]> www.infradead.org Git - users/jedix/linux-maple.git/commit
drivers/base/memory: Avoid overhead from for_each_present_section_nr()
authorGavin Shan <gshan@redhat.com>
Thu, 10 Apr 2025 12:51:10 +0000 (22:51 +1000)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Tue, 15 Apr 2025 16:19:49 +0000 (18:19 +0200)
commitb9792abb76ae1649080b8d48092c52e24c7bbdc2
treefe8d5b46def48e74f42b836dbb5227ebf9b134c9
parentbc2c46426f2d95e58c82f394531afdd034c8706c
drivers/base/memory: Avoid overhead from for_each_present_section_nr()

for_each_present_section_nr() was introduced to add_boot_memory_block()
by commit 61659efdb35c ("drivers/base/memory: improve add_boot_memory_block()").
It causes unnecessary overhead when the present sections are really
sparse. next_present_section_nr() called by the macro to find the next
present section, which is far away from the spanning sections in the
specified block. Too much time consumed by next_present_section_nr()
in this case, which can lead to softlockup as observed by Aditya Gupta
on IBM Power10 machine.

  watchdog: BUG: soft lockup - CPU#248 stuck for 22s! [swapper/248:1]
  Modules linked in:
  CPU: 248 UID: 0 PID: 1 Comm: swapper/248 Not tainted 6.15.0-rc1-next-20250408 #1 VOLUNTARY
  Hardware name: 9105-22A POWER10 (raw) 0x800200 opal:v7.1-107-gfda75d121942 PowerNV
  NIP:  c00000000209218c LR: c000000002092204 CTR: 0000000000000000
  REGS: c00040000418fa30 TRAP: 0900   Not tainted  (6.15.0-rc1-next-20250408)
  MSR:  9000000002009033 <SF,HV,VEC,EE,ME,IR,DR,RI,LE>  CR: 28000428  XER: 00000000
  CFAR: 0000000000000000 IRQMASK: 0
  GPR00: c000000002092204 c00040000418fcd0 c000000001b08100 0000000000000040
  GPR04: 0000000000013e00 c000c03ffebabb00 0000000000c03fff c000400fff587f80
  GPR08: 0000000000000000 00000000001196f7 0000000000000000 0000000028000428
  GPR12: 0000000000000000 c000000002e80000 c00000000001007c 0000000000000000
  GPR16: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
  GPR20: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
  GPR24: 0000000000000000 0000000000000000 0000000000000000 0000000000000000
  GPR28: c000000002df7f70 0000000000013dc0 c0000000011dd898 0000000008000000
  NIP [c00000000209218c] memory_dev_init+0x114/0x1e0
  LR [c000000002092204] memory_dev_init+0x18c/0x1e0
  Call Trace:
  [c00040000418fcd0] [c000000002092204] memory_dev_init+0x18c/0x1e0 (unreliable)
  [c00040000418fd50] [c000000002091348] driver_init+0x78/0xa4
  [c00040000418fd70] [c0000000020063ac] kernel_init_freeable+0x22c/0x370
  [c00040000418fde0] [c0000000000100a8] kernel_init+0x34/0x25c
  [c00040000418fe50] [c00000000000cd94] ret_from_kernel_user_thread+0x14/0x1c

Avoid the overhead by folding for_each_present_section_nr() to the outer
loop. add_boot_memory_block() is dropped after that.

Fixes: 61659efdb35c ("drivers/base/memory: improve add_boot_memory_block()")
Closes: https://lore.kernel.org/linux-mm/20250409180344.477916-1-adityag@linux.ibm.com
Reported-by: Aditya Gupta <adityag@linux.ibm.com>
Signed-off-by: Gavin Shan <gshan@redhat.com>
Acked-by: Oscar Salvador <osalvador@suse.de>
Tested-by: Aditya Gupta <adityag@linux.ibm.com>
Acked-by: David Hildenbrand <david@redhat.com>
Link: https://lore.kernel.org/r/20250410125110.1232329-1-gshan@redhat.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
drivers/base/memory.c