Skip to contentSkip to conversation

powerpc: select ARCH_HAS_MEMBARRIER_SYNC_CORE

PATCH Superseded Nicholas Piggin 3 participants 5 messages Jul 15, 2020 – Jul 16, 2020

Subsystems (3)

v2 of 3 Superseded Latest is v3 →

lore
Share & tools
lore

Nicholas Piggin

author reviewer
#0 lore raw

PATCH v2 1/1 powerpc: select ARCH_HAS_MEMBARRIER_SYNC_CORE

powerpc return from interrupt and return from system call sequences are context synchronising.

---
v2: add more comments

 .../features/sched/membarrier-sync-core/arch-support.txt   | 4 ++--
 arch/powerpc/Kconfig                                       | 1 +
 arch/powerpc/include/asm/exception-64e.h                   | 6 +++++-
 arch/powerpc/include/asm/exception-64s.h                   | 7 +++++++
 arch/powerpc/kernel/entry_32.S                             | 6 ++++++
 5 files changed, 21 insertions(+), 3 deletions(-)
Documentation/features/sched/membarrier-sync-core/arch-support.txt +2 -2
index 8a521a622966..52ad74a25f54 100644
@@ -5,7 +5,7 @@
 #
 # Architecture requirements
 #
-# * arm/arm64
+# * arm/arm64/powerpc
 #
 # Rely on implicit context synchronization as a result of exception return
 # when returning from IPI handler, and when returning to user-space.
@@ -45,7 +45,7 @@
     |       nios2: | TODO |
     |    openrisc: | TODO |
     |      parisc: | TODO |
-    |     powerpc: | TODO |
+    |     powerpc: |  ok  |
     |       riscv: | TODO |
     |        s390: | TODO |
     |          sh: | TODO |
arch/powerpc/Kconfig +1 -0
index 9fa23eb320ff..920c4e3ca4ef 100644
@@ -131,6 +131,7 @@ config PPC
 	select ARCH_HAS_PTE_DEVMAP		if PPC_BOOK3S_64
 	select ARCH_HAS_PTE_SPECIAL
 	select ARCH_HAS_MEMBARRIER_CALLBACKS
+	select ARCH_HAS_MEMBARRIER_SYNC_CORE
 	select ARCH_HAS_SCALED_CPUTIME		if VIRT_CPU_ACCOUNTING_NATIVE && PPC_BOOK3S_64
 	select ARCH_HAS_STRICT_KERNEL_RWX	if (PPC32 && !HIBERNATION)
 	select ARCH_HAS_TICK_BROADCAST		if GENERIC_CLOCKEVENTS_BROADCAST
arch/powerpc/include/asm/exception-64e.h +5 -1
index 54a98ef7d7fe..071d7ccb830f 100644
@@ -204,7 +204,11 @@ exc_##label##_book3e:
 	LOAD_REG_ADDR(r3,interrupt_base_book3e);\
 	ori	r3,r3,vector_offset@l;		\
 	mtspr	SPRN_IVOR##vector_number,r3;
-
+/*
+ * powerpc relies on return from interrupt/syscall being context synchronising
+ * (which rfi is) to support ARCH_HAS_MEMBARRIER_SYNC_CORE without additional
+ * additional synchronisation instructions.
+ */
 #define RFI_TO_KERNEL							\
 	rfi
 
arch/powerpc/include/asm/exception-64s.h +7 -0
index 47bd4ea0837d..a4704f405e8d 100644
@@ -68,6 +68,13 @@
  *
  * The nop instructions allow us to insert one or more instructions to flush the
  * L1-D cache when returning to userspace or a guest.
+ *
+ * powerpc relies on return from interrupt/syscall being context synchronising
+ * (which hrfid, rfid, and rfscv are) to support ARCH_HAS_MEMBARRIER_SYNC_CORE
+ * without additional additional synchronisation instructions. soft-masked
+ * interrupt replay does not include a context-synchronising rfid, but those
+ * always return to kernel, the context sync is only required for IPIs which
+ * return to user.
  */
 #define RFI_FLUSH_SLOT							\
 	RFI_FLUSH_FIXUP_SECTION;					\
arch/powerpc/kernel/entry_32.S +6 -0
index 217ebdf5b00b..23bb7352e7c3 100644
@@ -35,6 +35,12 @@
 
 #include "head_32.h"
 
+/*
+ * powerpc relies on return from interrupt/syscall being context synchronising
+ * (which rfi is) to support ARCH_HAS_MEMBARRIER_SYNC_CORE without additional
+ * additional synchronisation instructions.
+ */
+
 /*
  * Align to 4k in order to ensure that all functions modyfing srr0/srr1
  * fit into one page in order to not encounter a TLB miss between the
Signature · 1 line
2.23.0
#1 lore raw

arch/powerpc/include/asm/exception-64e.h exc_##label##_book3e: · line 210

Quoted 13 lines · Nicholas Piggin
arch/powerpc/include/asm/exception-64e.h +4 -1
index 54a98ef7d7fe..071d7ccb830f 100644
@@ -204,7 +204,11 @@ exc_##label##_book3e:
 	LOAD_REG_ADDR(r3,interrupt_base_book3e);\
 	ori	r3,r3,vector_offset@l;		\
 	mtspr	SPRN_IVOR##vector_number,r3;
-
+/*
+ * powerpc relies on return from interrupt/syscall being context synchronising
+ * (which rfi is) to support ARCH_HAS_MEMBARRIER_SYNC_CORE without additional
+ * additional synchronisation instructions.

s/additonal//

Andreas.

Signature · 3 lines
Andreas Schwab, [email protected]
GPG Key fingerprint = 7578 EB47 D4E5 4D69 2510  2552 DF73 E780 A9DA AEC1
"And now for something completely different."
author reviewer
#2 lore raw

Excerpts from Andreas Schwab's message of July 15, 2020 8:08 pm:

Quoted 17 lines · Nicholas Piggin
Quoted 6 lines
arch/powerpc/include/asm/exception-64e.h +0 -0
index 54a98ef7d7fe..071d7ccb830f 100644

arch/powerpc/include/asm/exception-64e.h exc_##label##_book3e: · line 210

Quoted 8 lines
arch/powerpc/include/asm/exception-64e.h +4 -1
@@ -204,7 +204,11 @@ exc_##label##_book3e:
 	LOAD_REG_ADDR(r3,interrupt_base_book3e);\
 	ori	r3,r3,vector_offset@l;		\
 	mtspr	SPRN_IVOR##vector_number,r3;
-
+/*
+ * powerpc relies on return from interrupt/syscall being context synchronising
+ * (which rfi is) to support ARCH_HAS_MEMBARRIER_SYNC_CORE without additional
+ * additional synchronisation instructions.

s/additonal//

Will fix in a respin.

Thanks,
Nick

#3 lore raw
In reply to Nicholas Piggin

----- On Jul 15, 2020, at 5:48 AM, Nicholas Piggin [email protected] wrote:
[...]

Quoted 7 lines

index 47bd4ea0837d..a4704f405e8d 100644

arch/powerpc/include/asm/exception-64s.h +0 -0
@@ -68,6 +68,13 @@
 *
 * The nop instructions allow us to insert one or more instructions to flush the
 * L1-D cache when returning to userspace or a guest.

arch/powerpc/include/asm/exception-64s.h line 80

Quoted 10 lines
arch/powerpc/include/asm/exception-64s.h +7 -0
+ *
+ * powerpc relies on return from interrupt/syscall being context synchronising
+ * (which hrfid, rfid, and rfscv are) to support ARCH_HAS_MEMBARRIER_SYNC_CORE
+ * without additional additional synchronisation instructions. soft-masked
+ * interrupt replay does not include a context-synchronising rfid, but those
+ * always return to kernel, the context sync is only required for IPIs which
+ * return to user.
 */
 #define RFI_FLUSH_SLOT							\
 	RFI_FLUSH_FIXUP_SECTION;					\

I suspect the statement "the context sync is only required for IPIs which return to user." is misleading.

As I recall that we need more than just context sync after IPI. We need context sync in return path of any trap/interrupt/system call which returns to user-space, else we'd need to add the proper core serializing barriers in the scheduler, as we had to do for lazy tlb on x86.

Or am I missing something ?

Thanks,

Mathieu

Signature · 3 lines
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com
author reviewer
#4 lore raw

Excerpts from Mathieu Desnoyers's message of July 15, 2020 10:27 pm:

Quoted 23 lines

----- On Jul 15, 2020, at 5:48 AM, Nicholas Piggin [email protected] wrote:
[...]

Quoted 7 lines

index 47bd4ea0837d..a4704f405e8d 100644

arch/powerpc/include/asm/exception-64s.h +0 -0
@@ -68,6 +68,13 @@
 *
 * The nop instructions allow us to insert one or more instructions to flush the
 * L1-D cache when returning to userspace or a guest.

arch/powerpc/include/asm/exception-64s.h line 80

Quoted 10 lines
arch/powerpc/include/asm/exception-64s.h +7 -0
+ *
+ * powerpc relies on return from interrupt/syscall being context synchronising
+ * (which hrfid, rfid, and rfscv are) to support ARCH_HAS_MEMBARRIER_SYNC_CORE
+ * without additional additional synchronisation instructions. soft-masked
+ * interrupt replay does not include a context-synchronising rfid, but those
+ * always return to kernel, the context sync is only required for IPIs which
+ * return to user.
 */
 #define RFI_FLUSH_SLOT							\
 	RFI_FLUSH_FIXUP_SECTION;					\

I suspect the statement "the context sync is only required for IPIs which return to user." is misleading.

Quoted 6 lines

As I recall that we need more than just context sync after IPI. We need context sync in return path of any trap/interrupt/system call which returns to user-space, else we'd need to add the proper core serializing barriers in the scheduler, as we had to do for lazy tlb on x86.

Or am I missing something ?

Maybe ambiguous wording. For IPIs, the context synch is only required for those which return to user. Other things also require context sync.

I will try to improve it.

Thanks,
Nick