misc/sgi-xp: Replace in_interrupt() usage
PATCH 1/1 misc/sgi-xp: Replace in_interrupt() usage
From: Thomas Gleixner <[email protected]>
The usage of in_interrupt() in xpc_partition_disengaged() is clearly intended to avoid canceling the timeout timer when the function is invoked from the timer callback.
While in_interrupt() is deprecated and ill defined as it does not provide what the name suggests it catches the intended case.
Add an argument to xpc_partition_disengaged() which is true if called from timer and otherwise false.
Use del_timer_sync() instead of del_singleshot_timer_sync() which is the same thing.
Note: This does not prevent reentrancy into the function as the function has no concurrency control and timer callback and regular task context callers can happen concurrently on different CPUs or the timer can interrupt the task context before it is able to cancel it.
While the only driver which is providing the arch_xpc_ops callbacks (xpc_uv) seems not to have a reentrancy problem and the only negative effect would be a double dev_info() entry in dmesg, the whole mechanism is conceptually broken.
But that's not subject of this cleanup endeavour and left as an exercise to the folks who might have interest to make that code fully correct.
[bigeasy: Add the argument, use del_timer_sync().]
--- drivers/misc/sgi-xp/xpc.h | 2 +- drivers/misc/sgi-xp/xpc_main.c | 8 ++++---- drivers/misc/sgi-xp/xpc_partition.c | 9 ++++----- 3 files changed, 9 insertions(+), 10 deletions(-)
drivers/misc/sgi-xp/xpc.h +1 -1
@@ -633,7 +633,7 @@ extern void *xpc_kmalloc_cacheline_aligned(size_t, gfp_t, void **);extern int xpc_setup_rsvd_page (void );extern void xpc_teardown_rsvd_page (void );extern int xpc_identify_activate_IRQ_sender (void ); -extern int xpc_partition_disengaged (struct xpc_partition *); +extern int xpc_partition_disengaged (struct xpc_partition *,bool from_timer);extern enum xp_retval xpc_mark_partition_active (struct xpc_partition *);extern void xpc_mark_partition_inactive (struct xpc_partition *);extern void xpc_discovery (void );
drivers/misc/sgi-xp/xpc_main.c +4 -4
@@ -179,7 +179,7 @@ xpc_timeout_partition_disengage(struct timer_list *t)DBUG_ON (time_is_after_jiffies (part->disengage_timeout)); - (void )xpc_partition_disengaged (part); +xpc_partition_disengaged (part, true);DBUG_ON (part->disengage_timeout !=0 );DBUG_ON (xpc_arch_ops.partition_engaged (XPC_PARTID (part)));
@@ -341,7 +341,7 @@ xpc_channel_mgr(struct xpc_partition *part) {while (part->act_state !=XPC_P_AS_DEACTIVATING ||atomic_read (&part->nchannels_active) >0 || - !xpc_partition_disengaged (part)) { + !xpc_partition_disengaged (part, false)) {xpc_process_sent_chctl_flags (part);
@@ -364,7 +364,7 @@ xpc_channel_mgr(struct xpc_partition *part) part->chctl.all_flags !=0 || (part->act_state ==XPC_P_AS_DEACTIVATING &&atomic_read (&part->nchannels_active) ==0 && - xpc_partition_disengaged(part)))); + xpc_partition_disengaged(part, false))));atomic_set (&part->channel_mgr_requests,1 ); } }
@@ -983,7 +983,7 @@ xpc_do_exit(enum xp_retval reason)for (partid =0 ; partid < xp_max_npartitions; partid++) { part = &xpc_partitions[partid]; -if (xpc_partition_disengaged (part) && +if (xpc_partition_disengaged (part, false) && part->act_state ==XPC_P_AS_INACTIVE ) {continue ; }
drivers/misc/sgi-xp/xpc_partition.c +4 -5
@@ -262,8 +262,7 @@ xpc_get_remote_rp(int nasid, unsigned long *discovered_nasids,* from us. Though we requested the remote partition to deactivate with regard * to us, we really only need to wait for the other side to disengage from us. */ -int -xpc_partition_disengaged (struct xpc_partition *part) +int xpc_partition_disengaged (struct xpc_partition *part,bool from_timer) {short partid =XPC_PARTID (part);int disengaged;
@@ -289,9 +288,9 @@ xpc_partition_disengaged(struct xpc_partition *part) } part->disengage_timeout =0 ; -/* cancel the timer function, provided it's not us */ -if (!in_interrupt ()) -del_singleshot_timer_sync (&part->disengage_timer); +/* Cancel the timer function if not called from it */ +if (!from_timer) +del_timer_sync (&part->disengage_timer);DBUG_ON (part->act_state !=XPC_P_AS_DEACTIVATING && part->act_state !=XPC_P_AS_INACTIVE );
Signature · 1 line
2.29.2
Greg Kroah-Hartman
maintainerQuoted 49 lines · Sebastian Andrzej Siewior
From: Thomas Gleixner <[email protected]>
The usage of
in_interrupt()inxpc_partition_disengaged()is clearly intended to avoid canceling the timeout timer when the function is invoked from the timer callback.While
in_interrupt()is deprecated and ill defined as it does not provide what the name suggests it catches the intended case.Add an argument to
xpc_partition_disengaged()which is true if called from timer and otherwise false.
Usedel_timer_sync()instead ofdel_singleshot_timer_sync()which is the same thing.Note: This does not prevent reentrancy into the function as the function has no concurrency control and timer callback and regular task context callers can happen concurrently on different CPUs or the timer can interrupt the task context before it is able to cancel it.
While the only driver which is providing the
arch_xpc_opscallbacks (xpc_uv) seems not to have a reentrancy problem and the only negative effect would be a doubledev_info()entry in dmesg, the whole mechanism is conceptually broken.But that's not subject of this cleanup endeavour and left as an exercise to the folks who might have interest to make that code fully correct.
[bigeasy: Add the argument, use
del_timer_sync().]Signed-off-by Thomas Gleixner <[email protected]>Signed-off-by Sebastian Andrzej Siewior <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Cc Cliff Whickman <[email protected]>Cc Arnd Bergmann <[email protected]>Cc Robin Holt <[email protected]>Cc Steve Wahl <[email protected]>Cc Dimitri Sivanich <[email protected]>Cc Russ Anderson <[email protected]>--- drivers/misc/sgi-xp/xpc.h | 2 +- drivers/misc/sgi-xp/xpc_main.c | 8 ++++---- drivers/misc/sgi-xp/xpc_partition.c | 9 ++++----- 3 files changed, 9 insertions(+), 10 deletions(-)drivers/misc/sgi-xp/xpc.h +0
-0
drivers/misc/sgi-xp/xpc.h extern void *xpc_kmalloc_cacheline_aligned(size_t, gfp_t, void **); · line 636
Quoted 5 lines
drivers/misc/sgi-xp/xpc.h +1
-1@@ -633,7 +633,7 @@ extern void *xpc_kmalloc_cacheline_aligned(size_t, gfp_t, void **);extern int xpc_setup_rsvd_page (void );extern void xpc_teardown_rsvd_page (void );extern int xpc_identify_activate_IRQ_sender (void ); -extern int xpc_partition_disengaged (struct xpc_partition *); +extern int xpc_partition_disengaged (struct xpc_partition *,bool from_timer);
These types of "flags" for functions are horrible as they do not describe what is happening when you read the places the function is called.
Instead, make it part of the function name itself:
xpc_partition_disengaged_from_timer()
and then handle it that way, by using a shared static function with the flag.
Otherwise this type of change just does not age well at all.
thanks,
greg k-h
PATCH v2 1/1 misc/sgi-xp: Replace in_interrupt() usage
From: Thomas Gleixner <[email protected]>
The usage of in_interrupt() in xpc_partition_disengaged() is clearly intended to avoid canceling the timeout timer when the function is invoked from the timer callback.
While in_interrupt() is deprecated and ill defined as it does not provide what the name suggests it catches the intended case.
Add an argument to xpc_partition_disengaged() which is true if called from timer and otherwise false.
Use del_timer_sync() instead of del_singleshot_timer_sync() which is the same thing.
Note: This does not prevent reentrancy into the function as the function has no concurrency control and timer callback and regular task context callers can happen concurrently on different CPUs or the timer can interrupt the task context before it is able to cancel it.
While the only driver which is providing the arch_xpc_ops callbacks (xpc_uv) seems not to have a reentrancy problem and the only negative effect would be a double dev_info() entry in dmesg, the whole mechanism is conceptually broken.
But that's not subject of this cleanup endeavour and left as an exercise to the folks who might have interest to make that code fully correct.
[bigeasy: Add the argument, use del_timer_sync().]
---
v1…v2: Use xpc_partition_disengaged_from_timer() and don't export the
argument. Suggested by Greg.
drivers/misc/sgi-xp/xpc.h | 1 + drivers/misc/sgi-xp/xpc_main.c | 2 +- drivers/misc/sgi-xp/xpc_partition.c | 20 +++++++++++++++----- 3 files changed, 17 insertions(+), 6 deletions(-)
drivers/misc/sgi-xp/xpc.h +1 -0
@@ -634,6 +634,7 @@ extern int xpc_setup_rsvd_page(void);extern void xpc_teardown_rsvd_page (void );extern int xpc_identify_activate_IRQ_sender (void );extern int xpc_partition_disengaged (struct xpc_partition *); +extern int xpc_partition_disengaged_from_timer (struct xpc_partition *part);extern enum xp_retval xpc_mark_partition_active (struct xpc_partition *);extern void xpc_mark_partition_inactive (struct xpc_partition *);extern void xpc_discovery (void );
drivers/misc/sgi-xp/xpc_main.c +1 -1
@@ -179,7 +179,7 @@ xpc_timeout_partition_disengage(struct timer_list *t)DBUG_ON (time_is_after_jiffies (part->disengage_timeout)); - (void )xpc_partition_disengaged (part); +xpc_partition_disengaged_from_timer (part);DBUG_ON (part->disengage_timeout !=0 );DBUG_ON (xpc_arch_ops.partition_engaged (XPC_PARTID (part)));
drivers/misc/sgi-xp/xpc_partition.c +15 -5
@@ -262,8 +262,8 @@ xpc_get_remote_rp(int nasid, unsigned long *discovered_nasids,* from us. Though we requested the remote partition to deactivate with regard * to us, we really only need to wait for the other side to disengage from us. */ -int -xpc_partition_disengaged (struct xpc_partition *part) +static int __xpc_partition_disengaged (struct xpc_partition *part, +bool from_timer) {short partid =XPC_PARTID (part);int disengaged;
@@ -289,9 +289,9 @@ xpc_partition_disengaged(struct xpc_partition *part) } part->disengage_timeout =0 ; -/* cancel the timer function, provided it's not us */ -if (!in_interrupt ()) -del_singleshot_timer_sync (&part->disengage_timer); +/* Cancel the timer function if not called from it */ +if (!from_timer) +del_timer_sync (&part->disengage_timer);DBUG_ON (part->act_state !=XPC_P_AS_DEACTIVATING && part->act_state !=XPC_P_AS_INACTIVE );
@@ -303,6 +303,16 @@ xpc_partition_disengaged(struct xpc_partition *part)return disengaged; } +int xpc_partition_disengaged (struct xpc_partition *part) +{ +return __xpc_partition_disengaged (part, false); +} + +int xpc_partition_disengaged_from_timer (struct xpc_partition *part) +{ +return __xpc_partition_disengaged (part, true); +} +/* * Mark specified partition as active. */
Signature · 1 line
2.29.2
Quoted 120 lines · Sebastian Andrzej Siewior
From: Thomas Gleixner <[email protected]>
The usage of
in_interrupt()inxpc_partition_disengaged()is clearly intended to avoid canceling the timeout timer when the function is invoked from the timer callback.While
in_interrupt()is deprecated and ill defined as it does not provide what the name suggests it catches the intended case.Add an argument to
xpc_partition_disengaged()which is true if called from timer and otherwise false.
Usedel_timer_sync()instead ofdel_singleshot_timer_sync()which is the same thing.Note: This does not prevent reentrancy into the function as the function has no concurrency control and timer callback and regular task context callers can happen concurrently on different CPUs or the timer can interrupt the task context before it is able to cancel it.
While the only driver which is providing the
arch_xpc_opscallbacks (xpc_uv) seems not to have a reentrancy problem and the only negative effect would be a doubledev_info()entry in dmesg, the whole mechanism is conceptually broken.But that's not subject of this cleanup endeavour and left as an exercise to the folks who might have interest to make that code fully correct.
[bigeasy: Add the argument, use
del_timer_sync().]Signed-off-by Thomas Gleixner <[email protected]>Signed-off-by Sebastian Andrzej Siewior <[email protected]>Cc Greg Kroah-Hartman <[email protected]>Cc Cliff Whickman <[email protected]>Cc Arnd Bergmann <[email protected]>Cc Robin Holt <[email protected]>Cc Steve Wahl <[email protected]>Cc Dimitri Sivanich <[email protected]>Cc Russ Anderson <[email protected]>---
v1…v2: Usexpc_partition_disengaged_from_timer()and don't export the
argument. Suggested by Greg.drivers/misc/sgi-xp/xpc.h | 1 + drivers/misc/sgi-xp/xpc_main.c | 2 +- drivers/misc/sgi-xp/xpc_partition.c | 20 +++++++++++++++----- 3 files changed, 17 insertions(+), 6 deletions(-)drivers/misc/sgi-xp/xpc.h +1
-0@@ -634,6 +634,7 @@ extern int xpc_setup_rsvd_page(void);extern void xpc_teardown_rsvd_page (void );extern int xpc_identify_activate_IRQ_sender (void );extern int xpc_partition_disengaged (struct xpc_partition *); +extern int xpc_partition_disengaged_from_timer (struct xpc_partition *part);extern enum xp_retval xpc_mark_partition_active (struct xpc_partition *);extern void xpc_mark_partition_inactive (struct xpc_partition *);extern void xpc_discovery (void );drivers/misc/sgi-xp/xpc_main.c +1
-1@@ -179,7 +179,7 @@ xpc_timeout_partition_disengage(struct timer_list *t)DBUG_ON (time_is_after_jiffies (part->disengage_timeout)); - (void )xpc_partition_disengaged (part); +xpc_partition_disengaged_from_timer (part);DBUG_ON (part->disengage_timeout !=0 );DBUG_ON (xpc_arch_ops.partition_engaged (XPC_PARTID (part)));drivers/misc/sgi-xp/xpc_partition.c +15
-5@@ -262,8 +262,8 @@ xpc_get_remote_rp(int nasid, unsigned long *discovered_nasids,* from us. Though we requested the remote partition to deactivate with regard * to us, we really only need to wait for the other side to disengage from us. */ -int -xpc_partition_disengaged (struct xpc_partition *part) +static int __xpc_partition_disengaged (struct xpc_partition *part, +bool from_timer) {short partid =XPC_PARTID (part);int disengaged;@@ -289,9 +289,9 @@ xpc_partition_disengaged(struct xpc_partition *part) } part->disengage_timeout =0 ; -/* cancel the timer function, provided it's not us */ -if (!in_interrupt ()) -del_singleshot_timer_sync (&part->disengage_timer); +/* Cancel the timer function if not called from it */ +if (!from_timer) +del_timer_sync (&part->disengage_timer);DBUG_ON (part->act_state !=XPC_P_AS_DEACTIVATING && part->act_state !=XPC_P_AS_INACTIVE );@@ -303,6 +303,16 @@ xpc_partition_disengaged(struct xpc_partition *part)return disengaged; } +int xpc_partition_disengaged (struct xpc_partition *part) +{ +return __xpc_partition_disengaged (part, false); +} + +int xpc_partition_disengaged_from_timer (struct xpc_partition *part) +{ +return __xpc_partition_disengaged (part, true); +} +/* * Mark specified partition as active. */ Signature · 1 line
2.29.2
Signature · 1 line
Steve Wahl, Hewlett Packard Enterprise