int adb_notify_sleep(struct pmu_sleep_notifier *self, int when);
static struct pmu_sleep_notifier adb_sleep_notifier = {
adb_notify_sleep,
SLEEP_LEVEL_ADB,
};
#endif
static int adb_scan_bus(void);
static int do_adb_reset_bus(void);
static void adbdev_init(void);
static int try_handler_change(int, int);
static struct adb_handler {
void (*handler)(unsigned char *, int, struct pt_regs *, int);
int original_address;
int handler_id;
int busy;
} adb_handler[16];
/*
* The adb_handler_sem mutex protects all accesses to the original_address
* and handler_id fields of adb_handler[i] for all i, and changes to the
* handler field.
* Accesses to the handler field are protected by the adb_handler_lock
* rwlock. It is held across all calls to any handler, so that by the
* time adb_unregister returns, we know that the old handler isn't being
* called.
*/
static DECLARE_MUTEX(adb_handler_sem);
static DEFINE_RWLOCK(adb_handler_lock);
#if 0
static void printADBreply(struct adb_request *req)
{
int i;
printk("adb reply (%d)", req->reply_len);
for(i = 0; i < req->reply_len; i++)
printk(" %x", req->reply[i]);
printk("\n");
}
#endif
static __inline__ void adb_wait_ms(unsigned int ms)
{
if (current->pid && adb_probe_task_pid &&
adb_probe_task_pid == current->pid)
msleep(ms);
else
mdelay(ms);
}
static int adb_scan_bus(void)
{
int i, highFree=0, noMovement;
int devmask = 0;
struct adb_request req;
/* assumes adb_handler[] is all zeroes at this point */
for (i = 1; i < 16; i++) {
/* see if there is anything at address i */
adb_request(&req, NULL, ADBREQ_SYNC | ADBREQ_REPLY, 1,
(i << 4) | 0xf);
if (req.reply_len > 1)
/* one or more devices at this address */
adb_handler[i].original_address = i;
else if (i > highFree)
highFree = i;
}
/* Note we reset noMovement to 0 each time we move a device */
for (noMovement = 1; noMovement < 2 && highFree > 0; noMovement++) {
for (i = 1; i < 16; i++) {
if (adb_handler[i].original_address == 0)
continue;
/*
* Send a "talk register 3" command to address i
* to provoke a collision if there is more than
* one device at this address.
*/
adb_request(&req, NULL, ADBREQ_SYNC | ADBREQ_REPLY, 1,
(i << 4) | 0xf);
/*
* Move the device(s) which didn't detect a
* collision to address `highFree'. Hopefully
* this only moves one device.
*/
adb_request(&req, NULL, ADBREQ_SYNC, 3,
(i<< 4) | 0xb, (highFree | 0x60), 0xfe);
/*
* See if anybody actually moved. This is suggested
* by HW TechNote 01:
*
* http://developer.apple.com/technotes/hw/hw_01.html
*/
adb_request(&req, NULL, ADBREQ_SYNC | ADBREQ_REPLY, 1,
(highFree << 4) | 0xf);
if (req.reply_len <= 1) continue;
/*
* Test whether there are any device(s) left
* at address i.
*/
adb_request(&req, NULL, ADBREQ_SYNC | ADBREQ_REPLY, 1,
(i << 4) | 0xf);
if (req.reply_len > 1) {
/*
* There are still one or more devices
* left at address i. Register the one(s)
* we moved to `highFree', and find a new
* value for highFree.
*/
adb_handler[highFree].original_address =
adb_handler[i].original_address;
while (highFree > 0 &&
adb_handler[highFree].original_address)
highFree--;
if (highFree <= 0)
break;
noMovement = 0;
}
else {
/*
* No devices left at address i; move the
* one(s) we moved to `highFree' back to i.
*/
adb_request(&req, NULL, ADBREQ_SYNC, 3,
(highFree << 4) | 0xb,
(i | 0x60), 0xfe);
}
}
}
/* Now fill in the handler_id field of the adb_handler entries. */
printk(KERN_DEBUG "adb devices:");
for (i = 1; i < 16; i++) {
if (adb_handler[i].original_address == 0)
continue;
adb_request(&req, NULL, ADBREQ_SYNC | ADBREQ_REPLY, 1,
(i << 4) | 0xf);
adb_handler[i].handler_id = req.reply[2];
printk(" [%d]: %d %x", i, adb_handler[i].original_address,
adb_handler[i].handler_id);
devmask |= 1 << i;
}
printk("\n");
return devmask;
}
/*
* This kernel task handles ADB probing. It dies once probing is
* completed.
*/
static int
adb_probe_task(void *x)
{
sigset_t blocked;
strcpy(current->comm, "kadbprobe");
sigfillset(&blocked);
sigprocmask(SIG_BLOCK, &blocked, NULL);
flush_signals(current);
printk(KERN_INFO "adb: starting probe task...\n");
do_adb_reset_bus();
printk(KERN_INFO "adb: finished probe task...\n");
adb_probe_task_pid = 0;
up(&adb_probe_mutex);
return 0;
}
static void
__adb_probe_task(void *data)
{
adb_probe_task_pid = kernel_thread(adb_probe_task, NULL, SIGCHLD | CLONE_KERNEL);
}
static DECLARE_WORK(adb_reset_work, __adb_probe_task, NULL);
int
adb_reset_bus(void)
{
if (__adb_probe_sync) {
do_adb_reset_bus();
return 0;
}
down(&adb_probe_mutex);
schedule_work(&adb_reset_work);
return 0;
}
int __init adb_init(void)
{
struct adb_driver *driver;
int i;
#ifdef CONFIG_PPC32
if (!machine_is(chrp) && !machine_is(powermac))
return 0;
#endif
#ifdef CONFIG_MAC
if (!MACH_IS_MAC)
return 0;
#endif
/* xmon may do early-init */
if (adb_inited)
return 0;
adb_inited = 1;
adb_controller = NULL;
i = 0;
while ((driver = adb_driver_list[i++]) != NULL) {
if (!driver->probe()) {
adb_controller = driver;
break;
}
}
if ((adb_controller == NULL) || adb_controller->init()) {
printk(KERN_WARNING "Warning: no ADB interface detected\n");
adb_controller = NULL;
} else {
#ifdef CONFIG_PM
pmu_register_sleep_notifier(&adb_sleep_notifier);
#endif /* CONFIG_PM */
#ifdef CONFIG_PPC
if (machine_is_compatible("AAPL,PowerBook1998") ||
machine_is_compatible("PowerBook1,1"))
sleepy_trackpad = 1;
#endif /* CONFIG_PPC */
init_completion(&adb_probe_task_comp);
adbdev_init();
adb_reset_bus();
}
return 0;
}
__initcall(adb_init);
#ifdef CONFIG_PM
/*
* notify clients before sleep and reset bus afterwards
*/
int
adb_notify_sleep(struct pmu_sleep_notifier *self, int when)
{
int ret;
switch (when) {
case PBOOK_SLEEP_REQUEST:
adb_got_sleep = 1;
/* We need to get a lock on the probe thread */
down(&adb_probe_mutex);
/* Stop autopoll */
if (adb_controller->autopoll)
adb_controller->autopoll(0);
[PATCH] Notifier chain update: API changes The kernel's implementation of notifier chains is unsafe. There is no protection against entries being added to or removed from a chain while the chain is in use. The issues were discussed in this thread: http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2 We noticed that notifier chains in the kernel fall into two basic usage classes: "Blocking" chains are always called from a process context and the callout routines are allowed to sleep; "Atomic" chains can be called from an atomic context and the callout routines are not allowed to sleep. We decided to codify this distinction and make it part of the API. Therefore this set of patches introduces three new, parallel APIs: one for blocking notifiers, one for atomic notifiers, and one for "raw" notifiers (which is really just the old API under a new name). New kinds of data structures are used for the heads of the chains, and new routines are defined for registration, unregistration, and calling a chain. The three APIs are explained in include/linux/notifier.h and their implementation is in kernel/sys.c. With atomic and blocking chains, the implementation guarantees that the chain links will not be corrupted and that chain callers will not get messed up by entries being added or removed. For raw chains the implementation provides no guarantees at all; users of this API must provide their own protections. (The idea was that situations may come up where the assumptions of the atomic and blocking APIs are not appropriate, so it should be possible for users to handle these things in their own way.) There are some limitations, which should not be too hard to live with. For atomic/blocking chains, registration and unregistration must always be done in a process context since the chain is protected by a mutex/rwsem. Also, a callout routine for a non-raw chain must not try to register or unregister entries on its own chain. (This did happen in a couple of places and the code had to be changed to avoid it.) Since atomic chains may be called from within an NMI handler, they cannot use spinlocks for synchronization. Instead we use RCU. The overhead falls almost entirely in the unregister routine, which is okay since unregistration is much less frequent that calling a chain. Here is the list of chains that we adjusted and their classifications. None of them use the raw API, so for the moment it is only a placeholder. ATOMIC CHAINS ------------- arch/i386/kernel/traps.c: i386die_chain arch/ia64/kernel/traps.c: ia64die_chain arch/powerpc/kernel/traps.c: powerpc_die_chain arch/sparc64/kernel/traps.c: sparc64die_chain arch/x86_64/kernel/traps.c: die_chain drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list kernel/panic.c: panic_notifier_list kernel/profile.c: task_free_notifier net/bluetooth/hci_core.c: hci_notifier net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain net/ipv6/addrconf.c: inet6addr_chain net/netfilter/nf_conntrack_core.c: nf_conntrack_chain net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain net/netlink/af_netlink.c: netlink_chain BLOCKING CHAINS --------------- arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain arch/s390/kernel/process.c: idle_chain arch/x86_64/kernel/process.c idle_notifier drivers/base/memory.c: memory_chain drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list drivers/macintosh/adb.c: adb_client_list drivers/macintosh/via-pmu.c sleep_notifier_list drivers/macintosh/via-pmu68k.c sleep_notifier_list drivers/macintosh/windfarm_core.c wf_client_list drivers/usb/core/notify.c usb_notifier_list drivers/video/fbmem.c fb_notifier_list kernel/cpu.c cpu_chain kernel/module.c module_notify_list kernel/profile.c munmap_notifier kernel/profile.c task_exit_notifier kernel/sys.c reboot_notifier_list net/core/dev.c netdev_chain net/decnet/dn_dev.c: dnaddr_chain net/ipv4/devinet.c: inetaddr_chain It's possible that some of these classifications are wrong. If they are, please let us know or submit a patch to fix them. Note that any chain that gets called very frequently should be atomic, because the rwsem read-locking used for blocking chains is very likely to incur cache misses on SMP systems. (However, if the chain's callout routines may sleep then the chain cannot be atomic.) The patch set was written by Alan Stern and Chandra Seetharaman, incorporating material written by Keith Owens and suggestions from Paul McKenney and Andrew Morton. [jes@sgi.com: restructure the notifier chain initialization macros] Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com> Signed-off-by: Jes Sorensen <jes@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
ret = blocking_notifier_call_chain(&adb_client_list,
ADB_MSG_POWERDOWN, NULL);
if (ret & NOTIFY_STOP_MASK) {
up(&adb_probe_mutex);
return PBOOK_SLEEP_REFUSE;
}
break;
case PBOOK_SLEEP_REJECT:
if (adb_got_sleep) {
adb_got_sleep = 0;
up(&adb_probe_mutex);
adb_reset_bus();
}
break;
case PBOOK_SLEEP_NOW:
break;
case PBOOK_WAKE:
adb_got_sleep = 0;
up(&adb_probe_mutex);
adb_reset_bus();
break;
}
return PBOOK_SLEEP_OK;
}
#endif /* CONFIG_PM */
static int
do_adb_reset_bus(void)
{
int ret, nret;
if (adb_controller == NULL)
return -ENXIO;
if (adb_controller->autopoll)
adb_controller->autopoll(0);
[PATCH] Notifier chain update: API changes The kernel's implementation of notifier chains is unsafe. There is no protection against entries being added to or removed from a chain while the chain is in use. The issues were discussed in this thread: http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2 We noticed that notifier chains in the kernel fall into two basic usage classes: "Blocking" chains are always called from a process context and the callout routines are allowed to sleep; "Atomic" chains can be called from an atomic context and the callout routines are not allowed to sleep. We decided to codify this distinction and make it part of the API. Therefore this set of patches introduces three new, parallel APIs: one for blocking notifiers, one for atomic notifiers, and one for "raw" notifiers (which is really just the old API under a new name). New kinds of data structures are used for the heads of the chains, and new routines are defined for registration, unregistration, and calling a chain. The three APIs are explained in include/linux/notifier.h and their implementation is in kernel/sys.c. With atomic and blocking chains, the implementation guarantees that the chain links will not be corrupted and that chain callers will not get messed up by entries being added or removed. For raw chains the implementation provides no guarantees at all; users of this API must provide their own protections. (The idea was that situations may come up where the assumptions of the atomic and blocking APIs are not appropriate, so it should be possible for users to handle these things in their own way.) There are some limitations, which should not be too hard to live with. For atomic/blocking chains, registration and unregistration must always be done in a process context since the chain is protected by a mutex/rwsem. Also, a callout routine for a non-raw chain must not try to register or unregister entries on its own chain. (This did happen in a couple of places and the code had to be changed to avoid it.) Since atomic chains may be called from within an NMI handler, they cannot use spinlocks for synchronization. Instead we use RCU. The overhead falls almost entirely in the unregister routine, which is okay since unregistration is much less frequent that calling a chain. Here is the list of chains that we adjusted and their classifications. None of them use the raw API, so for the moment it is only a placeholder. ATOMIC CHAINS ------------- arch/i386/kernel/traps.c: i386die_chain arch/ia64/kernel/traps.c: ia64die_chain arch/powerpc/kernel/traps.c: powerpc_die_chain arch/sparc64/kernel/traps.c: sparc64die_chain arch/x86_64/kernel/traps.c: die_chain drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list kernel/panic.c: panic_notifier_list kernel/profile.c: task_free_notifier net/bluetooth/hci_core.c: hci_notifier net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain net/ipv6/addrconf.c: inet6addr_chain net/netfilter/nf_conntrack_core.c: nf_conntrack_chain net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain net/netlink/af_netlink.c: netlink_chain BLOCKING CHAINS --------------- arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain arch/s390/kernel/process.c: idle_chain arch/x86_64/kernel/process.c idle_notifier drivers/base/memory.c: memory_chain drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list drivers/macintosh/adb.c: adb_client_list drivers/macintosh/via-pmu.c sleep_notifier_list drivers/macintosh/via-pmu68k.c sleep_notifier_list drivers/macintosh/windfarm_core.c wf_client_list drivers/usb/core/notify.c usb_notifier_list drivers/video/fbmem.c fb_notifier_list kernel/cpu.c cpu_chain kernel/module.c module_notify_list kernel/profile.c munmap_notifier kernel/profile.c task_exit_notifier kernel/sys.c reboot_notifier_list net/core/dev.c netdev_chain net/decnet/dn_dev.c: dnaddr_chain net/ipv4/devinet.c: inetaddr_chain It's possible that some of these classifications are wrong. If they are, please let us know or submit a patch to fix them. Note that any chain that gets called very frequently should be atomic, because the rwsem read-locking used for blocking chains is very likely to incur cache misses on SMP systems. (However, if the chain's callout routines may sleep then the chain cannot be atomic.) The patch set was written by Alan Stern and Chandra Seetharaman, incorporating material written by Keith Owens and suggestions from Paul McKenney and Andrew Morton. [jes@sgi.com: restructure the notifier chain initialization macros] Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com> Signed-off-by: Jes Sorensen <jes@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
nret = blocking_notifier_call_chain(&adb_client_list,
ADB_MSG_PRE_RESET, NULL);
if (nret & NOTIFY_STOP_MASK) {
if (adb_controller->autopoll)
adb_controller->autopoll(autopoll_devs);
return -EBUSY;
}
if (sleepy_trackpad) {
/* Let the trackpad settle down */
adb_wait_ms(500);
}
down(&adb_handler_sem);
write_lock_irq(&adb_handler_lock);
memset(adb_handler, 0, sizeof(adb_handler));
write_unlock_irq(&adb_handler_lock);
/* That one is still a bit synchronous, oh well... */
if (adb_controller->reset_bus)
ret = adb_controller->reset_bus();
else
ret = 0;
if (sleepy_trackpad) {
/* Let the trackpad settle down */
adb_wait_ms(1500);
}
if (!ret) {
autopoll_devs = adb_scan_bus();
if (adb_controller->autopoll)
adb_controller->autopoll(autopoll_devs);
}
up(&adb_handler_sem);
[PATCH] Notifier chain update: API changes The kernel's implementation of notifier chains is unsafe. There is no protection against entries being added to or removed from a chain while the chain is in use. The issues were discussed in this thread: http://marc.theaimsgroup.com/?l=linux-kernel&m=113018709002036&w=2 We noticed that notifier chains in the kernel fall into two basic usage classes: "Blocking" chains are always called from a process context and the callout routines are allowed to sleep; "Atomic" chains can be called from an atomic context and the callout routines are not allowed to sleep. We decided to codify this distinction and make it part of the API. Therefore this set of patches introduces three new, parallel APIs: one for blocking notifiers, one for atomic notifiers, and one for "raw" notifiers (which is really just the old API under a new name). New kinds of data structures are used for the heads of the chains, and new routines are defined for registration, unregistration, and calling a chain. The three APIs are explained in include/linux/notifier.h and their implementation is in kernel/sys.c. With atomic and blocking chains, the implementation guarantees that the chain links will not be corrupted and that chain callers will not get messed up by entries being added or removed. For raw chains the implementation provides no guarantees at all; users of this API must provide their own protections. (The idea was that situations may come up where the assumptions of the atomic and blocking APIs are not appropriate, so it should be possible for users to handle these things in their own way.) There are some limitations, which should not be too hard to live with. For atomic/blocking chains, registration and unregistration must always be done in a process context since the chain is protected by a mutex/rwsem. Also, a callout routine for a non-raw chain must not try to register or unregister entries on its own chain. (This did happen in a couple of places and the code had to be changed to avoid it.) Since atomic chains may be called from within an NMI handler, they cannot use spinlocks for synchronization. Instead we use RCU. The overhead falls almost entirely in the unregister routine, which is okay since unregistration is much less frequent that calling a chain. Here is the list of chains that we adjusted and their classifications. None of them use the raw API, so for the moment it is only a placeholder. ATOMIC CHAINS ------------- arch/i386/kernel/traps.c: i386die_chain arch/ia64/kernel/traps.c: ia64die_chain arch/powerpc/kernel/traps.c: powerpc_die_chain arch/sparc64/kernel/traps.c: sparc64die_chain arch/x86_64/kernel/traps.c: die_chain drivers/char/ipmi/ipmi_si_intf.c: xaction_notifier_list kernel/panic.c: panic_notifier_list kernel/profile.c: task_free_notifier net/bluetooth/hci_core.c: hci_notifier net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_chain net/ipv4/netfilter/ip_conntrack_core.c: ip_conntrack_expect_chain net/ipv6/addrconf.c: inet6addr_chain net/netfilter/nf_conntrack_core.c: nf_conntrack_chain net/netfilter/nf_conntrack_core.c: nf_conntrack_expect_chain net/netlink/af_netlink.c: netlink_chain BLOCKING CHAINS --------------- arch/powerpc/platforms/pseries/reconfig.c: pSeries_reconfig_chain arch/s390/kernel/process.c: idle_chain arch/x86_64/kernel/process.c idle_notifier drivers/base/memory.c: memory_chain drivers/cpufreq/cpufreq.c cpufreq_policy_notifier_list drivers/cpufreq/cpufreq.c cpufreq_transition_notifier_list drivers/macintosh/adb.c: adb_client_list drivers/macintosh/via-pmu.c sleep_notifier_list drivers/macintosh/via-pmu68k.c sleep_notifier_list drivers/macintosh/windfarm_core.c wf_client_list drivers/usb/core/notify.c usb_notifier_list drivers/video/fbmem.c fb_notifier_list kernel/cpu.c cpu_chain kernel/module.c module_notify_list kernel/profile.c munmap_notifier kernel/profile.c task_exit_notifier kernel/sys.c reboot_notifier_list net/core/dev.c netdev_chain net/decnet/dn_dev.c: dnaddr_chain net/ipv4/devinet.c: inetaddr_chain It's possible that some of these classifications are wrong. If they are, please let us know or submit a patch to fix them. Note that any chain that gets called very frequently should be atomic, because the rwsem read-locking used for blocking chains is very likely to incur cache misses on SMP systems. (However, if the chain's callout routines may sleep then the chain cannot be atomic.) The patch set was written by Alan Stern and Chandra Seetharaman, incorporating material written by Keith Owens and suggestions from Paul McKenney and Andrew Morton. [jes@sgi.com: restructure the notifier chain initialization macros] Signed-off-by: Alan Stern <stern@rowland.harvard.edu> Signed-off-by: Chandra Seetharaman <sekharan@us.ibm.com> Signed-off-by: Jes Sorensen <jes@sgi.com> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-03-27 17:16:30 +08:00
nret = blocking_notifier_call_chain(&adb_client_list,
ADB_MSG_POST_RESET, NULL);
if (nret & NOTIFY_STOP_MASK)
return -EBUSY;
return ret;
}
void
adb_poll(void)
{
if ((adb_controller == NULL)||(adb_controller->poll == NULL))
return;
adb_controller->poll();
}
static void
adb_probe_wakeup(struct adb_request *req)
{
complete(&adb_probe_task_comp);
}
/* Static request used during probe */
static struct adb_request adb_sreq;
static unsigned long adb_sreq_lock; // Use semaphore ! */
int
adb_request(struct adb_request *req, void (*done)(struct adb_request *),
int flags, int nbytes, ...)
{
va_list list;
int i, use_sreq;
int rc;
if ((adb_controller == NULL) || (adb_controller->send_request == NULL))
return -ENXIO;
if (nbytes < 1)
return -EINVAL;
if (req == NULL && (flags & ADBREQ_NOSEND))
return -EINVAL;
if (req == NULL) {
if (test_and_set_bit(0,&adb_sreq_lock)) {
printk("adb.c: Warning: contention on static request !\n");
return -EPERM;
}
req = &adb_sreq;
flags |= ADBREQ_SYNC;
use_sreq = 1;
} else
use_sreq = 0;
req->nbytes = nbytes+1;
req->done = done;
req->reply_expected = flags & ADBREQ_REPLY;
req->data[0] = ADB_PACKET;
va_start(list, nbytes);
for (i = 0; i < nbytes; ++i)
req->data[i+1] = va_arg(list, int);
va_end(list);
if (flags & ADBREQ_NOSEND)
return 0;
/* Synchronous requests send from the probe thread cause it to
* block. Beware that the "done" callback will be overriden !
*/
if ((flags & ADBREQ_SYNC) &&
(current->pid && adb_probe_task_pid &&
adb_probe_task_pid == current->pid)) {
req->done = adb_probe_wakeup;
rc = adb_controller->send_request(req, 0);
if (rc || req->complete)
goto bail;
wait_for_completion(&adb_probe_task_comp);
rc = 0;
goto bail;
}
rc = adb_controller->send_request(req, flags & ADBREQ_SYNC);
bail:
if (use_sreq)
clear_bit(0, &adb_sreq_lock);
return rc;
}
/* Ultimately this should return the number of devices with
the given default id.
And it does it now ! Note: changed behaviour: This function
will now register if default_id _and_ handler_id both match
but handler_id can be left to 0 to match with default_id only.
When handler_id is set, this function will try to adjust
the handler_id id it doesn't match. */
int
adb_register(int default_id, int handler_id, struct adb_ids *ids,
void (*handler)(unsigned char *, int, struct pt_regs *, int))
{
int i;
down(&adb_handler_sem);
ids->nids = 0;
for (i = 1; i < 16; i++) {
if ((adb_handler[i].original_address == default_id) &&
(!handler_id || (handler_id == adb_handler[i].handler_id) ||
try_handler_change(i, handler_id))) {
if (adb_handler[i].handler != 0) {
printk(KERN_ERR
"Two handlers for ADB device %d\n",
default_id);
continue;
}
write_lock_irq(&adb_handler_lock);
adb_handler[i].handler = handler;
write_unlock_irq(&adb_handler_lock);
ids->id[ids->nids++] = i;
}
}
up(&adb_handler_sem);
return ids->nids;
}
int
adb_unregister(int index)
{
int ret = -ENODEV;
down(&adb_handler_sem);
write_lock_irq(&adb_handler_lock);
if (adb_handler[index].handler) {
while(adb_handler[index].busy) {
write_unlock_irq(&adb_handler_lock);
yield();
write_lock_irq(&adb_handler_lock);
}
ret = 0;
adb_handler[index].handler = NULL;
}
write_unlock_irq(&adb_handler_lock);
up(&adb_handler_sem);
return ret;
}
void
adb_input(unsigned char *buf, int nb, struct pt_regs *regs, int autopoll)
{
int i, id;
static int dump_adb_input = 0;
unsigned long flags;
void (*handler)(unsigned char *, int, struct pt_regs *, int);
/* We skip keystrokes and mouse moves when the sleep process
* has been started. We stop autopoll, but this is another security
*/
if (adb_got_sleep)
return;
id = buf[0] >> 4;
if (dump_adb_input) {
printk(KERN_INFO "adb packet: ");
for (i = 0; i < nb; ++i)
printk(" %x", buf[i]);
printk(", id = %d\n", id);
}
write_lock_irqsave(&adb_handler_lock, flags);
handler = adb_handler[id].handler;
if (handler != NULL)
adb_handler[id].busy = 1;
write_unlock_irqrestore(&adb_handler_lock, flags);
if (handler != NULL) {
(*handler)(buf, nb, regs, autopoll);
wmb();
adb_handler[id].busy = 0;
}
}
/* Try to change handler to new_id. Will return 1 if successful. */
static int try_handler_change(int address, int new_id)
{
struct adb_request req;
if (adb_handler[address].handler_id == new_id)
return 1;
adb_request(&req, NULL, ADBREQ_SYNC, 3,
ADB_WRITEREG(address, 3), address | 0x20, new_id);
adb_request(&req, NULL, ADBREQ_SYNC | ADBREQ_REPLY, 1,
ADB_READREG(address, 3));
if (req.reply_len < 2)
return 0;
if (req.reply[2] != new_id)
return 0;
adb_handler[address].handler_id = req.reply[2];
return 1;
}
int
adb_try_handler_change(int address, int new_id)
{
int ret;
down(&adb_handler_sem);
ret = try_handler_change(address, new_id);
up(&adb_handler_sem);
return ret;
}
int
adb_get_infos(int address, int *original_address, int *handler_id)
{
down(&adb_handler_sem);
*original_address = adb_handler[address].original_address;
*handler_id = adb_handler[address].handler_id;
up(&adb_handler_sem);
return (*original_address != 0);
}
/*
* /dev/adb device driver.
*/
#define ADB_MAJOR 56 /* major number for /dev/adb */
struct adbdev_state {
spinlock_t lock;
atomic_t n_pending;
struct adb_request *completed;
wait_queue_head_t wait_queue;
int inuse;
};
static void adb_write_done(struct adb_request *req)
{
struct adbdev_state *state = (struct adbdev_state *) req->arg;
unsigned long flags;
if (!req->complete) {
req->reply_len = 0;
req->complete = 1;
}
spin_lock_irqsave(&state->lock, flags);
atomic_dec(&state->n_pending);
if (!state->inuse) {
kfree(req);
if (atomic_read(&state->n_pending) == 0) {
spin_unlock_irqrestore(&state->lock, flags);
kfree(state);
return;
}
} else {
struct adb_request **ap = &state->completed;
while (*ap != NULL)
ap = &(*ap)->next;
req->next = NULL;
*ap = req;
wake_up_interruptible(&state->wait_queue);
}
spin_unlock_irqrestore(&state->lock, flags);
}
static int
do_adb_query(struct adb_request *req)
{
int ret = -EINVAL;
switch(req->data[1])
{
case ADB_QUERY_GETDEVINFO:
if (req->nbytes < 3)
break;
down(&adb_handler_sem);
req->reply[0] = adb_handler[req->data[2]].original_address;
req->reply[1] = adb_handler[req->data[2]].handler_id;
up(&adb_handler_sem);
req->complete = 1;
req->reply_len = 2;
adb_write_done(req);
ret = 0;
break;
}
return ret;
}
static int adb_open(struct inode *inode, struct file *file)
{
struct adbdev_state *state;
if (iminor(inode) > 0 || adb_controller == NULL)
return -ENXIO;
state = kmalloc(sizeof(struct adbdev_state), GFP_KERNEL);
if (state == 0)
return -ENOMEM;
file->private_data = state;
spin_lock_init(&state->lock);
atomic_set(&state->n_pending, 0);
state->completed = NULL;
init_waitqueue_head(&state->wait_queue);
state->inuse = 1;
return 0;
}
static int adb_release(struct inode *inode, struct file *file)
{
struct adbdev_state *state = file->private_data;
unsigned long flags;
lock_kernel();
if (state) {
file->private_data = NULL;
spin_lock_irqsave(&state->lock, flags);
if (atomic_read(&state->n_pending) == 0
&& state->completed == NULL) {
spin_unlock_irqrestore(&state->lock, flags);
kfree(state);
} else {
state->inuse = 0;
spin_unlock_irqrestore(&state->lock, flags);
}
}
unlock_kernel();
return 0;
}
static ssize_t adb_read(struct file *file, char __user *buf,
size_t count, loff_t *ppos)
{
int ret = 0;
struct adbdev_state *state = file->private_data;
struct adb_request *req;
wait_queue_t wait = __WAITQUEUE_INITIALIZER(wait,current);
unsigned long flags;
if (count < 2)
return -EINVAL;
if (count > sizeof(req->reply))
count = sizeof(req->reply);
if (!access_ok(VERIFY_WRITE, buf, count))
return -EFAULT;
req = NULL;
spin_lock_irqsave(&state->lock, flags);
add_wait_queue(&state->wait_queue, &wait);
current->state = TASK_INTERRUPTIBLE;
for (;;) {
req = state->completed;
if (req != NULL)
state->completed = req->next;
else if (atomic_read(&state->n_pending) == 0)
ret = -EIO;
if (req != NULL || ret != 0)
break;
if (file->f_flags & O_NONBLOCK) {
ret = -EAGAIN;
break;
}
if (signal_pending(current)) {
ret = -ERESTARTSYS;
break;
}
spin_unlock_irqrestore(&state->lock, flags);
schedule();
spin_lock_irqsave(&state->lock, flags);
}
current->state = TASK_RUNNING;
remove_wait_queue(&state->wait_queue, &wait);
spin_unlock_irqrestore(&state->lock, flags);
if (ret)
return ret;
ret = req->reply_len;
if (ret > count)
ret = count;
if (ret > 0 && copy_to_user(buf, req->reply, ret))
ret = -EFAULT;
kfree(req);
return ret;
}
static ssize_t adb_write(struct file *file, const char __user *buf,
size_t count, loff_t *ppos)
{
int ret/*, i*/;
struct adbdev_state *state = file->private_data;
struct adb_request *req;
if (count < 2 || count > sizeof(req->data))
return -EINVAL;
if (adb_controller == NULL)
return -ENXIO;
if (!access_ok(VERIFY_READ, buf, count))
return -EFAULT;
req = (struct adb_request *) kmalloc(sizeof(struct adb_request),
GFP_KERNEL);
if (req == NULL)
return -ENOMEM;
req->nbytes = count;
req->done = adb_write_done;
req->arg = (void *) state;
req->complete = 0;
ret = -EFAULT;
if (copy_from_user(req->data, buf, count))
goto out;
atomic_inc(&state->n_pending);
/* If a probe is in progress or we are sleeping, wait for it to complete */
down(&adb_probe_mutex);
/* Queries are special requests sent to the ADB driver itself */
if (req->data[0] == ADB_QUERY) {
if (count > 1)
ret = do_adb_query(req);
else
ret = -EINVAL;
up(&adb_probe_mutex);
}
/* Special case for ADB_BUSRESET request, all others are sent to
the controller */
else if ((req->data[0] == ADB_PACKET)&&(count > 1)
&&(req->data[1] == ADB_BUSRESET)) {
ret = do_adb_reset_bus();
up(&adb_probe_mutex);
atomic_dec(&state->n_pending);
if (ret == 0)
ret = count;
goto out;
} else {
req->reply_expected = ((req->data[1] & 0xc) == 0xc);
if (adb_controller && adb_controller->send_request)
ret = adb_controller->send_request(req, 0);
else
ret = -ENXIO;
up(&adb_probe_mutex);
}
if (ret != 0) {
atomic_dec(&state->n_pending);
goto out;
}
return count;
out:
kfree(req);
return ret;
}
static struct file_operations adb_fops = {
.owner = THIS_MODULE,
.llseek = no_llseek,
.read = adb_read,
.write = adb_write,
.open = adb_open,
.release = adb_release,
};
static void
adbdev_init(void)
{
if (register_chrdev(ADB_MAJOR, "adb", &adb_fops)) {
printk(KERN_ERR "adb: unable to get major %d\n", ADB_MAJOR);
return;
}
devfs_mk_cdev(MKDEV(ADB_MAJOR, 0), S_IFCHR | S_IRUSR | S_IWUSR, "adb");
adb_dev_class = class_create(THIS_MODULE, "adb");
if (IS_ERR(adb_dev_class))
return;
class_device_create(adb_dev_class, NULL, MKDEV(ADB_MAJOR, 0), NULL, "adb");
}