Skip to contentSkip to conversation

arch/arc/kernel/smp.c:267:18: sparse: sparse: cast removes address space '__percpu' of expression

DISCUSSION kernel test robot 2 participants 3 messages Sep 25, 2024 – Oct 4, 2024 Synopsys ARC architecture

lore
Share & tools
lore

kernel test robot

author build bot
#0 lore raw
tree:   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
head:   684a64bf32b6e488004e0ad7f0d7e922798f65b6
commit: f2519d4d4fc4d36f2b58c5614357de9f5b4032fc ARC: Emulate one-byte cmpxchg
date:   12 days ago
config: arc-randconfig-r123-20240925 (https://download.01.org/0day-ci/archive/20240925/[email protected]/config)
compiler: arc-elf-gcc (GCC) 13.2.0
reproduce: (https://download.01.org/0day-ci/archive/20240925/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

sparse warnings: (new ones prefixed by >>)
arch/arc/kernel/smp.c:252:48: sparse: sparse: incorrect type in initializer (different address spaces) @@ expected unsigned long [noderef] __percpu *ipi_data_ptr @@ got unsigned long * @@
arch/arc/kernel/smp.c:252:48: sparse: expected unsigned long [noderef] __percpu *ipi_data_ptr
arch/arc/kernel/smp.c:252:48: sparse: got unsigned long *
arch/arc/kernel/smp.c:267:18: sparse: sparse: incorrect type in argument 1 (different address spaces) @@ expected void const volatile *v @@ got unsigned long [noderef] __percpu *__ai_ptr @@
arch/arc/kernel/smp.c:267:18: sparse: expected void const volatile *v
arch/arc/kernel/smp.c:267:18: sparse: got unsigned long [noderef] __percpu *__ai_ptr

>> arch/arc/kernel/smp.c:267:18: sparse: sparse: cast removes address space '__percpu' of expression
>> arch/arc/kernel/smp.c:267:18: sparse: sparse: cast removes address space '__percpu' of expression

arch/arc/kernel/smp.c:401:72: sparse: sparse: incorrect type in argument 4 (different address spaces) @@ expected void [noderef] __percpu *percpu_dev_id @@ got int *dev @@
arch/arc/kernel/smp.c:401:72: sparse: expected void [noderef] __percpu *percpu_dev_id
arch/arc/kernel/smp.c:401:72: sparse: got int *dev
arch/arc/kernel/smp.c:265:30: sparse: sparse: dereference of noderef expression

vim +/__percpu +267 arch/arc/kernel/smp.c

41195d236e8445 Vineet Gupta    2013-01-18  249  
ddf84433f411b6 Vineet Gupta    2013-11-25  250  static void ipi_send_msg_one(int cpu, enum ipi_msg_type msg)
41195d236e8445 Vineet Gupta    2013-01-18  251  {
f2a4aa5646687f Vineet Gupta    2013-11-26  252  	unsigned long __percpu *ipi_data_ptr = per_cpu_ptr(&ipi_data, cpu);
d8e8c7dda11f5d Vineet Gupta    2013-11-28  253  	unsigned long old, new;
41195d236e8445 Vineet Gupta    2013-01-18  254  	unsigned long flags;
41195d236e8445 Vineet Gupta    2013-01-18  255  
f2a4aa5646687f Vineet Gupta    2013-11-26  256  	pr_debug("%d Sending msg [%d] to %d\n", smp_processor_id(), msg, cpu);
f2a4aa5646687f Vineet Gupta    2013-11-26  257  
41195d236e8445 Vineet Gupta    2013-01-18  258  	local_irq_save(flags);
41195d236e8445 Vineet Gupta    2013-01-18  259  
d8e8c7dda11f5d Vineet Gupta    2013-11-28  260  	/*
d8e8c7dda11f5d Vineet Gupta    2013-11-28  261  	 * Atomically write new msg bit (in case others are writing too),
d8e8c7dda11f5d Vineet Gupta    2013-11-28  262  	 * and read back old value
d8e8c7dda11f5d Vineet Gupta    2013-11-28  263  	 */
d8e8c7dda11f5d Vineet Gupta    2013-11-28  264  	do {
c6ed4d84a2c49d Bang Li         2022-03-19  265  		new = old = *ipi_data_ptr;
d8e8c7dda11f5d Vineet Gupta    2013-11-28  266  		new |= 1U << msg;
d8e8c7dda11f5d Vineet Gupta    2013-11-28 @267  	} while (cmpxchg(ipi_data_ptr, old, new) != old);
41195d236e8445 Vineet Gupta    2013-01-18  268  
d8e8c7dda11f5d Vineet Gupta    2013-11-28  269  	/*
d8e8c7dda11f5d Vineet Gupta    2013-11-28  270  	 * Call the platform specific IPI kick function, but avoid if possible:
d8e8c7dda11f5d Vineet Gupta    2013-11-28  271  	 * Only do so if there's no pending msg from other concurrent sender(s).
82a423053eb3cf Changcheng Deng 2021-08-14  272  	 * Otherwise, receiver will see this msg as well when it takes the
d8e8c7dda11f5d Vineet Gupta    2013-11-28  273  	 * IPI corresponding to that msg. This is true, even if it is already in
d8e8c7dda11f5d Vineet Gupta    2013-11-28  274  	 * IPI handler, because !@old means it has not yet dequeued the msg(s)
d8e8c7dda11f5d Vineet Gupta    2013-11-28  275  	 * so @new msg can be a free-loader
d8e8c7dda11f5d Vineet Gupta    2013-11-28  276  	 */
d8e8c7dda11f5d Vineet Gupta    2013-11-28  277  	if (plat_smp_ops.ipi_send && !old)
ddf84433f411b6 Vineet Gupta    2013-11-25  278  		plat_smp_ops.ipi_send(cpu);
41195d236e8445 Vineet Gupta    2013-01-18  279  
41195d236e8445 Vineet Gupta    2013-01-18  280  	local_irq_restore(flags);
41195d236e8445 Vineet Gupta    2013-01-18  281  }
41195d236e8445 Vineet Gupta    2013-01-18  282  

:::::: The code at line 267 was first introduced by commit
:::::: d8e8c7dda11f5d5cf90495f2e89d917a83509bc0 ARC: [SMP] optimize IPI send and receive

:::::: TO: Vineet Gupta <[email protected]>
:::::: CC: Vineet Gupta <[email protected]>

Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
#1 lore raw
Quoted 13 lines · kernel test robot
tree:   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
head:   684a64bf32b6e488004e0ad7f0d7e922798f65b6
commit: f2519d4d4fc4d36f2b58c5614357de9f5b4032fc ARC: Emulate one-byte cmpxchg
date:   12 days ago
config: arc-randconfig-r123-20240925 (https://download.01.org/0day-ci/archive/20240925/[email protected]/config)
compiler: arc-elf-gcc (GCC) 13.2.0
reproduce: (https://download.01.org/0day-ci/archive/20240925/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

Quoted 9 lines

sparse warnings: (new ones prefixed by >>)
arch/arc/kernel/smp.c:252:48: sparse: sparse: incorrect type in initializer (different address spaces) @@ expected unsigned long [noderef] __percpu *ipi_data_ptr @@ got unsigned long * @@
arch/arc/kernel/smp.c:252:48: sparse: expected unsigned long [noderef] __percpu *ipi_data_ptr
arch/arc/kernel/smp.c:252:48: sparse: got unsigned long *
arch/arc/kernel/smp.c:267:18: sparse: sparse: incorrect type in argument 1 (different address spaces) @@ expected void const volatile *v @@ got unsigned long [noderef] __percpu *__ai_ptr @@
arch/arc/kernel/smp.c:267:18: sparse: expected void const volatile *v
arch/arc/kernel/smp.c:267:18: sparse: got unsigned long [noderef] __percpu *__ai_ptr

Quoted 2 lines
Quoted 2 lines

arch/arc/kernel/smp.c:267:18: sparse: sparse: cast removes address space '__percpu' of expression arch/arc/kernel/smp.c:267:18: sparse: sparse: cast removes address space '__percpu' of expression

I could "fix" this using __force. Is that a reasonable approach?

If I don't hear otherwise, I will take that approach.

Thanx, Paul

Quoted 51 lines

arch/arc/kernel/smp.c:401:72: sparse: sparse: incorrect type in argument 4 (different address spaces) @@ expected void [noderef] __percpu *percpu_dev_id @@ got int *dev @@
arch/arc/kernel/smp.c:401:72: sparse: expected void [noderef] __percpu *percpu_dev_id
arch/arc/kernel/smp.c:401:72: sparse: got int *dev
arch/arc/kernel/smp.c:265:30: sparse: sparse: dereference of noderef expression

vim +/__percpu +267 arch/arc/kernel/smp.c

41195d236e8445 Vineet Gupta    2013-01-18  249  
ddf84433f411b6 Vineet Gupta    2013-11-25  250  static void ipi_send_msg_one(int cpu, enum ipi_msg_type msg)
41195d236e8445 Vineet Gupta    2013-01-18  251  {
f2a4aa5646687f Vineet Gupta    2013-11-26  252  	unsigned long __percpu *ipi_data_ptr = per_cpu_ptr(&ipi_data, cpu);
d8e8c7dda11f5d Vineet Gupta    2013-11-28  253  	unsigned long old, new;
41195d236e8445 Vineet Gupta    2013-01-18  254  	unsigned long flags;
41195d236e8445 Vineet Gupta    2013-01-18  255  
f2a4aa5646687f Vineet Gupta    2013-11-26  256  	pr_debug("%d Sending msg [%d] to %d\n", smp_processor_id(), msg, cpu);
f2a4aa5646687f Vineet Gupta    2013-11-26  257  
41195d236e8445 Vineet Gupta    2013-01-18  258  	local_irq_save(flags);
41195d236e8445 Vineet Gupta    2013-01-18  259  
d8e8c7dda11f5d Vineet Gupta    2013-11-28  260  	/*
d8e8c7dda11f5d Vineet Gupta    2013-11-28  261  	 * Atomically write new msg bit (in case others are writing too),
d8e8c7dda11f5d Vineet Gupta    2013-11-28  262  	 * and read back old value
d8e8c7dda11f5d Vineet Gupta    2013-11-28  263  	 */
d8e8c7dda11f5d Vineet Gupta    2013-11-28  264  	do {
c6ed4d84a2c49d Bang Li         2022-03-19  265  		new = old = *ipi_data_ptr;
d8e8c7dda11f5d Vineet Gupta    2013-11-28  266  		new |= 1U << msg;
d8e8c7dda11f5d Vineet Gupta    2013-11-28 @267  	} while (cmpxchg(ipi_data_ptr, old, new) != old);
41195d236e8445 Vineet Gupta    2013-01-18  268  
d8e8c7dda11f5d Vineet Gupta    2013-11-28  269  	/*
d8e8c7dda11f5d Vineet Gupta    2013-11-28  270  	 * Call the platform specific IPI kick function, but avoid if possible:
d8e8c7dda11f5d Vineet Gupta    2013-11-28  271  	 * Only do so if there's no pending msg from other concurrent sender(s).
82a423053eb3cf Changcheng Deng 2021-08-14  272  	 * Otherwise, receiver will see this msg as well when it takes the
d8e8c7dda11f5d Vineet Gupta    2013-11-28  273  	 * IPI corresponding to that msg. This is true, even if it is already in
d8e8c7dda11f5d Vineet Gupta    2013-11-28  274  	 * IPI handler, because !@old means it has not yet dequeued the msg(s)
d8e8c7dda11f5d Vineet Gupta    2013-11-28  275  	 * so @new msg can be a free-loader
d8e8c7dda11f5d Vineet Gupta    2013-11-28  276  	 */
d8e8c7dda11f5d Vineet Gupta    2013-11-28  277  	if (plat_smp_ops.ipi_send && !old)
ddf84433f411b6 Vineet Gupta    2013-11-25  278  		plat_smp_ops.ipi_send(cpu);
41195d236e8445 Vineet Gupta    2013-01-18  279  
41195d236e8445 Vineet Gupta    2013-01-18  280  	local_irq_restore(flags);
41195d236e8445 Vineet Gupta    2013-01-18  281  }
41195d236e8445 Vineet Gupta    2013-01-18  282  

:::::: The code at line 267 was first introduced by commit
:::::: d8e8c7dda11f5d5cf90495f2e89d917a83509bc0 ARC: [SMP] optimize IPI send and receive

:::::: TO: Vineet Gupta <[email protected]>
:::::: CC: Vineet Gupta <[email protected]>

Signature · 2 lines
0-DAY CI Kernel Test Service
https://github.com/intel/lkp-tests/wiki
#2 lore raw
Quoted 21 lines · Paul E. McKenney
Quoted 20 lines · kernel test robot
tree:   https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
head:   684a64bf32b6e488004e0ad7f0d7e922798f65b6
commit: f2519d4d4fc4d36f2b58c5614357de9f5b4032fc ARC: Emulate one-byte cmpxchg
date:   12 days ago
config: arc-randconfig-r123-20240925 (https://download.01.org/0day-ci/archive/20240925/[email protected]/config)
compiler: arc-elf-gcc (GCC) 13.2.0
reproduce: (https://download.01.org/0day-ci/archive/20240925/[email protected]/reproduce)

If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/

sparse warnings: (new ones prefixed by >>)
arch/arc/kernel/smp.c:252:48: sparse: sparse: incorrect type in initializer (different address spaces) @@ expected unsigned long [noderef] __percpu *ipi_data_ptr @@ got unsigned long * @@
arch/arc/kernel/smp.c:252:48: sparse: expected unsigned long [noderef] __percpu *ipi_data_ptr
arch/arc/kernel/smp.c:252:48: sparse: got unsigned long *
arch/arc/kernel/smp.c:267:18: sparse: sparse: incorrect type in argument 1 (different address spaces) @@ expected void const volatile *v @@ got unsigned long [noderef] __percpu *__ai_ptr @@
arch/arc/kernel/smp.c:267:18: sparse: expected void const volatile *v
arch/arc/kernel/smp.c:267:18: sparse: got unsigned long [noderef] __percpu *__ai_ptr

Quoted 6 lines
Quoted 2 lines
Quoted 2 lines
Quoted 2 lines

arch/arc/kernel/smp.c:267:18: sparse: sparse: cast removes address space '__percpu' of expression arch/arc/kernel/smp.c:267:18: sparse: sparse: cast removes address space '__percpu' of expression

I could "fix" this using __force. Is that a reasonable approach?

If I don't hear otherwise, I will take that approach.

Having not heard otherwise, does the following fix this issue for you?

Thanx, Paul

------------------------------------------------------------------------

arch/arc/include/asm/cmpxchg.h +1 -1
index 58045c8983404..76f43db0890fc 100644
@@ -48,7 +48,7 @@
 									\
 	switch(sizeof((_p_))) {						\
 	case 1:								\
-		_prev_ = (__typeof__(*(ptr)))cmpxchg_emu_u8((volatile u8 *)_p_, (uintptr_t)_o_, (uintptr_t)_n_);	\
+		_prev_ = (__typeof__(*(ptr)))cmpxchg_emu_u8((volatile u8 *__force)_p_, (uintptr_t)_o_, (uintptr_t)_n_);	\
 		break;							\
 	case 4:								\
 		_prev_ = __cmpxchg(_p_, _o_, _n_);			\