Skip to content

L
linux
Commit 5292e24a authored by Jason Gunthorpe's avatar Jason Gunthorpe
Browse files
Options

mm/mmu_notifiers: Use 'interval_sub' as the variable for mmu_interval_notifier 

The 'interval_sub' is placed on the 'notifier_subscriptions' interval
tree.

This eliminates the poor name 'mni' for this variable.
Signed-off-by: default avatarJason Gunthorpe <jgg@mellanox.com>
parent 1991722a
Hide whitespace changes
Inline Side-by-side
Showing with 104 additions and 90 deletions
Documentation/vm/hmm.rst
View file @ 5292e24a
......@@ -149,14 +149,14 @@ CPU page table into a device page table; HMM helps keep both synchronized. A
device driver that wants to mirror a process address space must start with the
registration of a mmu_interval_notifier::
mni->ops = &driver_ops;
int mmu_interval_notifier_insert(struct mmu_interval_notifier *mni,
unsigned long start, unsigned long length,
struct mm_struct *mm);
int mmu_interval_notifier_insert(struct mmu_interval_notifier *interval_sub,
struct mm_struct *mm, unsigned long start,
unsigned long length,
const struct mmu_interval_notifier_ops *ops);
During the driver_ops->invalidate() callback the device driver must perform
the update action to the range (mark range read only, or fully unmap,
etc.). The device must complete the update before the driver callback returns.
During the ops->invalidate() callback the device driver must perform the
update action to the range (mark range read only, or fully unmap, etc.). The
device must complete the update before the driver callback returns.
When the device driver wants to populate a range of virtual addresses, it can
use::
......@@ -183,7 +183,7 @@ The usage pattern is::
struct hmm_range range;
...
range.notifier = &mni;
range.notifier = &interval_sub;
range.start = ...;
range.end = ...;
range.pfns = ...;
......@@ -191,11 +191,11 @@ The usage pattern is::
range.values = ...;
range.pfn_shift = ...;
if (!mmget_not_zero(mni->notifier.mm))
if (!mmget_not_zero(interval_sub->notifier.mm))
return -EFAULT;
again:
range.notifier_seq = mmu_interval_read_begin(&mni);
range.notifier_seq = mmu_interval_read_begin(&interval_sub);
down_read(&mm->mmap_sem);
ret = hmm_range_fault(&range, HMM_RANGE_SNAPSHOT);
if (ret) {
......
include/linux/mmu_notifier.h
View file @ 5292e24a
......@@ -237,7 +237,7 @@ struct mmu_notifier {
* was required but mmu_notifier_range_blockable(range) is false.
*/
struct mmu_interval_notifier_ops {
bool (*invalidate)(struct mmu_interval_notifier *mni,
bool (*invalidate)(struct mmu_interval_notifier *interval_sub,
const struct mmu_notifier_range *range,
unsigned long cur_seq);
};
......@@ -292,20 +292,21 @@ extern int __mmu_notifier_register(struct mmu_notifier *subscription,
extern void mmu_notifier_unregister(struct mmu_notifier *subscription,
struct mm_struct *mm);
unsigned long mmu_interval_read_begin(struct mmu_interval_notifier *mni);
int mmu_interval_notifier_insert(struct mmu_interval_notifier *mni,
unsigned long
mmu_interval_read_begin(struct mmu_interval_notifier *interval_sub);
int mmu_interval_notifier_insert(struct mmu_interval_notifier *interval_sub,
struct mm_struct *mm, unsigned long start,
unsigned long length,
const struct mmu_interval_notifier_ops *ops);
int mmu_interval_notifier_insert_locked(
struct mmu_interval_notifier *mni, struct mm_struct *mm,
struct mmu_interval_notifier *interval_sub, struct mm_struct *mm,
unsigned long start, unsigned long length,
const struct mmu_interval_notifier_ops *ops);
void mmu_interval_notifier_remove(struct mmu_interval_notifier *mni);
void mmu_interval_notifier_remove(struct mmu_interval_notifier *interval_sub);
/**
* mmu_interval_set_seq - Save the invalidation sequence
* @mni - The mni passed to invalidate
* @interval_sub - The subscription passed to invalidate
* @cur_seq - The cur_seq passed to the invalidate() callback
*
* This must be called unconditionally from the invalidate callback of a
......@@ -316,15 +317,16 @@ void mmu_interval_notifier_remove(struct mmu_interval_notifier *mni);
* If the caller does not call mmu_interval_read_begin() or
* mmu_interval_read_retry() then this call is not required.
*/
static inline void mmu_interval_set_seq(struct mmu_interval_notifier *mni,
unsigned long cur_seq)
static inline void
mmu_interval_set_seq(struct mmu_interval_notifier *interval_sub,
unsigned long cur_seq)
{
WRITE_ONCE(mni->invalidate_seq, cur_seq);
WRITE_ONCE(interval_sub->invalidate_seq, cur_seq);
}
/**
* mmu_interval_read_retry - End a read side critical section against a VA range
* mni: The range
* interval_sub: The subscription
* seq: The return of the paired mmu_interval_read_begin()
*
* This MUST be called under a user provided lock that is also held
......@@ -336,15 +338,16 @@ static inline void mmu_interval_set_seq(struct mmu_interval_notifier *mni,
* Returns true if an invalidation collided with this critical section, and
* the caller should retry.
*/
static inline bool mmu_interval_read_retry(struct mmu_interval_notifier *mni,
unsigned long seq)
static inline bool
mmu_interval_read_retry(struct mmu_interval_notifier *interval_sub,
unsigned long seq)
{
return mni->invalidate_seq != seq;
return interval_sub->invalidate_seq != seq;
}
/**
* mmu_interval_check_retry - Test if a collision has occurred
* mni: The range
* interval_sub: The subscription
* seq: The return of the matching mmu_interval_read_begin()
*
* This can be used in the critical section between mmu_interval_read_begin()
......@@ -359,11 +362,12 @@ static inline bool mmu_interval_read_retry(struct mmu_interval_notifier *mni,
* This call can be used as part of loops and other expensive operations to
* expedite a retry.
*/
static inline bool mmu_interval_check_retry(struct mmu_interval_notifier *mni,
unsigned long seq)
static inline bool
mmu_interval_check_retry(struct mmu_interval_notifier *interval_sub,
unsigned long seq)
{
/* Pairs with the WRITE_ONCE in mmu_interval_set_seq() */
return READ_ONCE(mni->invalidate_seq) != seq;
return READ_ONCE(interval_sub->invalidate_seq) != seq;
}
extern void __mmu_notifier_subscriptions_destroy(struct mm_struct *mm);
......
mm/mmu_notifier.c
View file @ 5292e24a
......@@ -81,7 +81,7 @@ struct mmu_notifier_subscriptions {
* seq & 1 # True if a writer exists
*
* The later state avoids some expensive work on inv_end in the common case of
* no mni monitoring the VA.
* no mmu_interval_notifier monitoring the VA.
*/
static bool
mn_itree_is_invalidating(struct mmu_notifier_subscriptions *subscriptions)
......@@ -114,13 +114,13 @@ mn_itree_inv_start_range(struct mmu_notifier_subscriptions *subscriptions,
}
static struct mmu_interval_notifier *
mn_itree_inv_next(struct mmu_interval_notifier *mni,
mn_itree_inv_next(struct mmu_interval_notifier *interval_sub,
const struct mmu_notifier_range *range)
{
struct interval_tree_node *node;
node = interval_tree_iter_next(&mni->interval_tree, range->start,
range->end - 1);
node = interval_tree_iter_next(&interval_sub->interval_tree,
range->start, range->end - 1);
if (!node)
return NULL;
return container_of(node, struct mmu_interval_notifier, interval_tree);
......@@ -128,7 +128,7 @@ mn_itree_inv_next(struct mmu_interval_notifier *mni,
static void mn_itree_inv_end(struct mmu_notifier_subscriptions *subscriptions)
{
struct mmu_interval_notifier *mni;
struct mmu_interval_notifier *interval_sub;
struct hlist_node *next;
spin_lock(&subscriptions->lock);
......@@ -147,15 +147,16 @@ static void mn_itree_inv_end(struct mmu_notifier_subscriptions *subscriptions)
* they are progressed. This arrangement for tree updates is used to
* avoid using a blocking lock during invalidate_range_start.
*/
hlist_for_each_entry_safe(mni, next, &subscriptions->deferred_list,
hlist_for_each_entry_safe(interval_sub, next,
&subscriptions->deferred_list,
deferred_item) {
if (RB_EMPTY_NODE(&mni->interval_tree.rb))
interval_tree_insert(&mni->interval_tree,
if (RB_EMPTY_NODE(&interval_sub->interval_tree.rb))
interval_tree_insert(&interval_sub->interval_tree,
&subscriptions->itree);
else
interval_tree_remove(&mni->interval_tree,
interval_tree_remove(&interval_sub->interval_tree,
&subscriptions->itree);
hlist_del(&mni->deferred_item);
hlist_del(&interval_sub->deferred_item);
}
spin_unlock(&subscriptions->lock);
......@@ -165,12 +166,12 @@ static void mn_itree_inv_end(struct mmu_notifier_subscriptions *subscriptions)
/**
* mmu_interval_read_begin - Begin a read side critical section against a VA
* range
* mni: The range to use
* interval_sub: The interval subscription
*
* mmu_iterval_read_begin()/mmu_iterval_read_retry() implement a
* collision-retry scheme similar to seqcount for the VA range under mni. If
* the mm invokes invalidation during the critical section then
* mmu_interval_read_retry() will return true.
* collision-retry scheme similar to seqcount for the VA range under
* subscription. If the mm invokes invalidation during the critical section
* then mmu_interval_read_retry() will return true.
*
* This is useful to obtain shadow PTEs where teardown or setup of the SPTEs
* require a blocking context. The critical region formed by this can sleep,
......@@ -181,25 +182,26 @@ static void mn_itree_inv_end(struct mmu_notifier_subscriptions *subscriptions)
*
* The return value should be passed to mmu_interval_read_retry().
*/
unsigned long mmu_interval_read_begin(struct mmu_interval_notifier *mni)
unsigned long
mmu_interval_read_begin(struct mmu_interval_notifier *interval_sub)
{
struct mmu_notifier_subscriptions *subscriptions =
mni->mm->notifier_subscriptions;
interval_sub->mm->notifier_subscriptions;
unsigned long seq;
bool is_invalidating;
/*
* If the mni has a different seq value under the user_lock than we
* started with then it has collided.
* If the subscription has a different seq value under the user_lock
* than we started with then it has collided.
*
* If the mni currently has the same seq value as the subscriptions
* seq, then it is currently between invalidate_start/end and is
* colliding.
* If the subscription currently has the same seq value as the
* subscriptions seq, then it is currently between
* invalidate_start/end and is colliding.
*
* The locking looks broadly like this:
* mn_tree_invalidate_start(): mmu_interval_read_begin():
* spin_lock
* seq = READ_ONCE(mni->invalidate_seq);
* seq = READ_ONCE(interval_sub->invalidate_seq);
* seq == subs->invalidate_seq
* spin_unlock
* spin_lock
......@@ -208,7 +210,7 @@ unsigned long mmu_interval_read_begin(struct mmu_interval_notifier *mni)
* op->invalidate_range():
* user_lock
* mmu_interval_set_seq()
* mni->invalidate_seq = seq
* interval_sub->invalidate_seq = seq
* user_unlock
*
* [Required: mmu_interval_read_retry() == true]
......@@ -220,7 +222,7 @@ unsigned long mmu_interval_read_begin(struct mmu_interval_notifier *mni)
*
* user_lock
* mmu_interval_read_retry():
* mni->invalidate_seq != seq
* interval_sub->invalidate_seq != seq
* user_unlock
*
* Barriers are not needed here as any races here are closed by an
......@@ -229,12 +231,12 @@ unsigned long mmu_interval_read_begin(struct mmu_interval_notifier *mni)
*/
spin_lock(&subscriptions->lock);
/* Pairs with the WRITE_ONCE in mmu_interval_set_seq() */
seq = READ_ONCE(mni->invalidate_seq);
seq = READ_ONCE(interval_sub->invalidate_seq);
is_invalidating = seq == subscriptions->invalidate_seq;
spin_unlock(&subscriptions->lock);
/*
* mni->invalidate_seq must always be set to an odd value via
* interval_sub->invalidate_seq must always be set to an odd value via
* mmu_interval_set_seq() using the provided cur_seq from
* mn_itree_inv_start_range(). This ensures that if seq does wrap we
* will always clear the below sleep in some reasonable time as
......@@ -266,13 +268,16 @@ static void mn_itree_release(struct mmu_notifier_subscriptions *subscriptions,
.start = 0,
.end = ULONG_MAX,
};
struct mmu_interval_notifier *mni;
struct mmu_interval_notifier *interval_sub;
unsigned long cur_seq;
bool ret;
for (mni = mn_itree_inv_start_range(subscriptions, &range, &cur_seq);
mni; mni = mn_itree_inv_next(mni, &range)) {
ret = mni->ops->invalidate(mni, &range, cur_seq);
for (interval_sub =
mn_itree_inv_start_range(subscriptions, &range, &cur_seq);
interval_sub;
interval_sub = mn_itree_inv_next(interval_sub, &range)) {
ret = interval_sub->ops->invalidate(interval_sub, &range,
cur_seq);
WARN_ON(!ret);
}
......@@ -434,14 +439,17 @@ void __mmu_notifier_change_pte(struct mm_struct *mm, unsigned long address,
static int mn_itree_invalidate(struct mmu_notifier_subscriptions *subscriptions,
const struct mmu_notifier_range *range)
{
struct mmu_interval_notifier *mni;
struct mmu_interval_notifier *interval_sub;
unsigned long cur_seq;
for (mni = mn_itree_inv_start_range(subscriptions, range, &cur_seq);
mni; mni = mn_itree_inv_next(mni, range)) {
for (interval_sub =
mn_itree_inv_start_range(subscriptions, range, &cur_seq);
interval_sub;
interval_sub = mn_itree_inv_next(interval_sub, range)) {
bool ret;
ret = mni->ops->invalidate(mni, range, cur_seq);
ret = interval_sub->ops->invalidate(interval_sub, range,
cur_seq);
if (!ret) {
if (WARN_ON(mmu_notifier_range_blockable(range)))
continue;
......@@ -878,20 +886,21 @@ void mmu_notifier_put(struct mmu_notifier *subscription)
EXPORT_SYMBOL_GPL(mmu_notifier_put);
static int __mmu_interval_notifier_insert(
struct mmu_interval_notifier *mni, struct mm_struct *mm,
struct mmu_interval_notifier *interval_sub, struct mm_struct *mm,
struct mmu_notifier_subscriptions *subscriptions, unsigned long start,
unsigned long length, const struct mmu_interval_notifier_ops *ops)
{
mni->mm = mm;
mni->ops = ops;
RB_CLEAR_NODE(&mni->interval_tree.rb);
mni->interval_tree.start = start;
interval_sub->mm = mm;
interval_sub->ops = ops;
RB_CLEAR_NODE(&interval_sub->interval_tree.rb);
interval_sub->interval_tree.start = start;
/*
* Note that the representation of the intervals in the interval tree
* considers the ending point as contained in the interval.
*/
if (length == 0 ||
check_add_overflow(start, length - 1, &mni->interval_tree.last))
check_add_overflow(start, length - 1,
&interval_sub->interval_tree.last))
return -EOVERFLOW;
/* Must call with a mmget() held */
......@@ -911,30 +920,31 @@ static int __mmu_interval_notifier_insert(
* possibility for live lock, instead defer the add to
* mn_itree_inv_end() so this algorithm is deterministic.
*
* In all cases the value for the mni->invalidate_seq should be
* In all cases the value for the interval_sub->invalidate_seq should be
* odd, see mmu_interval_read_begin()
*/
spin_lock(&subscriptions->lock);
if (subscriptions->active_invalidate_ranges) {
if (mn_itree_is_invalidating(subscriptions))
hlist_add_head(&mni->deferred_item,
hlist_add_head(&interval_sub->deferred_item,
&subscriptions->deferred_list);
else {
subscriptions->invalidate_seq |= 1;
interval_tree_insert(&mni->interval_tree,
interval_tree_insert(&interval_sub->interval_tree,
&subscriptions->itree);
}
mni->invalidate_seq = subscriptions->invalidate_seq;
interval_sub->invalidate_seq = subscriptions->invalidate_seq;
} else {
WARN_ON(mn_itree_is_invalidating(subscriptions));
/*
* The starting seq for a mni not under invalidation should be
* odd, not equal to the current invalidate_seq and
* The starting seq for a subscription not under invalidation
* should be odd, not equal to the current invalidate_seq and
* invalidate_seq should not 'wrap' to the new seq any time
* soon.
*/
mni->invalidate_seq = subscriptions->invalidate_seq - 1;
interval_tree_insert(&mni->interval_tree,
interval_sub->invalidate_seq =
subscriptions->invalidate_seq - 1;
interval_tree_insert(&interval_sub->interval_tree,
&subscriptions->itree);
}
spin_unlock(&subscriptions->lock);
......@@ -943,7 +953,7 @@ static int __mmu_interval_notifier_insert(
/**
* mmu_interval_notifier_insert - Insert an interval notifier
* @mni: Interval notifier to register
* @interval_sub: Interval subscription to register
* @start: Starting virtual address to monitor
* @length: Length of the range to monitor
* @mm : mm_struct to attach to
......@@ -956,7 +966,7 @@ static int __mmu_interval_notifier_insert(
* The caller must use the normal interval notifier read flow via
* mmu_interval_read_begin() to establish SPTEs for this range.
*/
int mmu_interval_notifier_insert(struct mmu_interval_notifier *mni,
int mmu_interval_notifier_insert(struct mmu_interval_notifier *interval_sub,
struct mm_struct *mm, unsigned long start,
unsigned long length,
const struct mmu_interval_notifier_ops *ops)
......@@ -973,13 +983,13 @@ int mmu_interval_notifier_insert(struct mmu_interval_notifier *mni,
return ret;
subscriptions = mm->notifier_subscriptions;
}
return __mmu_interval_notifier_insert(mni, mm, subscriptions, start,
length, ops);
return __mmu_interval_notifier_insert(interval_sub, mm, subscriptions,
start, length, ops);
}
EXPORT_SYMBOL_GPL(mmu_interval_notifier_insert);
int mmu_interval_notifier_insert_locked(
struct mmu_interval_notifier *mni, struct mm_struct *mm,
struct mmu_interval_notifier *interval_sub, struct mm_struct *mm,
unsigned long start, unsigned long length,
const struct mmu_interval_notifier_ops *ops)
{
......@@ -995,14 +1005,14 @@ int mmu_interval_notifier_insert_locked(
return ret;
subscriptions = mm->notifier_subscriptions;
}
return __mmu_interval_notifier_insert(mni, mm, subscriptions, start,
length, ops);
return __mmu_interval_notifier_insert(interval_sub, mm, subscriptions,
start, length, ops);
}
EXPORT_SYMBOL_GPL(mmu_interval_notifier_insert_locked);
/**
* mmu_interval_notifier_remove - Remove a interval notifier
* @mni: Interval notifier to unregister
* @interval_sub: Interval subscription to unregister
*
* This function must be paired with mmu_interval_notifier_insert(). It cannot
* be called from any ops callback.
......@@ -1010,9 +1020,9 @@ EXPORT_SYMBOL_GPL(mmu_interval_notifier_insert_locked);
* Once this returns ops callbacks are no longer running on other CPUs and
* will not be called in future.
*/
void mmu_interval_notifier_remove(struct mmu_interval_notifier *mni)
void mmu_interval_notifier_remove(struct mmu_interval_notifier *interval_sub)
{
struct mm_struct *mm = mni->mm;
struct mm_struct *mm = interval_sub->mm;
struct mmu_notifier_subscriptions *subscriptions =
mm->notifier_subscriptions;
unsigned long seq = 0;
......@@ -1025,16 +1035,16 @@ void mmu_interval_notifier_remove(struct mmu_interval_notifier *mni)
* remove is being called after insert put this on the
* deferred list, but before the deferred list was processed.
*/
if (RB_EMPTY_NODE(&mni->interval_tree.rb)) {
hlist_del(&mni->deferred_item);
if (RB_EMPTY_NODE(&interval_sub->interval_tree.rb)) {
hlist_del(&interval_sub->deferred_item);
} else {
hlist_add_head(&mni->deferred_item,
hlist_add_head(&interval_sub->deferred_item,
&subscriptions->deferred_list);
seq = subscriptions->invalidate_seq;
}
} else {
WARN_ON(RB_EMPTY_NODE(&mni->interval_tree.rb));
interval_tree_remove(&mni->interval_tree,
WARN_ON(RB_EMPTY_NODE(&interval_sub->interval_tree.rb));
interval_tree_remove(&interval_sub->interval_tree,
&subscriptions->itree);
}
spin_unlock(&subscriptions->lock);
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment
GitLab Nexedi Edition | About GitLab | About Nexedi | 沪ICP备2021021310号-2 | 沪ICP备2021021310号-7