enhance lock debugging
v3 cover enhance lock debugging
While hunting a locking problem in my core scheduling series I have added some debugging aids to spinlock handling making it easier to find the root cause for the problem.
Making use of the already existing lock profiling and enhancing it a little bit produces some really valuable diagnostic data e.g. when a NMI watchdog is triggering a crash.
Changes in V3:
- rebase to current staging (after realizing that patch 4 still applied, but resulting in patching a wrong function)
Changes in V2:
- multiple comments addressed
- added patch 5
Juergen Gross (5):
xen/spinlocks: in debug builds store cpu holding the lock
xen: add new CONFIG_DEBUG_LOCKS option
xen: print lock profile info in panic()
xen: modify lock profiling interface
xen: modify several static locks to unique names
tools/libxc/xc_misc.c | 1 + tools/misc/xenlockprof.c | 17 ++--- xen/Kconfig.debug | 10 ++- xen/arch/arm/xen.lds.S | 13 ++-- xen/arch/x86/domain.c | 2 +- xen/arch/x86/irq.c | 6 +- xen/arch/x86/time.c | 6 +- xen/arch/x86/xen.lds.S | 13 ++-- xen/common/domain.c | 4 +- xen/common/keyhandler.c | 8 +-- xen/common/perfc.c | 6 +- xen/common/spinlock.c | 161 ++++++++++++++++++++++++++++++-------------- xen/common/sysctl.c | 2 +- xen/common/trace.c | 6 +- xen/drivers/char/console.c | 10 +-- xen/include/public/sysctl.h | 11 ++- xen/include/xen/spinlock.h | 60 ++++++++++------- 17 files changed, 206 insertions(+), 130 deletions(-)
Signature · 7 lines
2.16.4 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v3 1/5 xen/spinlocks: in debug builds store cpu holding the lock
Add the cpu currently holding the lock to struct lock_debug. This makes analysis of locking errors easier and it can be tested whether the correct cpu is releasing a lock again.
---
V2: adjust types (Jan Beulich)
--- xen/common/spinlock.c | 33 ++++++++++++++++++++++++++------- xen/include/xen/spinlock.h | 17 ++++++++++++----- 2 files changed, 38 insertions(+), 12 deletions(-)
xen/common/spinlock.c +26 -7
@@ -13,9 +13,9 @@static atomic_t spin_debug__read_mostly =ATOMIC_INIT (0 ); -static void check_lock (struct lock_debug *debug) +static void check_lock (union lock_debug *debug) { -int irq_safe = !local_irq_is_enabled (); +bool irq_safe = !local_irq_is_enabled ();if (unlikely (atomic_read (&spin_debug) <=0 ) )return ;
@@ -43,18 +43,21 @@ static void check_lock(struct lock_debug *debug)*/ if (unlikely (debug->irq_safe != irq_safe) ) { -int seen =cmpxchg (&debug->irq_safe,-1 , irq_safe); +union lock_debug seen, new = {0 }; -if ( seen == !irq_safe ) + new.irq_safe = irq_safe; + seen.val =cmpxchg (&debug->val,LOCK_DEBUG_INITVAL , new.val); + +if ( !seen.unseen && seen.irq_safe == !irq_safe ) {printk ("CHECKLOCK FAILURE: prev irqsafe: %d, curr irqsafe %d\n" , - seen, irq_safe); + seen.irq_safe, irq_safe);BUG (); } } } -static void check_barrier (struct lock_debug *debug) +static void check_barrier (union lock_debug *debug) {if (unlikely (atomic_read (&spin_debug) <=0 ) )return ;
@@ -70,7 +73,18 @@ static void check_barrier(struct lock_debug *debug)* However, if we spin on an IRQ-unsafe lock with IRQs disabled then that * is clearly wrong, for the same reason outlined in check_lock() above. */ -BUG_ON (!local_irq_is_enabled () && (debug->irq_safe ==0 )); +BUG_ON (!local_irq_is_enabled () && !debug->irq_safe); +} + +static void got_lock (union lock_debug *debug) +{ + debug->cpu =smp_processor_id (); +} + +static void rel_lock (union lock_debug *debug) +{ +ASSERT (debug->cpu ==smp_processor_id ()); + debug->cpu =SPINLOCK_NO_CPU ; }void spin_debug_enable (void )
@@ -87,6 +101,8 @@ void spin_debug_disable(void)#define check_lock (l) ((void)0)#define check_barrier (l) ((void)0) +#define got_lock (l) ((void)0) +#define rel_lock (l) ((void)0)#endif
@@ -150,6 +166,7 @@ void inline _spin_lock_cb(spinlock_t *lock, void (*cb)(void *), void *data)cb (data);arch_lock_relax (); } +got_lock (&lock->debug);LOCK_PROFILE_GOT ;preempt_disable ();arch_lock_acquire_barrier ();
@@ -181,6 +198,7 @@ void _spin_unlock(spinlock_t *lock)arch_lock_release_barrier ();preempt_enable ();LOCK_PROFILE_REL ; +rel_lock (&lock->debug);add_sized (&lock->tickets.head,1 );arch_lock_signal (); }
@@ -224,6 +242,7 @@ int _spin_trylock(spinlock_t *lock)if (cmpxchg (&lock->tickets.head_tail, old.head_tail, new.head_tail) != old.head_tail )return 0 ; +got_lock (&lock->debug);#ifdef CONFIG_LOCK_PROFILE if (lock->profile) lock->profile->time_locked =NOW ();
xen/include/xen/spinlock.h +12 -5
@@ -6,14 +6,21 @@#include <asm/types.h> #ifndef NDEBUG -struct lock_debug { -s16 irq_safe;/* +1: IRQ-safe; 0: not IRQ-safe; -1: don't know yet */ +union lock_debug { +uint16_t val; +#define LOCK_DEBUG_INITVAL 0xffff +struct { +uint16_t cpu:12 ; +uint16_t pad:2 ; +bool irq_safe:1 ; +bool unseen:1 ; + }; }; -#define _LOCK_DEBUG { -1 } +#define _LOCK_DEBUG { LOCK_DEBUG_INITVAL }void spin_debug_enable (void );void spin_debug_disable (void );#else -struct lock_debug { }; +union lock_debug { };#define _LOCK_DEBUG { }#define spin_debug_enable () ((void)0)#define spin_debug_disable () ((void)0)
@@ -142,7 +149,7 @@ typedef struct spinlock {#define SPINLOCK_NO_CPU 0xfffu u16 recurse_cnt:4;#define SPINLOCK_MAX_RECURSE 0xfu -struct lock_debug debug; +union lock_debug debug;#ifdef CONFIG_LOCK_PROFILE struct lock_profile *profile;#endif
Signature · 7 lines
2.16.4 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 1/5] xen/spinlocks: in debug builds store cpu holding the lock
xen/include/xen/spinlock.h line 13
Quoted 13 lines · Juergen Gross
xen/include/xen/spinlock.h +5
-2@@ -6,14 +6,21 @@#include <asm/types.h> #ifndef NDEBUG -struct lock_debug { - s16 irq_safe; /* +1: IRQ-safe; 0: not IRQ-safe; -1: don't know yet */ +union lock_debug { + uint16_t val; +#define LOCK_DEBUG_INITVAL 0xffff + struct { + uint16_t cpu:12;
I'm afraid I have one more request: The literal 12 in struct spinlock is sitting right next to the SPINLOCK_NO_CPU definition, making it pretty unlikely that both author and reviewer of a change wouldn't notice one getting changed without adjustment to the other. The literal 12 here, though, is sufficiently remote to that other place, so there needs to be a connection established. Best I can think of right away is to have a #define for the 12, move SPINLOCK_NO_CPU's definition next to it, and use the new constant in both places.
xen/include/xen/spinlock.h line 14
Quoted 1 line
xen/include/xen/spinlock.h +1
-0+ uint16_t pad:2;
While at it, would you mind making this an unnamed field?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 1/5] xen/spinlocks: in debug builds store cpu holding the lock
Quoted 15 lines · Jan Beulich
xen/include/xen/spinlock.h line 13
Quoted 13 lines · Juergen Gross
xen/include/xen/spinlock.h +5
-2@@ -6,14 +6,21 @@#include <asm/types.h> #ifndef NDEBUG -struct lock_debug { - s16 irq_safe; /* +1: IRQ-safe; 0: not IRQ-safe; -1: don't know yet */ +union lock_debug { + uint16_t val; +#define LOCK_DEBUG_INITVAL 0xffff + struct { + uint16_t cpu:12;
Quoted 8 lines
I'm afraid I have one more request: The literal 12 in struct spinlock is sitting right next to the
SPINLOCK_NO_CPUdefinition, making it pretty unlikely that both author and reviewer of a change wouldn't notice one getting changed without adjustment to the other. The literal 12 here, though, is sufficiently remote to that other place, so there needs to be a connection established. Best I can think of right away is to have a #define for the 12, move SPINLOCK_NO_CPU's definition next to it, and use the new constant in both places.
Can do.
Quoted 4 lines
xen/include/xen/spinlock.h line 14
Quoted 2 lines
xen/include/xen/spinlock.h +1
-0+ uint16_t pad:2;While at it, would you mind making this an unnamed field?
NP. I guess the "2" should then be replaced to depend on the added #define above.
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 1/5] xen/spinlocks: in debug builds store cpu holding the lock
Quoted 28 lines · Juergen Gross
Quoted 15 lines · Jan Beulich
xen/include/xen/spinlock.h line 13
Quoted 13 lines · Juergen Gross
xen/include/xen/spinlock.h +5
-2@@ -6,14 +6,21 @@#include <asm/types.h> #ifndef NDEBUG -struct lock_debug { - s16 irq_safe; /* +1: IRQ-safe; 0: not IRQ-safe; -1: don't know yet */ +union lock_debug { + uint16_t val; +#define LOCK_DEBUG_INITVAL 0xffff + struct { + uint16_t cpu:12;Quoted 8 lines
I'm afraid I have one more request: The literal 12 in struct spinlock is sitting right next to the
SPINLOCK_NO_CPUdefinition, making it pretty unlikely that both author and reviewer of a change wouldn't notice one getting changed without adjustment to the other. The literal 12 here, though, is sufficiently remote to that other place, so there needs to be a connection established. Best I can think of right away is to have a #define for the 12, move SPINLOCK_NO_CPU's definition next to it, and use the new constant in both places.Can do.
Quoted 1 line
Quoted 6 lines
Quoted 3 lines
xen/include/xen/spinlock.h line 14
Quoted 1 line
xen/include/xen/spinlock.h +1
-0+ uint16_t pad:2;While at it, would you mind making this an unnamed field?
NP. I guess the "2" should then be replaced to depend on the added #define above.
Ah yes, I didn't even notice this yet.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v3 2/5 xen: add new CONFIG_DEBUG_LOCKS option
In reply to Juergen GrossInstead of enabling debugging for debug builds only add a dedicated Kconfig option for that purpose which defaults to DEBUG.
---
V2:
- rename to CONFIG_DEBUG_LOCKS (Jan Beulich)
--- xen/Kconfig.debug | 7 +++++++ xen/common/spinlock.c | 4 ++-- xen/include/xen/spinlock.h | 2 +- 3 files changed, 10 insertions(+), 3 deletions(-)
xen/Kconfig.debug +7 -0
@@ -51,6 +51,13 @@ config LOCK_PROFILE You can use serial console to print (and reset) using 'l' and 'L' respectively, or the 'xenlockprof' tool. +config DEBUG_LOCKS + bool "Lock debugging" + default DEBUG + ---help--- + Enable debugging features of lock handling. Some additional + checks will be performed when acquiring and releasing locks. + config PERF_COUNTERS bool "Performance Counters" ---help---
xen/common/spinlock.c +2 -2
@@ -9,7 +9,7 @@#include <asm/processor.h> #include <asm/atomic.h> -#ifndef NDEBUG +#ifdef CONFIG_DEBUG_LOCKS static atomic_t spin_debug__read_mostly =ATOMIC_INIT (0 );
@@ -97,7 +97,7 @@ void spin_debug_disable(void) atomic_dec(&spin_debug); } -#else /* defined(NDEBUG) */ +#else /* CONFIG_DEBUG_LOCKS */ #define check_lock (l) ((void)0)#define check_barrier (l) ((void)0)
xen/include/xen/spinlock.h +1 -1
@@ -5,7 +5,7 @@#include <asm/spinlock.h> #include <asm/types.h> -#ifndef NDEBUG +#ifdef CONFIG_DEBUG_LOCKS union lock_debug { uint16_t val; #define LOCK_DEBUG_INITVAL 0xffff
Signature · 7 lines
2.16.4 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 2/5] xen: add new CONFIG_DEBUG_LOCKS option
Quoted 4 lines · Juergen Gross
Instead of enabling debugging for debug builds only add a dedicated
Kconfigoption for that purpose which defaults to DEBUG.Signed-off-by Juergen Gross <[email protected]>
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v3 3/5 xen: print lock profile info in panic()
In reply to Juergen GrossPrint the lock profile data when the system crashes and add some more information for each lock data (lock address, cpu holding the lock). While at it use the PRI_stime format specifier for printing time data.
This is especially beneficial for watchdog triggered crashes in case of deadlocks.
In order to have the cpu holding the lock available let the lock profile config option select DEBUG_LOCKS.
As printing the lock profile data will make use of locking, too, we need to disable spinlock debugging before calling spinlock_profile_printall() from panic().
While at it remove a superfluous #ifdef CONFIG_LOCK_PROFILE and rename CONFIG_LOCK_PROFILE to CONFIG_DEBUG_LOCK_PROFILE.
Also move the .lockprofile.data section to init area in linker scripts as the data is no longer needed after boot.
---
V2:
- rename CONFIG_LOCK_PROFILE to CONFIG_DEBUG_LOCK_PROFILE (Jan Beulich)
- move .lockprofile.data section to init area in linker scripts
- use PRI_stime (Andrew Cooper)
- don't print "cpu=4095", but "not locked" (Andrew Cooper)
--- xen/Kconfig.debug | 3 ++- xen/arch/arm/xen.lds.S | 13 +++++++------ xen/arch/x86/domain.c | 2 +- xen/arch/x86/xen.lds.S | 13 +++++++------ xen/common/keyhandler.c | 2 +- xen/common/spinlock.c | 33 ++++++++++++++++++--------------- xen/common/sysctl.c | 2 +- xen/drivers/char/console.c | 4 +++- xen/include/xen/spinlock.h | 12 +++++++----- 9 files changed, 47 insertions(+), 37 deletions(-)
xen/Kconfig.debug +2 -1
@@ -44,8 +44,9 @@ config COVERAGE If unsure, say N here. -config LOCK_PROFILE +config DEBUG_LOCK_PROFILE bool "Lock Profiling" + select DEBUG_LOCKS ---help--- Lock profiling allows you to see how often locks are taken and blocked. You can use serial console to print (and reset) using 'l' and 'L'
xen/arch/arm/xen.lds.S +7 -6
@@ -54,12 +54,6 @@ SECTIONS *(.data.rel.ro) *(.data.rel.ro.*) -#ifdef CONFIG_LOCK_PROFILE - . = ALIGN(POINTER_ALIGN); - __lock_profile_start = .; - *(.lockprofile.data) - __lock_profile_end = .; -#endif . = ALIGN(POINTER_ALIGN); __param_start = .; *(.data.param)
@@ -173,6 +167,13 @@ SECTIONS . = ALIGN(4); *(.altinstr_replacement) +#ifdef CONFIG_DEBUG_LOCK_PROFILE + . = ALIGN(POINTER_ALIGN); + __lock_profile_start = .; + *(.lockprofile.data) + __lock_profile_end = .; +#endif + *(.init.data) *(.init.data.rel) *(.init.data.rel.*)
xen/arch/x86/domain.c +1 -1
@@ -305,7 +305,7 @@ struct domain *alloc_domain_struct(void)#endif -#ifndef CONFIG_LOCK_PROFILE +#ifndef CONFIG_DEBUG_LOCK_PROFILE BUILD_BUG_ON (sizeof (*d) >PAGE_SIZE );#endif d =alloc_xenheap_pages (order,MEMF_bits (bits));
xen/arch/x86/xen.lds.S +7 -6
@@ -128,12 +128,6 @@ SECTIONS *(.ex_table.pre) __stop___pre_ex_table = .; -#ifdef CONFIG_LOCK_PROFILE - . = ALIGN(POINTER_ALIGN); - __lock_profile_start = .; - *(.lockprofile.data) - __lock_profile_end = .; -#endif . = ALIGN(POINTER_ALIGN); __param_start = .; *(.data.param)
@@ -251,6 +245,13 @@ SECTIONS *(.altinstructions) __alt_instructions_end = .; +#ifdef CONFIG_DEBUG_LOCK_PROFILE + . = ALIGN(POINTER_ALIGN); + __lock_profile_start = .; + *(.lockprofile.data) + __lock_profile_end = .; +#endif + . = ALIGN(8); __ctors_start = .; *(.ctors)
xen/common/keyhandler.c +1 -1
@@ -62,7 +62,7 @@ static struct keyhandler { KEYHANDLER('P', perfc_reset, "reset performance counters", 0),#endif -#ifdef CONFIG_LOCK_PROFILE +#ifdef CONFIG_DEBUG_LOCK_PROFILE KEYHANDLER('l', spinlock_profile_printall, "print lock profile info", 1), KEYHANDLER('L', spinlock_profile_reset, "reset lock profile info", 0),#endif
xen/common/spinlock.c +18 -15
@@ -106,7 +106,7 @@ void spin_debug_disable(void)#endif -#ifdef CONFIG_LOCK_PROFILE +#ifdef CONFIG_DEBUG_LOCK_PROFILE #define LOCK_PROFILE_REL \ if (lock->profile) \
@@ -243,7 +243,7 @@ int _spin_trylock(spinlock_t *lock) old.head_tail, new.head_tail) != old.head_tail )return 0 ;got_lock (&lock->debug); -#ifdef CONFIG_LOCK_PROFILE +#ifdef CONFIG_DEBUG_LOCK_PROFILE if (lock->profile) lock->profile->time_locked =NOW ();#endif
@@ -258,7 +258,7 @@ int _spin_trylock(spinlock_t *lock)void _spin_barrier (spinlock_t *lock) {spinlock_tickets_t sample; -#ifdef CONFIG_LOCK_PROFILE +#ifdef CONFIG_DEBUG_LOCK_PROFILE s_time_t block =NOW ();#endif
@@ -269,7 +269,7 @@ void _spin_barrier(spinlock_t *lock) {while (observe_head (&lock->tickets) == sample.head )arch_lock_relax (); -#ifdef CONFIG_LOCK_PROFILE +#ifdef CONFIG_DEBUG_LOCK_PROFILE if ( lock->profile ) { lock->profile->time_block +=NOW () - block;
@@ -327,7 +327,7 @@ void _spin_unlock_recursive(spinlock_t *lock) } } -#ifdef CONFIG_LOCK_PROFILE +#ifdef CONFIG_DEBUG_LOCK_PROFILE struct lock_profile_anc { struct lock_profile_qhead *head_q; /* first head of this type */
@@ -362,14 +362,19 @@ static void spinlock_profile_iterate(lock_profile_subfunc *sub, void *par)static void spinlock_profile_print_elem (struct lock_profile *data,int32_t type,int32_t idx,void *par) { -if ( type ==LOCKPROF_TYPE_GLOBAL ) -printk ("%s %s:\n" , lock_profile_ancs[type].name, data->name); +struct spinlock *lock = data->lock; + +printk ("%s " , lock_profile_ancs[type].name); +if ( type !=LOCKPROF_TYPE_GLOBAL ) +printk ("%d " , idx); +printk ("%s: addr=%p, lockval=%08x, " , data->name, lock, + lock->tickets.head_tail); +if ( lock->debug.cpu ==SPINLOCK_NO_CPU ) +printk ("not locked\n" );else -printk ("%s %d %s:\n" , lock_profile_ancs[type].name, idx, data->name); -printk (" lock:%12" PRId64"(%08X:%08X), block:%12" PRId64"(%08X:%08X)\n" , - data->lock_cnt, (u32)(data->time_hold >>32 ), (u32 )data->time_hold, - data->block_cnt, (u32)(data->time_block >>32 ), - (u32 )data->time_block); +printk ("cpu=%d\n" , lock->debug.cpu); +printk (" lock:%" PRId64"(%" PRI_stime"), block:%" PRId64"(%" PRI_stime")\n" , + data->lock_cnt, data->time_hold, data->block_cnt, data->time_block); }void spinlock_profile_printall (unsigned char key)
@@ -488,7 +493,6 @@ void _lock_profile_deregister_struct(spin_unlock (&lock_profile_lock); } -#ifdef CONFIG_LOCK_PROFILE static int __init lock_prof_init (void ) {struct lock_profile **q;
@@ -507,6 +511,5 @@ static int __init lock_prof_init(void) return 0; }__initcall (lock_prof_init); -#endif -#endif /* LOCK_PROFILE */ +#endif /* CONFIG_DEBUG_LOCK_PROFILE */
xen/common/sysctl.c +1 -1
@@ -119,7 +119,7 @@ long do_sysctl(XEN_GUEST_HANDLE_PARAM(xen_sysctl_t) u_sysctl)break ;#endif -#ifdef CONFIG_LOCK_PROFILE +#ifdef CONFIG_DEBUG_LOCK_PROFILE case XEN_SYSCTL_lockprof_op: ret =spinlock_profile_control (&op->u.lockprof_op);break ;
xen/drivers/char/console.c +3 -1
@@ -1170,7 +1170,9 @@ void panic(const char *fmt, ...)unsigned long flags;static DEFINE_SPINLOCK (lock);static char buf[128 ]; - + +spin_debug_disable (); +spinlock_profile_printall ('\0' );debugtrace_dump ();/* Protects buf[] and ensure multi-line message prints atomically. */
xen/include/xen/spinlock.h +7 -5
@@ -1,6 +1,7 @@#ifndef __SPINLOCK_H__#define __SPINLOCK_H__ +#include <xen/time.h> #include <asm/system.h> #include <asm/spinlock.h> #include <asm/types.h>
@@ -26,7 +27,7 @@ union lock_debug { };#define spin_debug_disable () ((void)0)#endif -#ifdef CONFIG_LOCK_PROFILE +#ifdef CONFIG_DEBUG_LOCK_PROFILE #include <public/sysctl.h>
@@ -72,9 +73,9 @@ struct lock_profile {struct spinlock *lock;/* the lock itself */ u64 lock_cnt;/* # of complete locking ops */ u64 block_cnt;/* # of complete wait for lock */ -s64 time_hold;/* cumulated lock time */ -s64 time_block;/* cumulated wait time */ -s64 time_locked;/* system time of last locking */ +s_time_t time_hold;/* cumulated lock time */ +s_time_t time_block;/* cumulated wait time */ +s_time_t time_locked;/* system time of last locking */ };struct lock_profile_qhead {
@@ -130,6 +131,7 @@ struct lock_profile_qhead { };#define spin_lock_init_prof (s, l) spin_lock_init(&((s)->l))#define lock_profile_register_struct (type, ptr, idx, print)#define lock_profile_deregister_struct (type, ptr) +#define spinlock_profile_printall (key)#endif
@@ -150,7 +152,7 @@ typedef struct spinlock { u16 recurse_cnt:4;#define SPINLOCK_MAX_RECURSE 0xfuunion lock_debug debug; -#ifdef CONFIG_LOCK_PROFILE +#ifdef CONFIG_DEBUG_LOCK_PROFILE struct lock_profile *profile;#endif } spinlock_t;
Signature · 7 lines
2.16.4 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 3/5] xen: print lock profile info in panic()
Quoted 3 lines · Juergen Gross
V2:
- renameCONFIG_LOCK_PROFILEtoCONFIG_DEBUG_LOCK_PROFILE(Jan Beulich)
- move .lockprofile.data section to init area in linker scripts
How can this be correct, when you don't change lock_prof_init() at all? The function builds a linked list from the entries in the section, and then hands the head of this list to _lock_profile_register_struct(). I guess I must be missing something. Anyway, everything else here looks good to me.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 3/5] xen: print lock profile info in panic()
Quoted 10 lines · Jan Beulich
Quoted 3 lines · Juergen Gross
V2:
- renameCONFIG_LOCK_PROFILEtoCONFIG_DEBUG_LOCK_PROFILE(Jan Beulich)
- move .lockprofile.data section to init area in linker scriptsHow can this be correct, when you don't change
lock_prof_init()at all? The function builds a linked list from the entries in the section, and then hands the head of this list to_lock_profile_register_struct(). I guess I must be missing something. Anyway, everything else here looks good to me.
The .lockprofile.data section contains only the pointers to the elements put into the linked list. And those pointers are no longer needed afterwards.
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Jan Beulich
Acked-byRe: [Xen-devel] [PATCH v3 3/5] xen: print lock profile info in panic()
Quoted 12 lines · Juergen Gross
Quoted 10 lines · Jan Beulich
Quoted 3 lines · Juergen Gross
V2:
- renameCONFIG_LOCK_PROFILEtoCONFIG_DEBUG_LOCK_PROFILE(Jan Beulich)
- move .lockprofile.data section to init area in linker scriptsHow can this be correct, when you don't change
lock_prof_init()at all? The function builds a linked list from the entries in the section, and then hands the head of this list to_lock_profile_register_struct(). I guess I must be missing something. Anyway, everything else here looks good to me.
Quoted 3 lines
The .lockprofile.data section contains only the pointers to the elements put into the linked list. And those pointers are no longer needed afterwards.
Oh, yes, I see now. Arguably worth making a separate change, but since it's this way now
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v3 4/5 xen: modify lock profiling interface
In reply to Juergen GrossToday adding locks located in a struct to lock profiling requires a unique type index for each structure. This makes it hard to add any new structure as the related sysctl interface needs to be changed, too.
Instead of using an index the already existing struct name specified in lock_profile_register_struct() can be used as an identifier instead.
Modify the sysctl interface to use the struct name instead of the type index and adapt the related coding accordingly. Instead of an array of struct anchors for lock profiling we now use a linked list for that purpose. Use the special idx value -1 for cases where the idx isn't relevant (global locks) and shouldn't be printed.
Increment the sysctl.h interface version as an interface is being modified.
Add the missing setting of profiling time to xc_lockprof_query().
Add freeing element data when deregistering a structure.
---
V2:
- add comment about memory allocation needs (Jan Beulich)
- check success of memory allocation (Jan Beulich)
--- tools/libxc/xc_misc.c | 1 + tools/misc/xenlockprof.c | 17 ++------ xen/common/domain.c | 4 +- xen/common/spinlock.c | 95 ++++++++++++++++++++++++++++++++------------- xen/include/public/sysctl.h | 11 ++---- xen/include/xen/spinlock.h | 29 ++++++++------ 6 files changed, 95 insertions(+), 62 deletions(-)
tools/libxc/xc_misc.c +1 -0
@@ -533,6 +533,7 @@ int xc_lockprof_query(xc_interface *xch, rc =do_sysctl (xch, &sysctl); *n_elems = sysctl.u.lockprof_op.nr_elem; + *time = sysctl.u.lockprof_op.time;return rc; }
tools/misc/xenlockprof.c +4 -13
@@ -88,19 +88,10 @@ int main(int argc, char *argv[]) sb =0 ;for ( j =0 ; j < i; j++ ) { -switch ( data[j].type ) - { -case LOCKPROF_TYPE_GLOBAL : -sprintf (name,"global lock %s" , data[j].name); -break ; -case LOCKPROF_TYPE_PERDOM : -sprintf (name,"domain %d lock %s" , data[j].idx, data[j].name); -break ; -default : -sprintf (name,"unknown type(%d) %d lock %s" , data[j].type, - data[j].idx, data[j].name); -break ; - } +if ( data[j].idx ==LOCKPROF_IDX_NONE ) +sprintf (name,"%s %s" , data[j].type, data[j].name); +else +sprintf (name,"%s %d %s" , data[j].type, data[j].idx, data[j].name); l = (double )(data[j].lock_time) /1E+09 ; b = (double )(data[j].block_time) /1E+09 ; sl += l;
xen/common/domain.c +2 -2
@@ -290,7 +290,7 @@ static void _domain_destroy(struct domain *d)xsm_free_security_domain (d); -lock_profile_deregister_struct (LOCKPROF_TYPE_PERDOM , d); +lock_profile_deregister_struct (d,"Domain" );free_domain_struct (d); }
@@ -369,7 +369,7 @@ struct domain *domain_create(domid_t domid, d->max_vcpus = config->max_vcpus; } -lock_profile_register_struct (LOCKPROF_TYPE_PERDOM , d, domid,"Domain" ); +lock_profile_register_struct (d, domid,"Domain" );if ( (err =xsm_alloc_security_domain (d)) !=0 ) goto fail;
xen/common/spinlock.c +67 -28
170 lines not on this page · raw message
xen/include/public/sysctl.h +4 -7
@@ -35,7 +35,7 @@#include "domctl.h" #include "physdev.h" -#define XEN_SYSCTL_INTERFACE_VERSION 0x00000012 +#define XEN_SYSCTL_INTERFACE_VERSION 0x00000013/* * Read console content from Xen buffer ring.
@@ -433,14 +433,11 @@ struct xen_sysctl_page_offline_op {/* Sub-operations: */ #define XEN_SYSCTL_LOCKPROF_reset 1/* Reset all profile data to zero. */ #define XEN_SYSCTL_LOCKPROF_query 2/* Get lock profile information. */ -/* Record-type: */ -#define LOCKPROF_TYPE_GLOBAL 0/* global lock, idx meaningless */ -#define LOCKPROF_TYPE_PERDOM 1/* per-domain lock, idx is domid */ -#define LOCKPROF_TYPE_N 2/* number of types */ struct xen_sysctl_lockprof_data { -char name[40 ];/* lock name (may include up to 2 %d specifiers) */ -int32_t type;/* LOCKPROF_TYPE_??? */ +char name[40 ];/* lock name */ +char type[20 ];/* e.g. "domain" */ int32_t idx;/* index (e.g. domain id) */ +#define LOCKPROF_IDX_NONE -1uint64_aligned_t lock_cnt;/* # of locking succeeded */ uint64_aligned_t block_cnt;/* # of wait for lock */ uint64_aligned_t lock_time;/* nsecs lock held */
xen/include/xen/spinlock.h +17 -12
@@ -50,19 +50,24 @@ union lock_debug { };with ptr being the main structure pointer and lock the spinlock field +It should be noted that this will need to allocate memory, so interrupts +must be enabled. +- each structure has to be added to profiling with -lock_profile_register_struct(type, ptr, idx, print); +lock_profile_register_struct(ptr, idx, print); with: -type: something like LOCKPROF_TYPE_PERDOM ptr: pointer to the structure idx: index of that structure, e.g. domid print: descriptive string like "domain" +It should be noted that this will might need to allocate memory, so +interrupts must be enabled. +- removing of a structure is done via -lock_profile_deregister_struct(type, ptr); +lock_profile_deregister_struct(ptr, print); */ struct spinlock ;
@@ -108,14 +113,14 @@ struct lock_profile_qhead { (s)->profile_head.elem_q = prof; \ }while (0 ) -void _lock_profile_register_struct( - int32_t, struct lock_profile_qhead *, int32_t, char *); -void _lock_profile_deregister_struct (int32_t ,struct lock_profile_qhead *); +void _lock_profile_register_struct(struct lock_profile_qhead *, int32_t, +const char *); +void _lock_profile_deregister_struct (struct lock_profile_qhead *,const char *); -#define lock_profile_register_struct (type, ptr, idx, print) \ - _lock_profile_register_struct(type, &((ptr)->profile_head), idx, print) -#define lock_profile_deregister_struct (type, ptr) \ - _lock_profile_deregister_struct(type, &((ptr)->profile_head)) +#define lock_profile_register_struct (ptr, idx, print) \ + _lock_profile_register_struct(&((ptr)->profile_head), idx, print) +#define lock_profile_deregister_struct (ptr, print) \ + _lock_profile_deregister_struct(&((ptr)->profile_head), print)extern int spinlock_profile_control (struct xen_sysctl_lockprof_op *pc);extern void spinlock_profile_printall (unsigned char key);
@@ -129,8 +134,8 @@ struct lock_profile_qhead { };#define DEFINE_SPINLOCK (l) spinlock_t l = SPIN_LOCK_UNLOCKED#define spin_lock_init_prof (s, l) spin_lock_init(&((s)->l)) -#define lock_profile_register_struct (type, ptr, idx, print) -#define lock_profile_deregister_struct (type, ptr) +#define lock_profile_register_struct (ptr, idx, print) +#define lock_profile_deregister_struct (ptr, print)#define spinlock_profile_printall (key)#endif
Signature · 7 lines
2.16.4 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 4/5] xen: modify lock profiling interface
Quoted 4 lines · Juergen Gross
Today adding locks located in a struct to lock profiling requires a unique type index for each structure. This makes it hard to add any new structure as the related sysctl interface needs to be changed, too.
Quoted 8 lines
Instead of using an index the already existing struct name specified in
lock_profile_register_struct()can be used as an identifier instead.Modify the sysctl interface to use the struct name instead of the type index and adapt the related coding accordingly. Instead of an array of struct anchors for lock profiling we now use a linked list for that purpose. Use the special idx value -1 for cases where the idx isn't relevant (global locks) and shouldn't be printed.
Just to make explicit what was implied by my v1 reply: I can see why you want to do this, but I'm not really a friend of string literals in the hypercall interface, when binary values can also do. So to me this looks to be a move in the wrong direction. Therefore, while I'm fine reviewing such a change, I'm not very likely to eventually ack it.
Quoted 6 lines
xen/common/spinlock.c +0
-2@@ -465,31 +466,70 @@ int spinlock_profile_control(struct xen_sysctl_lockprof_op *pc)return rc; } -void _lock_profile_register_struct ( -int32_t type,struct lock_profile_qhead *qhead,int32_t idx,char *name)
xen/common/spinlock.c int spinlock_profile_control(struct xen_sysctl_lockprof_op *pc) · line 477
Quoted 10 lines
xen/common/spinlock.c +8
-1+static struct lock_profile_anc *find_prof_anc (const char *name) { - qhead->idx = idx; +struct lock_profile_anc *anc; + +for ( anc = lock_profile_ancs; anc; anc = anc->next ) +if ( !strcmp (anc->name, name) ) +return anc; +return NULL ; +}
Blank line before main return statement please.
Quoted 4 lines
xen/common/spinlock.c +4
-0+void _lock_profile_register_struct (struct lock_profile_qhead *qhead, +int32_t idx,const char *name) +{ +struct lock_profile_anc *anc;
xen/common/spinlock.c int spinlock_profile_control(struct xen_sysctl_lockprof_op *pc) · line 489
Quoted 10 lines
xen/common/spinlock.c +6
-3+spin_lock (&lock_profile_lock); - qhead->head_q = lock_profile_ancs[type].head_q; - lock_profile_ancs[type].head_q = qhead; - lock_profile_ancs[type].name = name; + + anc =find_prof_anc (name); +if ( !anc ) + { + anc = xzalloc(struct lock_profile_anc);
Hmm, another allocation with a lock held. We try to avoid such as much as possible, and it doesn't look overly difficult to avoid it here.
Quoted 33 lines
xen/common/spinlock.c +20
-4+if ( !anc ) + goto out; + anc->name = name; + anc->next = lock_profile_ancs; + lock_profile_ancs = anc; + } + + qhead->idx = idx; + qhead->head_q = anc->head_q; + anc->head_q = qhead; + + out:spin_unlock (&lock_profile_lock); } -void _lock_profile_deregister_struct ( -int32_t type,struct lock_profile_qhead *qhead) +void _lock_profile_deregister_struct (struct lock_profile_qhead *qhead, +const char *name) { +struct lock_profile_anc *anc;struct lock_profile_qhead **q; +struct lock_profile *elem;spin_lock (&lock_profile_lock); -for ( q = &lock_profile_ancs[type].head_q; *q; q = &(*q)->head_q ) + + anc =find_prof_anc (name); +if ( anc ) { -if ( *q == qhead ) +for ( q = &anc->head_q; *q; q = &(*q)->head_q ) {
xen/common/spinlock.c int spinlock_profile_control(struct xen_sysctl_lockprof_op *pc) · line 526
Quoted 10 lines
xen/common/spinlock.c +8
-2- *q = qhead->head_q; -break ; +if ( *q == qhead ) + { + *q = qhead->head_q; +while ( qhead->elem_q ) + { + elem = qhead->elem_q; + qhead->elem_q = elem->next; +xfree (elem);
Which assumes the global list would never get handed here. Probably fine.
Quoted 11 lines
xen/include/xen/spinlock.h +3
-0@@ -50,19 +50,24 @@ union lock_debug { }; with ptr being the main structure pointer and lock the spinlock field + It should be noted that this will need to allocate memory, so interrupts + must be enabled. + - each structure has to be added to profiling with
xen/include/xen/spinlock.h union lock_debug { }; · line 65
Quoted 10 lines
xen/include/xen/spinlock.h +2
-2-lock_profile_register_struct (type, ptr, idx, print); +lock_profile_register_struct (ptr, idx, print); with: - type: something like LOCKPROF_TYPE_PERDOM ptr: pointer to the structure idx: index of that structure, e.g. domid print: descriptive string like "domain" + It should be noted that this will might need to allocate memory, so
Nit: "will" or "might", but not both.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 4/5] xen: modify lock profiling interface
Quoted 14 lines · Jan Beulich
Quoted 4 lines · Juergen Gross
Today adding locks located in a struct to lock profiling requires a unique type index for each structure. This makes it hard to add any new structure as the related sysctl interface needs to be changed, too.
Quoted 8 lines
Instead of using an index the already existing struct name specified in
lock_profile_register_struct()can be used as an identifier instead.Modify the sysctl interface to use the struct name instead of the type index and adapt the related coding accordingly. Instead of an array of struct anchors for lock profiling we now use a linked list for that purpose. Use the special idx value -1 for cases where the idx isn't relevant (global locks) and shouldn't be printed.
Quoted 6 lines
Just to make explicit what was implied by my v1 reply: I can see why you want to do this, but I'm not really a friend of string literals in the hypercall interface, when binary values can also do. So to me this looks to be a move in the wrong direction. Therefore, while I'm fine reviewing such a change, I'm not very likely to eventually ack it.
I'll write two example patches for adding support of lock profiling in a new structure, one with patch 4 of this series applied and one for the interface without that patch. This should make clear why I'm really in favor of this patch.
Quoted 5 lines
Quoted 5 lines
xen/common/spinlock.c +0
-0@@ -465,31 +466,70 @@ int spinlock_profile_control(struct xen_sysctl_lockprof_op *pc)return rc; }
Quoted 14 lines
xen/common/spinlock.c int spinlock_profile_control(struct xen_sysctl_lockprof_op *pc) · line 477
Quoted 12 lines
xen/common/spinlock.c +8
-3-void _lock_profile_register_struct ( -int32_t type,struct lock_profile_qhead *qhead,int32_t idx,char *name) +static struct lock_profile_anc *find_prof_anc (const char *name) { - qhead->idx = idx; +struct lock_profile_anc *anc; + +for ( anc = lock_profile_ancs; anc; anc = anc->next ) +if ( !strcmp (anc->name, name) ) +return anc; +return NULL ; +}Blank line before main return statement please.
Yes.
Quoted 11 lines
Quoted 11 lines
xen/common/spinlock.c +6
-3+void _lock_profile_register_struct (struct lock_profile_qhead *qhead, +int32_t idx,const char *name) +{ +struct lock_profile_anc *anc; +spin_lock (&lock_profile_lock); - qhead->head_q = lock_profile_ancs[type].head_q; - lock_profile_ancs[type].head_q = qhead; - lock_profile_ancs[type].name = name; +
Quoted 8 lines
xen/common/spinlock.c int spinlock_profile_control(struct xen_sysctl_lockprof_op *pc) · line 489
Quoted 4 lines
xen/common/spinlock.c +4
-0+ anc =find_prof_anc (name); +if ( !anc ) + { + anc = xzalloc(struct lock_profile_anc);Hmm, another allocation with a lock held. We try to avoid such as much as possible, and it doesn't look overly difficult to avoid it here.
I'll modify it.
Quoted 39 lines
Quoted 39 lines
xen/common/spinlock.c +23
-6+if ( !anc ) + goto out; + anc->name = name; + anc->next = lock_profile_ancs; + lock_profile_ancs = anc; + } + + qhead->idx = idx; + qhead->head_q = anc->head_q; + anc->head_q = qhead; + + out:spin_unlock (&lock_profile_lock); } -void _lock_profile_deregister_struct ( -int32_t type,struct lock_profile_qhead *qhead) +void _lock_profile_deregister_struct (struct lock_profile_qhead *qhead, +const char *name) { +struct lock_profile_anc *anc;struct lock_profile_qhead **q; +struct lock_profile *elem;spin_lock (&lock_profile_lock); -for ( q = &lock_profile_ancs[type].head_q; *q; q = &(*q)->head_q ) + + anc =find_prof_anc (name); +if ( anc ) { -if ( *q == qhead ) +for ( q = &anc->head_q; *q; q = &(*q)->head_q ) { - *q = qhead->head_q; -break ; +if ( *q == qhead ) + { + *q = qhead->head_q;
Quoted 8 lines
xen/common/spinlock.c int spinlock_profile_control(struct xen_sysctl_lockprof_op *pc) · line 526
Quoted 5 lines
xen/common/spinlock.c +5
-0+while ( qhead->elem_q ) + { + elem = qhead->elem_q; + qhead->elem_q = elem->next; +xfree (elem);Which assumes the global list would never get handed here. Probably fine.
I can add an ASSERT() to make this very clear.
Quoted 21 lines
Quoted 21 lines
xen/include/xen/spinlock.h +4
-2@@ -50,19 +50,24 @@ union lock_debug { }; with ptr being the main structure pointer and lock the spinlock field + It should be noted that this will need to allocate memory, so interrupts + must be enabled. + - each structure has to be added to profiling with -lock_profile_register_struct (type ,ptr ,idx ,lock_profile_register_struct (ptr ,idx ,
Quoted 3 lines
xen/include/xen/spinlock.h union lock_debug { }; · line 65
Quoted 1 line
xen/include/xen/spinlock.h +1
-0+ It should be noted that this will might need to allocate memory, soNit: "will" or "might", but not both.
Indeed.
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 4/5] xen: modify lock profiling interface
Quoted 22 lines · Juergen Gross
Quoted 14 lines · Jan Beulich
Quoted 4 lines · Juergen Gross
Today adding locks located in a struct to lock profiling requires a unique type index for each structure. This makes it hard to add any new structure as the related sysctl interface needs to be changed, too.
Quoted 8 lines
Instead of using an index the already existing struct name specified in
lock_profile_register_struct()can be used as an identifier instead.Modify the sysctl interface to use the struct name instead of the type index and adapt the related coding accordingly. Instead of an array of struct anchors for lock profiling we now use a linked list for that purpose. Use the special idx value -1 for cases where the idx isn't relevant (global locks) and shouldn't be printed.
Quoted 6 lines
Just to make explicit what was implied by my v1 reply: I can see why you want to do this, but I'm not really a friend of string literals in the hypercall interface, when binary values can also do. So to me this looks to be a move in the wrong direction. Therefore, while I'm fine reviewing such a change, I'm not very likely to eventually ack it.
Quoted 4 lines
I'll write two example patches for adding support of lock profiling in a new structure, one with patch 4 of this series applied and one for the interface without that patch. This should make clear why I'm really in favor of this patch.
Well, I can easily see how one will be quite a bit smaller than the other, but patch size / intrusiveness is not the only aspect to consider. IOW I'm afraid the difference won't change my opinion on string literals in hypercall interfaces. But recall, you don't depend on me acking this patch of yours, there are enough other people who can if they are happy with such a model.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v3 5/5 xen: modify several static locks to unique names
In reply to Juergen GrossIn order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
--- xen/arch/x86/irq.c | 6 +++--- xen/arch/x86/time.c | 6 +++--- xen/common/keyhandler.c | 6 +++--- xen/common/perfc.c | 6 +++--- xen/common/trace.c | 6 +++--- xen/drivers/char/console.c | 6 +++--- 6 files changed, 18 insertions(+), 18 deletions(-)
xen/arch/x86/irq.c +3 -3
@@ -867,14 +867,14 @@ void set_direct_apic_vector(void alloc_direct_apic_vector (uint8_t *vector,void (*handler)(struct cpu_user_regs *)) { -static DEFINE_SPINLOCK (lock); +static DEFINE_SPINLOCK (apic_alloc_lock); -spin_lock (&lock); +spin_lock (&apic_alloc_lock);if (*vector ==0 ) { *vector =alloc_hipriority_vector ();set_direct_apic_vector (*vector, handler); } -spin_unlock (&lock); +spin_unlock (&apic_alloc_lock); }void do_IRQ (struct cpu_user_regs *regs)
xen/arch/x86/time.c +3 -3
@@ -1426,9 +1426,9 @@ static void tsc_check_slave(void *unused)static void tsc_check_reliability (void ) {unsigned int cpu =smp_processor_id (); -static DEFINE_SPINLOCK (lock); +static DEFINE_SPINLOCK (tsc_check_lock); -spin_lock (&lock); +spin_lock (&tsc_check_lock); tsc_check_count++;smp_call_function (tsc_check_slave,NULL ,0 );
@@ -1439,7 +1439,7 @@ static void tsc_check_reliability(void)while ( !cpumask_empty (&tsc_check_cpumask) )cpu_relax (); -spin_unlock (&lock); +spin_unlock (&tsc_check_lock); }/*
xen/common/keyhandler.c +3 -3
@@ -371,9 +371,9 @@ static void read_clocks(unsigned char key)static u64 sumdif_stime =0 , maxdif_stime =0 ;static u64 sumdif_cycles =0 , maxdif_cycles =0 ;static u32 count =0 ; -static DEFINE_SPINLOCK (lock); +static DEFINE_SPINLOCK (read_clocks_lock); -spin_lock (&lock); +spin_lock (&read_clocks_lock);smp_call_function (read_clocks_slave,NULL ,0 );
@@ -408,7 +408,7 @@ static void read_clocks(unsigned char key) min_cycles =per_cpu (read_cycles_time, min_cycles_cpu); max_cycles =per_cpu (read_cycles_time, max_cycles_cpu); -spin_unlock (&lock); +spin_unlock (&read_clocks_lock); dif_stime = max_stime - min_stime;if ( dif_stime > maxdif_stime )
xen/common/perfc.c +3 -3
@@ -241,10 +241,10 @@ static int perfc_copy_info(XEN_GUEST_HANDLE_64(xen_sysctl_perfc_desc_t) desc,/* Dom0 control of perf counters */ int perfc_control (struct xen_sysctl_perfc_op *pc) { -static DEFINE_SPINLOCK (lock); +static DEFINE_SPINLOCK (perfc_control_lock);int rc; -spin_lock (&lock); +spin_lock (&perfc_control_lock);switch ( pc->cmd ) {
@@ -262,7 +262,7 @@ int perfc_control(struct xen_sysctl_perfc_op *pc)break ; } -spin_unlock (&lock); +spin_unlock (&perfc_control_lock); pc->nr_counters =NR_PERFCTRS ; pc->nr_vals = perfc_nbr_vals;
xen/common/trace.c +3 -3
@@ -368,10 +368,10 @@ void __init init_trace_bufs(void)*/ int tb_control (struct xen_sysctl_tbuf_op *tbc) { -static DEFINE_SPINLOCK (lock); +static DEFINE_SPINLOCK (tb_control_lock);int rc =0 ; -spin_lock (&lock); +spin_lock (&tb_control_lock);switch ( tbc->cmd ) {
@@ -432,7 +432,7 @@ int tb_control(struct xen_sysctl_tbuf_op *tbc)break ; } -spin_unlock (&lock); +spin_unlock (&tb_control_lock);return rc; }
xen/drivers/char/console.c +3 -3
@@ -1168,7 +1168,7 @@ void panic(const char *fmt, ...) {va_list args;unsigned long flags; -static DEFINE_SPINLOCK (lock); +static DEFINE_SPINLOCK (panic_lock);static char buf[128 ];spin_debug_disable ();
@@ -1176,7 +1176,7 @@ void panic(const char *fmt, ...)debugtrace_dump ();/* Protects buf[] and ensure multi-line message prints atomically. */ -spin_lock_irqsave (&lock, flags); +spin_lock_irqsave (&panic_lock, flags);va_start (args, fmt); (void )vsnprintf (buf,sizeof (buf), fmt, args);
@@ -1196,7 +1196,7 @@ void panic(const char *fmt, ...)printk ("Reboot in five seconds...\n" );#endif -spin_unlock_irqrestore (&lock, flags); +spin_unlock_irqrestore (&panic_lock, flags);debugger_trap_immediate ();
Signature · 7 lines
2.16.4 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define _LOCK_PROFILE(name) { 0, #name, &name, 0, 0, 0, 0, 0 }
so I'd rather suggest making this a derivation of a new
#define _LOCK_PROFILE_NAME(lock, name) { 0, #name, &lock, 0, 0, 0, 0, 0 }
if there's no other (transparent) way of disambiguating the names.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 9 lines · Jan Beulich
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
Quoted 7 lines
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }if there's no other (transparent) way of disambiguating the names.
This will require to use a different DEFINE_SPINLOCK() variant, so e.g. DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and add "@<func>" to the lock profiling name. Is this okay?
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 14 lines · Juergen Gross
Quoted 13 lines · Jan Beulich
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
Quoted 7 lines
Quoted 3 lines
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }if there's no other (transparent) way of disambiguating the names.
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and add "@<func>" to the lock profiling name. Is this okay?
To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 19 lines · Jan Beulich
Quoted 18 lines · Juergen Gross
Quoted 16 lines · Jan Beulich
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }if there's no other (transparent) way of disambiguating the names.
Quoted 7 lines
Quoted 3 lines
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and add "@<func>" to the lock profiling name. Is this okay?To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
NP with me. It was Andrew who asked for a way to differentiate between multiple locks. I'm not depending in any way on this patch.
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
In reply to Jan BeulichQuoted 19 lines · Jan Beulich
Quoted 18 lines · Juergen Gross
Quoted 16 lines · Jan Beulich
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }if there's no other (transparent) way of disambiguating the names.
Quoted 7 lines
Quoted 3 lines
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and add "@<func>" to the lock profiling name. Is this okay?To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
I think I have found a way: I could add __FILE__ and __LINE__ data to struct lock_profile. In lock_prof_init() I could look for non-unique lock names and mark those to be printed with the __FILE__ and __LINE__ data added to the names.
Would you be fine with this approach?
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 28 lines · Juergen Gross
Quoted 19 lines · Jan Beulich
Quoted 18 lines · Juergen Gross
Quoted 16 lines · Jan Beulich
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }if there's no other (transparent) way of disambiguating the names.
Quoted 7 lines
Quoted 3 lines
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and add "@<func>" to the lock profiling name. Is this okay?To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
Quoted 6 lines
I think I have found a way: I could add
__FILE__and__LINE__data to structlock_profile. Inlock_prof_init()I could look for non-unique lock names and mark those to be printed with the__FILE__and__LINE__data added to the names.Would you be fine with this approach?
I would be, but I'm afraid Andrew won't (as with any new uses of __LINE__). I wonder if __func__ or __FUNCTION__ could be used - the main question is how they behave when used outside of a function. If either would be NULL (rather than triggering a diagnostic), it might be usable here. Of course this would be fragile if just based on observed (rather than documented) behavior.
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 34 lines · Jan Beulich
Quoted 33 lines · Juergen Gross
Quoted 19 lines · Jan Beulich
Quoted 18 lines · Juergen Gross
Quoted 16 lines · Jan Beulich
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }if there's no other (transparent) way of disambiguating the names.
Quoted 7 lines
Quoted 3 lines
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and add "@<func>" to the lock profiling name. Is this okay?To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
I think I have found a way: I could add
__FILE__and__LINE__data to structlock_profile. Inlock_prof_init()I could look for non-unique lock names and mark those to be printed with the__FILE__and__LINE__data added to the names.
Quoted 8 lines
Quoted 1 line
Would you be fine with this approach?
I would be, but I'm afraid Andrew won't (as with any new uses of
__LINE__). I wonder if__func__or__FUNCTION__could be used - the main question is how they behave when used outside of a function. If either would be NULL (rather than triggering a diagnostic), it might be usable here. Of course this would be fragile if just based on observed (rather than documented) behavior.
With gcc 7.4.1 it fails:
/home/gross/xen/xen/include/xen/spinlock.h:80:41: error: ‘__func__’ is not defined outside of function scope [-Werror]
#define _LOCK_PROFILE(name) { 0, #name, __func__, &name, 0, 0, 0, 0, 0 }
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 44 lines · Juergen Gross
Quoted 34 lines · Jan Beulich
Quoted 33 lines · Juergen Gross
Quoted 19 lines · Jan Beulich
Quoted 18 lines · Juergen Gross
Quoted 16 lines · Jan Beulich
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }if there's no other (transparent) way of disambiguating the names.
Quoted 7 lines
Quoted 3 lines
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and add "@<func>" to the lock profiling name. Is this okay?To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
I think I have found a way: I could add
__FILE__and__LINE__data to structlock_profile. Inlock_prof_init()I could look for non-unique lock names and mark those to be printed with the__FILE__and__LINE__data added to the names.Quoted 8 lines
Quoted 1 line
Would you be fine with this approach?
I would be, but I'm afraid Andrew won't (as with any new uses of
__LINE__). I wonder if__func__or__FUNCTION__could be used - the main question is how they behave when used outside of a function. If either would be NULL (rather than triggering a diagnostic), it might be usable here. Of course this would be fragile if just based on observed (rather than documented) behavior.
Quoted 5 lines
With gcc 7.4.1 it fails:
/home/gross/xen/xen/include/xen/spinlock.h:80:41: error: ‘__func__’ is not defined outside of function scope [-Werror]
#define_LOCK_PROFILE(name){ 0, #name,__func__, &name, 0, 0, 0, 0, 0 }
Right, as I was afraid of. But __PRETTY_FUNCTION__ looks to be suitable (as per the gcc doc, and as per there being clear indications that clang also supports this).
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 47 lines · Jan Beulich
Quoted 46 lines · Juergen Gross
Quoted 34 lines · Jan Beulich
Quoted 33 lines · Juergen Gross
Quoted 19 lines · Jan Beulich
Quoted 18 lines · Juergen Gross
Quoted 16 lines · Jan Beulich
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }if there's no other (transparent) way of disambiguating the names.
Quoted 7 lines
Quoted 3 lines
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and add "@<func>" to the lock profiling name. Is this okay?To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
I think I have found a way: I could add
__FILE__and__LINE__data to structlock_profile. Inlock_prof_init()I could look for non-unique lock names and mark those to be printed with the__FILE__and__LINE__data added to the names.Quoted 8 lines
Quoted 1 line
Would you be fine with this approach?
I would be, but I'm afraid Andrew won't (as with any new uses of
__LINE__). I wonder if__func__or__FUNCTION__could be used - the main question is how they behave when used outside of a function. If either would be NULL (rather than triggering a diagnostic), it might be usable here. Of course this would be fragile if just based on observed (rather than documented) behavior.With gcc 7.4.1 it fails:
Quoted 7 lines
Quoted 3 lines
/home/gross/xen/xen/include/xen/spinlock.h:80:41: error: ‘__func__’ is not defined outside of function scope [-Werror]
#define_LOCK_PROFILE(name){ 0, #name,__func__, &name, 0, 0, 0, 0, 0 }Right, as I was afraid of. But
__PRETTY_FUNCTION__looks to be suitable (as per the gcc doc, and as per there being clear indications that clang also supports this).
Yes, this is working. Will modify the patch.
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
In reply to Jan BeulichQuoted 25 lines · Jan Beulich
Quoted 24 lines · Juergen Gross
Quoted 23 lines · Jan Beulich
Quoted 11 lines · Juergen Gross
Quoted 10 lines · Jan Beulich
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }Quoted 8 lines
Quoted 5 lines
so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }if there's no other (transparent) way of disambiguating the names.
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and add "@<func>" to the lock profiling name. Is this okay?To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
Quoted 7 lines
Quoted 6 lines
I think I have found a way: I could add
__FILE__and__LINE__data to structlock_profile. Inlock_prof_init()I could look for non-unique lock names and mark those to be printed with the__FILE__and__LINE__data added to the names.Would you be fine with this approach?
I would be, but I'm afraid Andrew won't (as with any new uses of
__LINE__).
The ok-ness of using __LINE__ is inversely proportional to the likelihood of developing a livepatch for this particular build of Xen, and what additional patching delta it would cause through unrelated changes.
We still have __LINE__ in a few cases in release builds because the utility has been deemed to outweigh the additional livepatch overhead. A specific example is x86_emulate(), where the extra patching delta is 0 given that it is a single function.
For stuff which wouldn't be active in production builds, and unlikely to be on the majority of debug builds, then it should be fine to use.
The one remaining problematic use of __LINE__ is in domain_crash() and it needs to disappear because it tends to have a massive unrelated-patch-delta for release builds.
~Andrew
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 31 lines · Andrew Cooper
Quoted 30 lines · Jan Beulich
Quoted 29 lines · Juergen Gross
Quoted 16 lines · Jan Beulich
Quoted 15 lines · Juergen Gross
Quoted 14 lines · Jan Beulich
Quoted 2 lines · Juergen Gross
In order to have unique names when doing lock profiling several local locks "lock" need to be renamed.
But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }Quoted 7 lines
Quoted 4 lines
Quoted 1 line
if there's no other (transparent) way of disambiguating the names.
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and add "@<func>" to the lock profiling name. Is this okay?To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
I think I have found a way: I could add
__FILE__and__LINE__data to structlock_profile. Inlock_prof_init()I could look for non-unique lock names and mark those to be printed with the__FILE__and__LINE__data added to the names.
Quoted 6 lines
Quoted 2 lines
Quoted 1 line
Would you be fine with this approach?
I would be, but I'm afraid Andrew won't (as with any new uses of
__LINE__).The ok-ness of using
__LINE__is inversely proportional to the likelihood of developing a livepatch for this particular build of Xen, and what additional patching delta it would cause through unrelated changes.
Not related to this patch, but to __LINE__ and livepatching: have you considered to use the "#line" directive to avoid unrelated diffs?
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 40 lines · Juergen Gross
Quoted 39 lines · Andrew Cooper
Quoted 37 lines · Jan Beulich
Quoted 28 lines · Juergen Gross
Quoted 19 lines · Jan Beulich
Quoted 18 lines · Juergen Gross
Quoted 17 lines · Jan Beulich
Quoted 3 lines · Juergen Gross
In order to have unique names when doing lock profiling several local
locks "lock" need to be renamed.But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }if there's no other (transparent) way of disambiguating the names.
Quoted 8 lines
Quoted 5 lines
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and
add "@<func>" to the lock profiling name. Is this okay?To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
Quoted 6 lines
I think I have found a way: I could add
__FILE__and__LINE__data to structlock_profile. Inlock_prof_init()I could look for non-unique lock names and mark those to be printed with the__FILE__and__LINE__data added to the names.Would you be fine with this approach?
I would be, but I'm afraid Andrew won't (as with any new uses of
__LINE__).
Quoted 7 lines
Quoted 4 lines
The ok-ness of using
__LINE__is inversely proportional to the likelihood of developing a livepatch for this particular build of Xen, and what additional patching delta it would cause through unrelated changes.Not related to this patch, but to
__LINE__and livepatching: have you considered to use the "#line" directive to avoid unrelated diffs?
There are ways to play with __LINE__, yes. #line was brought up in the original discussion.
As a thought experiment, how would you expect this to be used to simplify a livepatch?
~Andrew
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v3 5/5] xen: modify several static locks to unique names
Quoted 46 lines · Andrew Cooper
Quoted 45 lines · Juergen Gross
Quoted 43 lines · Andrew Cooper
Quoted 29 lines · Jan Beulich
Quoted 28 lines · Juergen Gross
Quoted 27 lines · Jan Beulich
Quoted 17 lines · Juergen Gross
Quoted 16 lines · Jan Beulich
Quoted 3 lines · Juergen Gross
In order to have unique names when doing lock profiling several local
locks "lock" need to be renamed.But these are all named simply "lock" for a good reason, including because they're all function scope symbols (and typically the functions are all sufficiently short). The issue stems from the dual use of "name" in
#define
_LOCK_PROFILE(name){ 0, #name, &name, 0, 0, 0, 0, 0 }so I'd rather suggest making this a derivation of a new
#define
_LOCK_PROFILE_NAME(lock, name){ 0, #name, &lock, 0, 0, 0, 0, 0 }Quoted 6 lines
Quoted 1 line
if there's no other (transparent) way of disambiguating the names.
This will require to use a different
DEFINE_SPINLOCK()variant, so e.g.DEFINE_SPINLOCK_LOCAL(), which will then include the needed "static" and
add "@<func>" to the lock profiling name. Is this okay?To be frank - not really. I dislike both, and would hence prefer to stick to what there is currently, until someone invents a transparent way to disambiguate these. I'm sorry for being unhelpful here.
Quoted 8 lines
Quoted 6 lines
I think I have found a way: I could add
__FILE__and__LINE__data to structlock_profile. Inlock_prof_init()I could look for non-unique lock names and mark those to be printed with the__FILE__and__LINE__data added to the names.Would you be fine with this approach?
I would be, but I'm afraid Andrew won't (as with any new uses of
__LINE__).The ok-ness of using
__LINE__is inversely proportional to the likelihood of developing a livepatch for this particular build of Xen, and what additional patching delta it would cause through unrelated changes.
Quoted 8 lines
Quoted 2 lines
Not related to this patch, but to
__LINE__and livepatching: have you considered to use the "#line" directive to avoid unrelated diffs?There are ways to play with
__LINE__, yes. #line was brought up in the original discussion.As a thought experiment, how would you expect this to be used to simplify a livepatch?
It should be rather strait forward to write a tool adding #line directives to a patch recovering the previous line numbers in the code following a modification which added or removed lines.
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel