* [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order
@ 2025-03-21 12:57 Philippe Mathieu-Daudé
2025-03-21 12:57 ` [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO Philippe Mathieu-Daudé
` (4 more replies)
0 siblings, 5 replies; 19+ messages in thread
From: Philippe Mathieu-Daudé @ 2025-03-21 12:57 UTC (permalink / raw)
To: Richard Henderson, qemu-devel
Cc: Paolo Bonzini, Pierrick Bouvier, Alex Bennée,
Anton Johansson, Philippe Mathieu-Daudé
Hi,
In this series we replace the TCG_GUEST_DEFAULT_MO definition
from "cpu-param.h" by a 'guest_default_memory_order' field in
TCGCPUOps.
Since tcg_req_mo() now accesses tcg_ctx, this impact the
cpu_req_mo() calls in accel/tcg/{cputlb,user-exec}.c.
The long term goal is to be able to use targets with distinct
guest memory order restrictions.
Philippe Mathieu-Daudé (4):
tcg: Always define TCG_GUEST_DEFAULT_MO
tcg: Simplify tcg_req_mo() macro
tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo
tcg: Define guest_default_memory_order in TCGCPUOps
accel/tcg/internal-target.h | 13 ++-----------
include/accel/tcg/cpu-ops.h | 8 ++++++++
target/alpha/cpu-param.h | 3 ---
target/arm/cpu-param.h | 3 ---
target/avr/cpu-param.h | 2 --
target/hppa/cpu-param.h | 8 --------
target/i386/cpu-param.h | 3 ---
target/loongarch/cpu-param.h | 2 --
target/microblaze/cpu-param.h | 3 ---
target/mips/cpu-param.h | 2 --
target/openrisc/cpu-param.h | 2 --
target/ppc/cpu-param.h | 2 --
target/riscv/cpu-param.h | 2 --
target/s390x/cpu-param.h | 6 ------
target/sparc/cpu-param.h | 23 -----------------------
target/xtensa/cpu-param.h | 3 ---
accel/tcg/tcg-all.c | 3 ---
accel/tcg/translate-all.c | 6 +-----
target/alpha/cpu.c | 3 +++
target/arm/cpu.c | 3 +++
target/arm/tcg/cpu-v7m.c | 3 +++
target/avr/cpu.c | 1 +
target/hexagon/cpu.c | 2 ++
target/hppa/cpu.c | 8 ++++++++
target/i386/tcg/tcg-cpu.c | 5 +++++
target/loongarch/cpu.c | 2 ++
target/m68k/cpu.c | 3 +++
target/microblaze/cpu.c | 3 +++
target/mips/cpu.c | 2 ++
target/openrisc/cpu.c | 2 ++
target/ppc/cpu_init.c | 2 ++
target/riscv/tcg/tcg-cpu.c | 2 ++
target/rx/cpu.c | 3 +++
target/s390x/cpu.c | 6 ++++++
target/sh4/cpu.c | 3 +++
target/sparc/cpu.c | 23 +++++++++++++++++++++++
target/tricore/cpu.c | 2 ++
target/xtensa/cpu.c | 3 +++
38 files changed, 92 insertions(+), 83 deletions(-)
--
2.47.1
^ permalink raw reply [flat|nested] 19+ messages in thread
* [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO
2025-03-21 12:57 [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order Philippe Mathieu-Daudé
@ 2025-03-21 12:57 ` Philippe Mathieu-Daudé
2025-03-21 14:39 ` Anton Johansson via
` (2 more replies)
2025-03-21 12:57 ` [PATCH-for-10.1 2/4] tcg: Simplify tcg_req_mo() macro Philippe Mathieu-Daudé
` (3 subsequent siblings)
4 siblings, 3 replies; 19+ messages in thread
From: Philippe Mathieu-Daudé @ 2025-03-21 12:57 UTC (permalink / raw)
To: Richard Henderson, qemu-devel
Cc: Paolo Bonzini, Pierrick Bouvier, Alex Bennée,
Anton Johansson, Philippe Mathieu-Daudé
We only require the TCG_GUEST_DEFAULT_MO for MTTCG-enabled
frontends, otherwise we use a default value of TCG_MO_ALL.
In order to simplify, require the definition for all targets,
defining it for hexagon, m68k, rx, sh4 and tricore.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
---
target/hexagon/cpu-param.h | 3 +++
target/m68k/cpu-param.h | 3 +++
target/rx/cpu-param.h | 3 +++
target/sh4/cpu-param.h | 3 +++
target/tricore/cpu-param.h | 3 +++
accel/tcg/translate-all.c | 4 ----
6 files changed, 15 insertions(+), 4 deletions(-)
diff --git a/target/hexagon/cpu-param.h b/target/hexagon/cpu-param.h
index 45ee7b46409..2d57ea6caf9 100644
--- a/target/hexagon/cpu-param.h
+++ b/target/hexagon/cpu-param.h
@@ -23,4 +23,7 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 36
#define TARGET_VIRT_ADDR_SPACE_BITS 32
+/* MTTCG not yet supported: require strict ordering */
+#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
+
#endif
diff --git a/target/m68k/cpu-param.h b/target/m68k/cpu-param.h
index 7afbf6d302d..1a909eaa13e 100644
--- a/target/m68k/cpu-param.h
+++ b/target/m68k/cpu-param.h
@@ -17,4 +17,7 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 32
#define TARGET_VIRT_ADDR_SPACE_BITS 32
+/* MTTCG not yet supported: require strict ordering */
+#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
+
#endif
diff --git a/target/rx/cpu-param.h b/target/rx/cpu-param.h
index ef1970a09e9..2ce199164d7 100644
--- a/target/rx/cpu-param.h
+++ b/target/rx/cpu-param.h
@@ -24,4 +24,7 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 32
#define TARGET_VIRT_ADDR_SPACE_BITS 32
+/* MTTCG not yet supported: require strict ordering */
+#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
+
#endif
diff --git a/target/sh4/cpu-param.h b/target/sh4/cpu-param.h
index 2b6e11dd0ac..1bc90d4695e 100644
--- a/target/sh4/cpu-param.h
+++ b/target/sh4/cpu-param.h
@@ -16,4 +16,7 @@
# define TARGET_VIRT_ADDR_SPACE_BITS 32
#endif
+/* MTTCG not yet supported: require strict ordering */
+#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
+
#endif
diff --git a/target/tricore/cpu-param.h b/target/tricore/cpu-param.h
index 790242ef3d2..923459370cc 100644
--- a/target/tricore/cpu-param.h
+++ b/target/tricore/cpu-param.h
@@ -12,4 +12,7 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 32
#define TARGET_VIRT_ADDR_SPACE_BITS 32
+/* MTTCG not yet supported: require strict ordering */
+#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
+
#endif
diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
index 82bc16bd535..fb9f83dbba3 100644
--- a/accel/tcg/translate-all.c
+++ b/accel/tcg/translate-all.c
@@ -349,11 +349,7 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
tcg_ctx->tlb_dyn_max_bits = CPU_TLB_DYN_MAX_BITS;
#endif
tcg_ctx->insn_start_words = TARGET_INSN_START_WORDS;
-#ifdef TCG_GUEST_DEFAULT_MO
tcg_ctx->guest_mo = TCG_GUEST_DEFAULT_MO;
-#else
- tcg_ctx->guest_mo = TCG_MO_ALL;
-#endif
restart_translate:
trace_translate_block(tb, pc, tb->tc.ptr);
--
2.47.1
^ permalink raw reply related [flat|nested] 19+ messages in thread
* [PATCH-for-10.1 2/4] tcg: Simplify tcg_req_mo() macro
2025-03-21 12:57 [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order Philippe Mathieu-Daudé
2025-03-21 12:57 ` [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO Philippe Mathieu-Daudé
@ 2025-03-21 12:57 ` Philippe Mathieu-Daudé
2025-03-21 14:40 ` Anton Johansson via
` (2 more replies)
2025-03-21 12:57 ` [RFC PATCH-for-10.1 3/4] tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo Philippe Mathieu-Daudé
` (2 subsequent siblings)
4 siblings, 3 replies; 19+ messages in thread
From: Philippe Mathieu-Daudé @ 2025-03-21 12:57 UTC (permalink / raw)
To: Richard Henderson, qemu-devel
Cc: Paolo Bonzini, Pierrick Bouvier, Alex Bennée,
Anton Johansson, Philippe Mathieu-Daudé
Now that TCG_GUEST_DEFAULT_MO is always defined,
simplify the tcg_req_mo() macro.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
---
accel/tcg/internal-target.h | 9 +--------
accel/tcg/tcg-all.c | 3 ---
2 files changed, 1 insertion(+), 11 deletions(-)
diff --git a/accel/tcg/internal-target.h b/accel/tcg/internal-target.h
index 2cdf11c905e..1cb35dba99e 100644
--- a/accel/tcg/internal-target.h
+++ b/accel/tcg/internal-target.h
@@ -50,17 +50,10 @@ G_NORETURN void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr);
* memory ordering vs the host memory ordering. A non-zero
* result indicates that some barrier is required.
*
- * If TCG_GUEST_DEFAULT_MO is not defined, assume that the
- * guest requires strict ordering.
- *
* This is a macro so that it's constant even without optimization.
*/
-#ifdef TCG_GUEST_DEFAULT_MO
-# define tcg_req_mo(type) \
+#define tcg_req_mo(type) \
((type) & TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO)
-#else
-# define tcg_req_mo(type) ((type) & ~TCG_TARGET_DEFAULT_MO)
-#endif
/**
* cpu_req_mo:
diff --git a/accel/tcg/tcg-all.c b/accel/tcg/tcg-all.c
index c1a30b01219..cb632cc8cc7 100644
--- a/accel/tcg/tcg-all.c
+++ b/accel/tcg/tcg-all.c
@@ -77,9 +77,6 @@ static bool default_mttcg_enabled(void)
return false;
}
#ifdef TARGET_SUPPORTS_MTTCG
-# ifndef TCG_GUEST_DEFAULT_MO
-# error "TARGET_SUPPORTS_MTTCG without TCG_GUEST_DEFAULT_MO"
-# endif
return true;
#else
return false;
--
2.47.1
^ permalink raw reply related [flat|nested] 19+ messages in thread
* [RFC PATCH-for-10.1 3/4] tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo
2025-03-21 12:57 [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order Philippe Mathieu-Daudé
2025-03-21 12:57 ` [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO Philippe Mathieu-Daudé
2025-03-21 12:57 ` [PATCH-for-10.1 2/4] tcg: Simplify tcg_req_mo() macro Philippe Mathieu-Daudé
@ 2025-03-21 12:57 ` Philippe Mathieu-Daudé
2025-03-21 14:39 ` Anton Johansson via
2025-03-21 15:16 ` Richard Henderson
2025-03-21 12:57 ` [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps Philippe Mathieu-Daudé
2025-03-21 14:40 ` [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order Anton Johansson via
4 siblings, 2 replies; 19+ messages in thread
From: Philippe Mathieu-Daudé @ 2025-03-21 12:57 UTC (permalink / raw)
To: Richard Henderson, qemu-devel
Cc: Paolo Bonzini, Pierrick Bouvier, Alex Bennée,
Anton Johansson, Philippe Mathieu-Daudé
In order to use TCG with multiple targets, replace the
compile time use of TCG_GUEST_DEFAULT_MO by a runtime
access to tcg_ctx->guest_mo.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
---
accel/tcg/internal-target.h | 4 +---
1 file changed, 1 insertion(+), 3 deletions(-)
diff --git a/accel/tcg/internal-target.h b/accel/tcg/internal-target.h
index 1cb35dba99e..014ee756f9d 100644
--- a/accel/tcg/internal-target.h
+++ b/accel/tcg/internal-target.h
@@ -49,11 +49,9 @@ G_NORETURN void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr);
* Filter @type to the barrier that is required for the guest
* memory ordering vs the host memory ordering. A non-zero
* result indicates that some barrier is required.
- *
- * This is a macro so that it's constant even without optimization.
*/
#define tcg_req_mo(type) \
- ((type) & TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO)
+ ((type) & tcg_ctx->guest_mo & ~TCG_TARGET_DEFAULT_MO)
/**
* cpu_req_mo:
--
2.47.1
^ permalink raw reply related [flat|nested] 19+ messages in thread
* [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps
2025-03-21 12:57 [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order Philippe Mathieu-Daudé
` (2 preceding siblings ...)
2025-03-21 12:57 ` [RFC PATCH-for-10.1 3/4] tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo Philippe Mathieu-Daudé
@ 2025-03-21 12:57 ` Philippe Mathieu-Daudé
2025-03-21 13:20 ` Philippe Mathieu-Daudé
` (3 more replies)
2025-03-21 14:40 ` [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order Anton Johansson via
4 siblings, 4 replies; 19+ messages in thread
From: Philippe Mathieu-Daudé @ 2025-03-21 12:57 UTC (permalink / raw)
To: Richard Henderson, qemu-devel
Cc: Paolo Bonzini, Pierrick Bouvier, Alex Bennée,
Anton Johansson, Philippe Mathieu-Daudé
Add the TCGCPUOps::guest_default_memory_order field and have
each target initialize it.
Use it to set TCGContext::guest_mo in tb_gen_code(), removing
the need for the TCG_GUEST_DEFAULT_MO definition.
Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
---
include/accel/tcg/cpu-ops.h | 8 ++++++++
target/alpha/cpu-param.h | 3 ---
target/arm/cpu-param.h | 3 ---
target/avr/cpu-param.h | 2 --
target/hexagon/cpu-param.h | 3 ---
target/hppa/cpu-param.h | 8 --------
target/i386/cpu-param.h | 3 ---
target/loongarch/cpu-param.h | 2 --
target/m68k/cpu-param.h | 3 ---
target/microblaze/cpu-param.h | 3 ---
target/mips/cpu-param.h | 2 --
target/openrisc/cpu-param.h | 2 --
target/ppc/cpu-param.h | 2 --
target/riscv/cpu-param.h | 2 --
target/rx/cpu-param.h | 3 ---
target/s390x/cpu-param.h | 6 ------
target/sh4/cpu-param.h | 3 ---
target/sparc/cpu-param.h | 23 -----------------------
target/tricore/cpu-param.h | 3 ---
target/xtensa/cpu-param.h | 3 ---
accel/tcg/translate-all.c | 2 +-
target/alpha/cpu.c | 3 +++
target/arm/cpu.c | 3 +++
target/arm/tcg/cpu-v7m.c | 3 +++
target/avr/cpu.c | 1 +
target/hexagon/cpu.c | 2 ++
target/hppa/cpu.c | 8 ++++++++
target/i386/tcg/tcg-cpu.c | 5 +++++
target/loongarch/cpu.c | 2 ++
target/m68k/cpu.c | 3 +++
target/microblaze/cpu.c | 3 +++
target/mips/cpu.c | 2 ++
target/openrisc/cpu.c | 2 ++
target/ppc/cpu_init.c | 2 ++
target/riscv/tcg/tcg-cpu.c | 2 ++
target/rx/cpu.c | 3 +++
target/s390x/cpu.c | 6 ++++++
target/sh4/cpu.c | 3 +++
target/sparc/cpu.c | 23 +++++++++++++++++++++++
target/tricore/cpu.c | 2 ++
target/xtensa/cpu.c | 3 +++
41 files changed, 90 insertions(+), 80 deletions(-)
diff --git a/include/accel/tcg/cpu-ops.h b/include/accel/tcg/cpu-ops.h
index f60e5303f21..5fd299cefb6 100644
--- a/include/accel/tcg/cpu-ops.h
+++ b/include/accel/tcg/cpu-ops.h
@@ -16,8 +16,16 @@
#include "exec/memop.h"
#include "exec/mmu-access-type.h"
#include "exec/vaddr.h"
+#include "tcg/tcg-mo.h"
struct TCGCPUOps {
+
+ /**
+ * @guest_default_memory_order: default barrier that is required
+ * for the guest memory ordering.
+ */
+ TCGBar guest_default_memory_order;
+
/**
* @initialize: Initialize TCG state
*
diff --git a/target/alpha/cpu-param.h b/target/alpha/cpu-param.h
index ff06e41497a..c74556d2667 100644
--- a/target/alpha/cpu-param.h
+++ b/target/alpha/cpu-param.h
@@ -25,7 +25,4 @@
# define TARGET_VIRT_ADDR_SPACE_BITS (30 + TARGET_PAGE_BITS)
#endif
-/* Alpha processors have a weak memory model */
-#define TCG_GUEST_DEFAULT_MO (0)
-
#endif
diff --git a/target/arm/cpu-param.h b/target/arm/cpu-param.h
index 896b35bd6d5..55de4d54544 100644
--- a/target/arm/cpu-param.h
+++ b/target/arm/cpu-param.h
@@ -38,7 +38,4 @@
# define TARGET_PAGE_BITS_MIN 10
#endif /* !CONFIG_USER_ONLY */
-/* ARM processors have a weak memory model */
-#define TCG_GUEST_DEFAULT_MO (0)
-
#endif
diff --git a/target/avr/cpu-param.h b/target/avr/cpu-param.h
index 81f3f49ee1f..11e827109c0 100644
--- a/target/avr/cpu-param.h
+++ b/target/avr/cpu-param.h
@@ -31,6 +31,4 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 24
#define TARGET_VIRT_ADDR_SPACE_BITS 24
-#define TCG_GUEST_DEFAULT_MO 0
-
#endif
diff --git a/target/hexagon/cpu-param.h b/target/hexagon/cpu-param.h
index 2d57ea6caf9..45ee7b46409 100644
--- a/target/hexagon/cpu-param.h
+++ b/target/hexagon/cpu-param.h
@@ -23,7 +23,4 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 36
#define TARGET_VIRT_ADDR_SPACE_BITS 32
-/* MTTCG not yet supported: require strict ordering */
-#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
-
#endif
diff --git a/target/hppa/cpu-param.h b/target/hppa/cpu-param.h
index 7ed6b5741e7..e0b2c7c9157 100644
--- a/target/hppa/cpu-param.h
+++ b/target/hppa/cpu-param.h
@@ -19,12 +19,4 @@
#define TARGET_PAGE_BITS 12
-/* PA-RISC 1.x processors have a strong memory model. */
-/*
- * ??? While we do not yet implement PA-RISC 2.0, those processors have
- * a weak memory model, but with TLB bits that force ordering on a per-page
- * basis. It's probably easier to fall back to a strong memory model.
- */
-#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
-
#endif
diff --git a/target/i386/cpu-param.h b/target/i386/cpu-param.h
index b0e884c5d70..909bc027923 100644
--- a/target/i386/cpu-param.h
+++ b/target/i386/cpu-param.h
@@ -22,7 +22,4 @@
#endif
#define TARGET_PAGE_BITS 12
-/* The x86 has a strong memory model with some store-after-load re-ordering */
-#define TCG_GUEST_DEFAULT_MO (TCG_MO_ALL & ~TCG_MO_ST_LD)
-
#endif
diff --git a/target/loongarch/cpu-param.h b/target/loongarch/cpu-param.h
index 52437946e56..071567712b3 100644
--- a/target/loongarch/cpu-param.h
+++ b/target/loongarch/cpu-param.h
@@ -13,6 +13,4 @@
#define TARGET_PAGE_BITS 12
-#define TCG_GUEST_DEFAULT_MO (0)
-
#endif
diff --git a/target/m68k/cpu-param.h b/target/m68k/cpu-param.h
index 1a909eaa13e..7afbf6d302d 100644
--- a/target/m68k/cpu-param.h
+++ b/target/m68k/cpu-param.h
@@ -17,7 +17,4 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 32
#define TARGET_VIRT_ADDR_SPACE_BITS 32
-/* MTTCG not yet supported: require strict ordering */
-#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
-
#endif
diff --git a/target/microblaze/cpu-param.h b/target/microblaze/cpu-param.h
index c866ec6c149..6a0714bb3d7 100644
--- a/target/microblaze/cpu-param.h
+++ b/target/microblaze/cpu-param.h
@@ -27,7 +27,4 @@
/* FIXME: MB uses variable pages down to 1K but linux only uses 4k. */
#define TARGET_PAGE_BITS 12
-/* MicroBlaze is always in-order. */
-#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
-
#endif
diff --git a/target/mips/cpu-param.h b/target/mips/cpu-param.h
index 11b3ac0ac63..35fb6ea7243 100644
--- a/target/mips/cpu-param.h
+++ b/target/mips/cpu-param.h
@@ -25,6 +25,4 @@
#define TARGET_PAGE_BITS_MIN 12
#endif
-#define TCG_GUEST_DEFAULT_MO (0)
-
#endif
diff --git a/target/openrisc/cpu-param.h b/target/openrisc/cpu-param.h
index 37627f2c394..3011bf5fcca 100644
--- a/target/openrisc/cpu-param.h
+++ b/target/openrisc/cpu-param.h
@@ -12,6 +12,4 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 32
#define TARGET_VIRT_ADDR_SPACE_BITS 32
-#define TCG_GUEST_DEFAULT_MO (0)
-
#endif
diff --git a/target/ppc/cpu-param.h b/target/ppc/cpu-param.h
index 6c4525fdf3c..2cee113ddd3 100644
--- a/target/ppc/cpu-param.h
+++ b/target/ppc/cpu-param.h
@@ -38,6 +38,4 @@
# define TARGET_PAGE_BITS 12
#endif
-#define TCG_GUEST_DEFAULT_MO 0
-
#endif
diff --git a/target/riscv/cpu-param.h b/target/riscv/cpu-param.h
index fba30e966a8..a80310ef2c5 100644
--- a/target/riscv/cpu-param.h
+++ b/target/riscv/cpu-param.h
@@ -26,6 +26,4 @@
* - M mode HLV/HLVX/HSV 0b111
*/
-#define TCG_GUEST_DEFAULT_MO 0
-
#endif
diff --git a/target/rx/cpu-param.h b/target/rx/cpu-param.h
index 2ce199164d7..ef1970a09e9 100644
--- a/target/rx/cpu-param.h
+++ b/target/rx/cpu-param.h
@@ -24,7 +24,4 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 32
#define TARGET_VIRT_ADDR_SPACE_BITS 32
-/* MTTCG not yet supported: require strict ordering */
-#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
-
#endif
diff --git a/target/s390x/cpu-param.h b/target/s390x/cpu-param.h
index 5c331ec424c..a5f798eeae7 100644
--- a/target/s390x/cpu-param.h
+++ b/target/s390x/cpu-param.h
@@ -12,10 +12,4 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 64
#define TARGET_VIRT_ADDR_SPACE_BITS 64
-/*
- * The z/Architecture has a strong memory model with some
- * store-after-load re-ordering.
- */
-#define TCG_GUEST_DEFAULT_MO (TCG_MO_ALL & ~TCG_MO_ST_LD)
-
#endif
diff --git a/target/sh4/cpu-param.h b/target/sh4/cpu-param.h
index 1bc90d4695e..2b6e11dd0ac 100644
--- a/target/sh4/cpu-param.h
+++ b/target/sh4/cpu-param.h
@@ -16,7 +16,4 @@
# define TARGET_VIRT_ADDR_SPACE_BITS 32
#endif
-/* MTTCG not yet supported: require strict ordering */
-#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
-
#endif
diff --git a/target/sparc/cpu-param.h b/target/sparc/cpu-param.h
index 6952ee2b826..6e8e2a51469 100644
--- a/target/sparc/cpu-param.h
+++ b/target/sparc/cpu-param.h
@@ -21,27 +21,4 @@
# define TARGET_VIRT_ADDR_SPACE_BITS 32
#endif
-/*
- * From Oracle SPARC Architecture 2015:
- *
- * Compatibility notes: The PSO memory model described in SPARC V8 and
- * SPARC V9 compatibility architecture specifications was never implemented
- * in a SPARC V9 implementation and is not included in the Oracle SPARC
- * Architecture specification.
- *
- * The RMO memory model described in the SPARC V9 specification was
- * implemented in some non-Sun SPARC V9 implementations, but is not
- * directly supported in Oracle SPARC Architecture 2015 implementations.
- *
- * Therefore always use TSO in QEMU.
- *
- * D.5 Specification of Partial Store Order (PSO)
- * ... [loads] are followed by an implied MEMBAR #LoadLoad | #LoadStore.
- *
- * D.6 Specification of Total Store Order (TSO)
- * ... PSO with the additional requirement that all [stores] are followed
- * by an implied MEMBAR #StoreStore.
- */
-#define TCG_GUEST_DEFAULT_MO (TCG_MO_LD_LD | TCG_MO_LD_ST | TCG_MO_ST_ST)
-
#endif
diff --git a/target/tricore/cpu-param.h b/target/tricore/cpu-param.h
index 923459370cc..790242ef3d2 100644
--- a/target/tricore/cpu-param.h
+++ b/target/tricore/cpu-param.h
@@ -12,7 +12,4 @@
#define TARGET_PHYS_ADDR_SPACE_BITS 32
#define TARGET_VIRT_ADDR_SPACE_BITS 32
-/* MTTCG not yet supported: require strict ordering */
-#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
-
#endif
diff --git a/target/xtensa/cpu-param.h b/target/xtensa/cpu-param.h
index 5e4848ad059..06d85218b84 100644
--- a/target/xtensa/cpu-param.h
+++ b/target/xtensa/cpu-param.h
@@ -16,7 +16,4 @@
#define TARGET_VIRT_ADDR_SPACE_BITS 32
#endif
-/* Xtensa processors have a weak memory model */
-#define TCG_GUEST_DEFAULT_MO (0)
-
#endif
diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
index fb9f83dbba3..26442e83776 100644
--- a/accel/tcg/translate-all.c
+++ b/accel/tcg/translate-all.c
@@ -349,7 +349,7 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
tcg_ctx->tlb_dyn_max_bits = CPU_TLB_DYN_MAX_BITS;
#endif
tcg_ctx->insn_start_words = TARGET_INSN_START_WORDS;
- tcg_ctx->guest_mo = TCG_GUEST_DEFAULT_MO;
+ tcg_ctx->guest_mo = cpu->cc->tcg_ops->guest_default_memory_order;
restart_translate:
trace_translate_block(tb, pc, tb->tc.ptr);
diff --git a/target/alpha/cpu.c b/target/alpha/cpu.c
index 584c2aa76bd..e5e14976f51 100644
--- a/target/alpha/cpu.c
+++ b/target/alpha/cpu.c
@@ -239,6 +239,9 @@ static const TCGCPUOps alpha_tcg_ops = {
.synchronize_from_tb = alpha_cpu_synchronize_from_tb,
.restore_state_to_opc = alpha_restore_state_to_opc,
+ /* Alpha processors have a weak memory model */
+ .guest_default_memory_order = 0,
+
#ifdef CONFIG_USER_ONLY
.record_sigsegv = alpha_cpu_record_sigsegv,
.record_sigbus = alpha_cpu_record_sigbus,
diff --git a/target/arm/cpu.c b/target/arm/cpu.c
index 01786ac7879..8b9f2acf82b 100644
--- a/target/arm/cpu.c
+++ b/target/arm/cpu.c
@@ -2675,6 +2675,9 @@ static const TCGCPUOps arm_tcg_ops = {
.debug_excp_handler = arm_debug_excp_handler,
.restore_state_to_opc = arm_restore_state_to_opc,
+ /* ARM processors have a weak memory model */
+ .guest_default_memory_order = 0,
+
#ifdef CONFIG_USER_ONLY
.record_sigsegv = arm_cpu_record_sigsegv,
.record_sigbus = arm_cpu_record_sigbus,
diff --git a/target/arm/tcg/cpu-v7m.c b/target/arm/tcg/cpu-v7m.c
index c4dd3092726..df6b7198944 100644
--- a/target/arm/tcg/cpu-v7m.c
+++ b/target/arm/tcg/cpu-v7m.c
@@ -238,6 +238,9 @@ static const TCGCPUOps arm_v7m_tcg_ops = {
.debug_excp_handler = arm_debug_excp_handler,
.restore_state_to_opc = arm_restore_state_to_opc,
+ /* ARM processors have a weak memory model */
+ .guest_default_memory_order = 0,
+
#ifdef CONFIG_USER_ONLY
.record_sigsegv = arm_cpu_record_sigsegv,
.record_sigbus = arm_cpu_record_sigbus,
diff --git a/target/avr/cpu.c b/target/avr/cpu.c
index 834c7082aa7..24e52e28f44 100644
--- a/target/avr/cpu.c
+++ b/target/avr/cpu.c
@@ -216,6 +216,7 @@ static const TCGCPUOps avr_tcg_ops = {
.cpu_exec_halt = avr_cpu_has_work,
.tlb_fill = avr_cpu_tlb_fill,
.do_interrupt = avr_cpu_do_interrupt,
+ .guest_default_memory_order = 0,
};
static void avr_cpu_class_init(ObjectClass *oc, void *data)
diff --git a/target/hexagon/cpu.c b/target/hexagon/cpu.c
index 766b6786511..34734b0edb0 100644
--- a/target/hexagon/cpu.c
+++ b/target/hexagon/cpu.c
@@ -324,6 +324,8 @@ static const TCGCPUOps hexagon_tcg_ops = {
.translate_code = hexagon_translate_code,
.synchronize_from_tb = hexagon_cpu_synchronize_from_tb,
.restore_state_to_opc = hexagon_restore_state_to_opc,
+ /* MTTCG not yet supported: require strict ordering */
+ .guest_default_memory_order = TCG_MO_ALL,
};
static void hexagon_cpu_class_init(ObjectClass *c, void *data)
diff --git a/target/hppa/cpu.c b/target/hppa/cpu.c
index 2a85495d02f..997bd69db19 100644
--- a/target/hppa/cpu.c
+++ b/target/hppa/cpu.c
@@ -257,6 +257,14 @@ static const TCGCPUOps hppa_tcg_ops = {
.synchronize_from_tb = hppa_cpu_synchronize_from_tb,
.restore_state_to_opc = hppa_restore_state_to_opc,
+ /* PA-RISC 1.x processors have a strong memory model. */
+ /*
+ * ??? While we do not yet implement PA-RISC 2.0, those processors have
+ * a weak memory model, but with TLB bits that force ordering on a per-page
+ * basis. It's probably easier to fall back to a strong memory model.
+ */
+ .guest_default_memory_order = TCG_MO_ALL,
+
#ifndef CONFIG_USER_ONLY
.tlb_fill_align = hppa_cpu_tlb_fill_align,
.cpu_exec_interrupt = hppa_cpu_exec_interrupt,
diff --git a/target/i386/tcg/tcg-cpu.c b/target/i386/tcg/tcg-cpu.c
index b8aff825eec..4a76c475971 100644
--- a/target/i386/tcg/tcg-cpu.c
+++ b/target/i386/tcg/tcg-cpu.c
@@ -128,6 +128,11 @@ static const TCGCPUOps x86_tcg_ops = {
.debug_check_breakpoint = x86_debug_check_breakpoint,
.need_replay_interrupt = x86_need_replay_interrupt,
#endif /* !CONFIG_USER_ONLY */
+
+ /*
+ * The x86 has a strong memory model with some store-after-load re-ordering
+ */
+ .guest_default_memory_order = TCG_MO_ALL & ~TCG_MO_ST_LD,
};
static void x86_tcg_cpu_init_ops(AccelCPUClass *accel_cpu, CPUClass *cc)
diff --git a/target/loongarch/cpu.c b/target/loongarch/cpu.c
index ea1665e2705..c39ff056157 100644
--- a/target/loongarch/cpu.c
+++ b/target/loongarch/cpu.c
@@ -869,6 +869,8 @@ static const TCGCPUOps loongarch_tcg_ops = {
.synchronize_from_tb = loongarch_cpu_synchronize_from_tb,
.restore_state_to_opc = loongarch_restore_state_to_opc,
+ .guest_default_memory_order = 0,
+
#ifndef CONFIG_USER_ONLY
.tlb_fill = loongarch_cpu_tlb_fill,
.cpu_exec_interrupt = loongarch_cpu_exec_interrupt,
diff --git a/target/m68k/cpu.c b/target/m68k/cpu.c
index 0065e1c1ca5..e96b379e266 100644
--- a/target/m68k/cpu.c
+++ b/target/m68k/cpu.c
@@ -593,6 +593,9 @@ static const TCGCPUOps m68k_tcg_ops = {
.translate_code = m68k_translate_code,
.restore_state_to_opc = m68k_restore_state_to_opc,
+ /* MTTCG not yet supported: require strict ordering */
+ .guest_default_memory_order = TCG_MO_ALL,
+
#ifndef CONFIG_USER_ONLY
.tlb_fill = m68k_cpu_tlb_fill,
.cpu_exec_interrupt = m68k_cpu_exec_interrupt,
diff --git a/target/microblaze/cpu.c b/target/microblaze/cpu.c
index f3bebea856e..4b9ef6e52c4 100644
--- a/target/microblaze/cpu.c
+++ b/target/microblaze/cpu.c
@@ -432,6 +432,9 @@ static const TCGCPUOps mb_tcg_ops = {
.synchronize_from_tb = mb_cpu_synchronize_from_tb,
.restore_state_to_opc = mb_restore_state_to_opc,
+ /* MicroBlaze is always in-order. */
+ .guest_default_memory_order = TCG_MO_ALL,
+
#ifndef CONFIG_USER_ONLY
.tlb_fill = mb_cpu_tlb_fill,
.cpu_exec_interrupt = mb_cpu_exec_interrupt,
diff --git a/target/mips/cpu.c b/target/mips/cpu.c
index b207106dd79..5ddc9bbb829 100644
--- a/target/mips/cpu.c
+++ b/target/mips/cpu.c
@@ -554,6 +554,8 @@ static const TCGCPUOps mips_tcg_ops = {
.synchronize_from_tb = mips_cpu_synchronize_from_tb,
.restore_state_to_opc = mips_restore_state_to_opc,
+ .guest_default_memory_order = 0,
+
#if !defined(CONFIG_USER_ONLY)
.tlb_fill = mips_cpu_tlb_fill,
.cpu_exec_interrupt = mips_cpu_exec_interrupt,
diff --git a/target/openrisc/cpu.c b/target/openrisc/cpu.c
index e8abf1f8b5c..6a878aaadd8 100644
--- a/target/openrisc/cpu.c
+++ b/target/openrisc/cpu.c
@@ -248,6 +248,8 @@ static const TCGCPUOps openrisc_tcg_ops = {
.synchronize_from_tb = openrisc_cpu_synchronize_from_tb,
.restore_state_to_opc = openrisc_restore_state_to_opc,
+ .guest_default_memory_order = 0,
+
#ifndef CONFIG_USER_ONLY
.tlb_fill = openrisc_cpu_tlb_fill,
.cpu_exec_interrupt = openrisc_cpu_exec_interrupt,
diff --git a/target/ppc/cpu_init.c b/target/ppc/cpu_init.c
index 8b590e7f17c..28fbbb8d3c1 100644
--- a/target/ppc/cpu_init.c
+++ b/target/ppc/cpu_init.c
@@ -7490,6 +7490,8 @@ static const TCGCPUOps ppc_tcg_ops = {
.translate_code = ppc_translate_code,
.restore_state_to_opc = ppc_restore_state_to_opc,
+ .guest_default_memory_order = 0,
+
#ifdef CONFIG_USER_ONLY
.record_sigsegv = ppc_cpu_record_sigsegv,
#else
diff --git a/target/riscv/tcg/tcg-cpu.c b/target/riscv/tcg/tcg-cpu.c
index 5aef9eef366..fb903992faa 100644
--- a/target/riscv/tcg/tcg-cpu.c
+++ b/target/riscv/tcg/tcg-cpu.c
@@ -139,6 +139,8 @@ static const TCGCPUOps riscv_tcg_ops = {
.synchronize_from_tb = riscv_cpu_synchronize_from_tb,
.restore_state_to_opc = riscv_restore_state_to_opc,
+ .guest_default_memory_order = 0,
+
#ifndef CONFIG_USER_ONLY
.tlb_fill = riscv_cpu_tlb_fill,
.cpu_exec_interrupt = riscv_cpu_exec_interrupt,
diff --git a/target/rx/cpu.c b/target/rx/cpu.c
index 0ba0d55ab5b..6a24e7e9136 100644
--- a/target/rx/cpu.c
+++ b/target/rx/cpu.c
@@ -212,6 +212,9 @@ static const TCGCPUOps rx_tcg_ops = {
.cpu_exec_interrupt = rx_cpu_exec_interrupt,
.cpu_exec_halt = rx_cpu_has_work,
.do_interrupt = rx_cpu_do_interrupt,
+
+ /* MTTCG not yet supported: require strict ordering */
+ .guest_default_memory_order = TCG_MO_ALL,
};
static void rx_cpu_class_init(ObjectClass *klass, void *data)
diff --git a/target/s390x/cpu.c b/target/s390x/cpu.c
index d73142600bf..12fd853c00a 100644
--- a/target/s390x/cpu.c
+++ b/target/s390x/cpu.c
@@ -360,6 +360,12 @@ static const TCGCPUOps s390_tcg_ops = {
.debug_excp_handler = s390x_cpu_debug_excp_handler,
.do_unaligned_access = s390x_cpu_do_unaligned_access,
#endif /* !CONFIG_USER_ONLY */
+
+ /*
+ * The z/Architecture has a strong memory model with some
+ * store-after-load re-ordering.
+ */
+ .guest_default_memory_order = TCG_MO_ALL & ~TCG_MO_ST_LD,
};
#endif /* CONFIG_TCG */
diff --git a/target/sh4/cpu.c b/target/sh4/cpu.c
index ce84bdf539a..ce9ed75107a 100644
--- a/target/sh4/cpu.c
+++ b/target/sh4/cpu.c
@@ -267,6 +267,9 @@ static const TCGCPUOps superh_tcg_ops = {
.synchronize_from_tb = superh_cpu_synchronize_from_tb,
.restore_state_to_opc = superh_restore_state_to_opc,
+ /* MTTCG not yet supported: require strict ordering */
+ .guest_default_memory_order = TCG_MO_ALL,
+
#ifndef CONFIG_USER_ONLY
.tlb_fill = superh_cpu_tlb_fill,
.cpu_exec_interrupt = superh_cpu_exec_interrupt,
diff --git a/target/sparc/cpu.c b/target/sparc/cpu.c
index 57161201173..39bd0c42855 100644
--- a/target/sparc/cpu.c
+++ b/target/sparc/cpu.c
@@ -1005,6 +1005,29 @@ static const TCGCPUOps sparc_tcg_ops = {
.synchronize_from_tb = sparc_cpu_synchronize_from_tb,
.restore_state_to_opc = sparc_restore_state_to_opc,
+ /*
+ * From Oracle SPARC Architecture 2015:
+ *
+ * Compatibility notes: The PSO memory model described in SPARC V8 and
+ * SPARC V9 compatibility architecture specifications was never
+ * implemented in a SPARC V9 implementation and is not included in the
+ * Oracle SPARC Architecture specification.
+ *
+ * The RMO memory model described in the SPARC V9 specification was
+ * implemented in some non-Sun SPARC V9 implementations, but is not
+ * directly supported in Oracle SPARC Architecture 2015 implementations.
+ *
+ * Therefore always use TSO in QEMU.
+ *
+ * D.5 Specification of Partial Store Order (PSO)
+ * ... [loads] are followed by an implied MEMBAR #LoadLoad | #LoadStore.
+ *
+ * D.6 Specification of Total Store Order (TSO)
+ * ... PSO with the additional requirement that all [stores] are followed
+ * by an implied MEMBAR #StoreStore.
+ */
+ .guest_default_memory_order = TCG_MO_LD_LD | TCG_MO_LD_ST | TCG_MO_ST_ST,
+
#ifndef CONFIG_USER_ONLY
.tlb_fill = sparc_cpu_tlb_fill,
.cpu_exec_interrupt = sparc_cpu_exec_interrupt,
diff --git a/target/tricore/cpu.c b/target/tricore/cpu.c
index 16acc4ecb92..e0a48065948 100644
--- a/target/tricore/cpu.c
+++ b/target/tricore/cpu.c
@@ -179,6 +179,8 @@ static const TCGCPUOps tricore_tcg_ops = {
.tlb_fill = tricore_cpu_tlb_fill,
.cpu_exec_interrupt = tricore_cpu_exec_interrupt,
.cpu_exec_halt = tricore_cpu_has_work,
+ /* MTTCG not yet supported: require strict ordering */
+ .guest_default_memory_order = TCG_MO_ALL,
};
static void tricore_cpu_class_init(ObjectClass *c, void *data)
diff --git a/target/xtensa/cpu.c b/target/xtensa/cpu.c
index 7663b62d01e..dd9061ba469 100644
--- a/target/xtensa/cpu.c
+++ b/target/xtensa/cpu.c
@@ -237,6 +237,9 @@ static const TCGCPUOps xtensa_tcg_ops = {
.debug_excp_handler = xtensa_breakpoint_handler,
.restore_state_to_opc = xtensa_restore_state_to_opc,
+ /* Xtensa processors have a weak memory model */
+ .guest_default_memory_order = 0,
+
#ifndef CONFIG_USER_ONLY
.tlb_fill = xtensa_cpu_tlb_fill,
.cpu_exec_interrupt = xtensa_cpu_exec_interrupt,
--
2.47.1
^ permalink raw reply related [flat|nested] 19+ messages in thread
* Re: [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps
2025-03-21 12:57 ` [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps Philippe Mathieu-Daudé
@ 2025-03-21 13:20 ` Philippe Mathieu-Daudé
2025-03-21 14:34 ` Anton Johansson via
` (2 subsequent siblings)
3 siblings, 0 replies; 19+ messages in thread
From: Philippe Mathieu-Daudé @ 2025-03-21 13:20 UTC (permalink / raw)
To: Richard Henderson, qemu-devel
Cc: Paolo Bonzini, Pierrick Bouvier, Alex Bennée,
Anton Johansson
On 21/3/25 13:57, Philippe Mathieu-Daudé wrote:
> Add the TCGCPUOps::guest_default_memory_order field and have
> each target initialize it.
>
> Use it to set TCGContext::guest_mo in tb_gen_code(), removing
> the need for the TCG_GUEST_DEFAULT_MO definition.
>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> ---
> include/accel/tcg/cpu-ops.h | 8 ++++++++
> target/alpha/cpu-param.h | 3 ---
> target/arm/cpu-param.h | 3 ---
> target/avr/cpu-param.h | 2 --
> target/hexagon/cpu-param.h | 3 ---
> target/hppa/cpu-param.h | 8 --------
> target/i386/cpu-param.h | 3 ---
> target/loongarch/cpu-param.h | 2 --
> target/m68k/cpu-param.h | 3 ---
> target/microblaze/cpu-param.h | 3 ---
> target/mips/cpu-param.h | 2 --
> target/openrisc/cpu-param.h | 2 --
> target/ppc/cpu-param.h | 2 --
> target/riscv/cpu-param.h | 2 --
> target/rx/cpu-param.h | 3 ---
> target/s390x/cpu-param.h | 6 ------
> target/sh4/cpu-param.h | 3 ---
> target/sparc/cpu-param.h | 23 -----------------------
> target/tricore/cpu-param.h | 3 ---
> target/xtensa/cpu-param.h | 3 ---
> accel/tcg/translate-all.c | 2 +-
> target/alpha/cpu.c | 3 +++
> target/arm/cpu.c | 3 +++
> target/arm/tcg/cpu-v7m.c | 3 +++
> target/avr/cpu.c | 1 +
> target/hexagon/cpu.c | 2 ++
> target/hppa/cpu.c | 8 ++++++++
> target/i386/tcg/tcg-cpu.c | 5 +++++
> target/loongarch/cpu.c | 2 ++
> target/m68k/cpu.c | 3 +++
> target/microblaze/cpu.c | 3 +++
> target/mips/cpu.c | 2 ++
> target/openrisc/cpu.c | 2 ++
> target/ppc/cpu_init.c | 2 ++
> target/riscv/tcg/tcg-cpu.c | 2 ++
> target/rx/cpu.c | 3 +++
> target/s390x/cpu.c | 6 ++++++
> target/sh4/cpu.c | 3 +++
> target/sparc/cpu.c | 23 +++++++++++++++++++++++
> target/tricore/cpu.c | 2 ++
> target/xtensa/cpu.c | 3 +++
> 41 files changed, 90 insertions(+), 80 deletions(-)
Missing one TCG_GUEST_DEFAULT_MO mention in docs:
-- >8 --
diff --git a/docs/devel/multi-thread-tcg.rst
b/docs/devel/multi-thread-tcg.rst
index b0f473961dd..14a2a9dc7b5 100644
--- a/docs/devel/multi-thread-tcg.rst
+++ b/docs/devel/multi-thread-tcg.rst
@@ -31,2 +31,2 @@ combinations where the host memory model is able to
accommodate the
-guest (TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO is zero) and the
-guest has had the required work done to support this safely
+guest (TCGCPUOps::guest_default_memory_order & ~TCG_TARGET_DEFAULT_MO
is zero)
+and the guest has had the required work done to support this safely
---
^ permalink raw reply related [flat|nested] 19+ messages in thread
* Re: [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps
2025-03-21 12:57 ` [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps Philippe Mathieu-Daudé
2025-03-21 13:20 ` Philippe Mathieu-Daudé
@ 2025-03-21 14:34 ` Anton Johansson via
2025-03-21 15:25 ` Richard Henderson
2025-03-21 17:47 ` Pierrick Bouvier
3 siblings, 0 replies; 19+ messages in thread
From: Anton Johansson via @ 2025-03-21 14:34 UTC (permalink / raw)
To: Philippe Mathieu-Daudé
Cc: Richard Henderson, qemu-devel, Paolo Bonzini, Pierrick Bouvier,
Alex Bennée
On 21/03/25, Philippe Mathieu-Daudé wrote:
> Add the TCGCPUOps::guest_default_memory_order field and have
> each target initialize it.
>
> Use it to set TCGContext::guest_mo in tb_gen_code(), removing
> the need for the TCG_GUEST_DEFAULT_MO definition.
>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> ---
> include/accel/tcg/cpu-ops.h | 8 ++++++++
> target/alpha/cpu-param.h | 3 ---
> target/arm/cpu-param.h | 3 ---
> target/avr/cpu-param.h | 2 --
> target/hexagon/cpu-param.h | 3 ---
> target/hppa/cpu-param.h | 8 --------
> target/i386/cpu-param.h | 3 ---
> target/loongarch/cpu-param.h | 2 --
> target/m68k/cpu-param.h | 3 ---
> target/microblaze/cpu-param.h | 3 ---
> target/mips/cpu-param.h | 2 --
> target/openrisc/cpu-param.h | 2 --
> target/ppc/cpu-param.h | 2 --
> target/riscv/cpu-param.h | 2 --
> target/rx/cpu-param.h | 3 ---
> target/s390x/cpu-param.h | 6 ------
> target/sh4/cpu-param.h | 3 ---
> target/sparc/cpu-param.h | 23 -----------------------
> target/tricore/cpu-param.h | 3 ---
> target/xtensa/cpu-param.h | 3 ---
> accel/tcg/translate-all.c | 2 +-
> target/alpha/cpu.c | 3 +++
> target/arm/cpu.c | 3 +++
> target/arm/tcg/cpu-v7m.c | 3 +++
> target/avr/cpu.c | 1 +
> target/hexagon/cpu.c | 2 ++
> target/hppa/cpu.c | 8 ++++++++
> target/i386/tcg/tcg-cpu.c | 5 +++++
> target/loongarch/cpu.c | 2 ++
> target/m68k/cpu.c | 3 +++
> target/microblaze/cpu.c | 3 +++
> target/mips/cpu.c | 2 ++
> target/openrisc/cpu.c | 2 ++
> target/ppc/cpu_init.c | 2 ++
> target/riscv/tcg/tcg-cpu.c | 2 ++
> target/rx/cpu.c | 3 +++
> target/s390x/cpu.c | 6 ++++++
> target/sh4/cpu.c | 3 +++
> target/sparc/cpu.c | 23 +++++++++++++++++++++++
> target/tricore/cpu.c | 2 ++
> target/xtensa/cpu.c | 3 +++
> 41 files changed, 90 insertions(+), 80 deletions(-)
This is a nice change!:)
Reviewed-by: Anton Johansson <anjo@rev.ng>
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [RFC PATCH-for-10.1 3/4] tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo
2025-03-21 12:57 ` [RFC PATCH-for-10.1 3/4] tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo Philippe Mathieu-Daudé
@ 2025-03-21 14:39 ` Anton Johansson via
2025-03-21 14:46 ` Philippe Mathieu-Daudé
2025-03-21 15:16 ` Richard Henderson
1 sibling, 1 reply; 19+ messages in thread
From: Anton Johansson via @ 2025-03-21 14:39 UTC (permalink / raw)
To: Philippe Mathieu-Daudé
Cc: Richard Henderson, qemu-devel, Paolo Bonzini, Pierrick Bouvier,
Alex Bennée
On 21/03/25, Philippe Mathieu-Daudé wrote:
> In order to use TCG with multiple targets, replace the
> compile time use of TCG_GUEST_DEFAULT_MO by a runtime
> access to tcg_ctx->guest_mo.
>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> ---
> accel/tcg/internal-target.h | 4 +---
> 1 file changed, 1 insertion(+), 3 deletions(-)
>
> diff --git a/accel/tcg/internal-target.h b/accel/tcg/internal-target.h
> index 1cb35dba99e..014ee756f9d 100644
> --- a/accel/tcg/internal-target.h
> +++ b/accel/tcg/internal-target.h
> @@ -49,11 +49,9 @@ G_NORETURN void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr);
> * Filter @type to the barrier that is required for the guest
> * memory ordering vs the host memory ordering. A non-zero
> * result indicates that some barrier is required.
> - *
> - * This is a macro so that it's constant even without optimization.
> */
> #define tcg_req_mo(type) \
> - ((type) & TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO)
> + ((type) & tcg_ctx->guest_mo & ~TCG_TARGET_DEFAULT_MO)
>
> /**
> * cpu_req_mo:
> --
> 2.47.1
>
I'll parrot some older feedback by Richard
https://lore.kernel.org/all/92cc9335-10bf-4a74-9eb4-249de5545dc5@linaro.org/
but tcg_ctx->guest_mo can be undefined or bogus here. Could we not
take cpu and access cpu->cc->tcg_ops->guest_default_memory_order
directly instead?
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO
2025-03-21 12:57 ` [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO Philippe Mathieu-Daudé
@ 2025-03-21 14:39 ` Anton Johansson via
2025-03-21 15:14 ` Richard Henderson
2025-03-21 17:44 ` Pierrick Bouvier
2 siblings, 0 replies; 19+ messages in thread
From: Anton Johansson via @ 2025-03-21 14:39 UTC (permalink / raw)
To: Philippe Mathieu-Daudé
Cc: Richard Henderson, qemu-devel, Paolo Bonzini, Pierrick Bouvier,
Alex Bennée
On 21/03/25, Philippe Mathieu-Daudé wrote:
> We only require the TCG_GUEST_DEFAULT_MO for MTTCG-enabled
> frontends, otherwise we use a default value of TCG_MO_ALL.
>
> In order to simplify, require the definition for all targets,
> defining it for hexagon, m68k, rx, sh4 and tricore.
>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> ---
> target/hexagon/cpu-param.h | 3 +++
> target/m68k/cpu-param.h | 3 +++
> target/rx/cpu-param.h | 3 +++
> target/sh4/cpu-param.h | 3 +++
> target/tricore/cpu-param.h | 3 +++
> accel/tcg/translate-all.c | 4 ----
> 6 files changed, 15 insertions(+), 4 deletions(-)
Reviewed-by: Anton Johansson <anjo@rev.ng>
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order
2025-03-21 12:57 [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order Philippe Mathieu-Daudé
` (3 preceding siblings ...)
2025-03-21 12:57 ` [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps Philippe Mathieu-Daudé
@ 2025-03-21 14:40 ` Anton Johansson via
4 siblings, 0 replies; 19+ messages in thread
From: Anton Johansson via @ 2025-03-21 14:40 UTC (permalink / raw)
To: Philippe Mathieu-Daudé
Cc: Richard Henderson, qemu-devel, Paolo Bonzini, Pierrick Bouvier,
Alex Bennée
On 21/03/25, Philippe Mathieu-Daudé wrote:
> Hi,
>
> In this series we replace the TCG_GUEST_DEFAULT_MO definition
> from "cpu-param.h" by a 'guest_default_memory_order' field in
> TCGCPUOps.
>
> Since tcg_req_mo() now accesses tcg_ctx, this impact the
> cpu_req_mo() calls in accel/tcg/{cputlb,user-exec}.c.
>
> The long term goal is to be able to use targets with distinct
> guest memory order restrictions.
>
> Philippe Mathieu-Daudé (4):
> tcg: Always define TCG_GUEST_DEFAULT_MO
> tcg: Simplify tcg_req_mo() macro
> tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo
> tcg: Define guest_default_memory_order in TCGCPUOps
>
> accel/tcg/internal-target.h | 13 ++-----------
> include/accel/tcg/cpu-ops.h | 8 ++++++++
> target/alpha/cpu-param.h | 3 ---
> target/arm/cpu-param.h | 3 ---
> target/avr/cpu-param.h | 2 --
> target/hppa/cpu-param.h | 8 --------
> target/i386/cpu-param.h | 3 ---
> target/loongarch/cpu-param.h | 2 --
> target/microblaze/cpu-param.h | 3 ---
> target/mips/cpu-param.h | 2 --
> target/openrisc/cpu-param.h | 2 --
> target/ppc/cpu-param.h | 2 --
> target/riscv/cpu-param.h | 2 --
> target/s390x/cpu-param.h | 6 ------
> target/sparc/cpu-param.h | 23 -----------------------
> target/xtensa/cpu-param.h | 3 ---
> accel/tcg/tcg-all.c | 3 ---
> accel/tcg/translate-all.c | 6 +-----
> target/alpha/cpu.c | 3 +++
> target/arm/cpu.c | 3 +++
> target/arm/tcg/cpu-v7m.c | 3 +++
> target/avr/cpu.c | 1 +
> target/hexagon/cpu.c | 2 ++
> target/hppa/cpu.c | 8 ++++++++
> target/i386/tcg/tcg-cpu.c | 5 +++++
> target/loongarch/cpu.c | 2 ++
> target/m68k/cpu.c | 3 +++
> target/microblaze/cpu.c | 3 +++
> target/mips/cpu.c | 2 ++
> target/openrisc/cpu.c | 2 ++
> target/ppc/cpu_init.c | 2 ++
> target/riscv/tcg/tcg-cpu.c | 2 ++
> target/rx/cpu.c | 3 +++
> target/s390x/cpu.c | 6 ++++++
> target/sh4/cpu.c | 3 +++
> target/sparc/cpu.c | 23 +++++++++++++++++++++++
> target/tricore/cpu.c | 2 ++
> target/xtensa/cpu.c | 3 +++
> 38 files changed, 92 insertions(+), 83 deletions(-)
>
> --
> 2.47.1
>
Reviewed-by: Anton Johansson <anjo@rev.ng>
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [PATCH-for-10.1 2/4] tcg: Simplify tcg_req_mo() macro
2025-03-21 12:57 ` [PATCH-for-10.1 2/4] tcg: Simplify tcg_req_mo() macro Philippe Mathieu-Daudé
@ 2025-03-21 14:40 ` Anton Johansson via
2025-03-21 15:14 ` Richard Henderson
2025-03-21 17:44 ` Pierrick Bouvier
2 siblings, 0 replies; 19+ messages in thread
From: Anton Johansson via @ 2025-03-21 14:40 UTC (permalink / raw)
To: Philippe Mathieu-Daudé
Cc: Richard Henderson, qemu-devel, Paolo Bonzini, Pierrick Bouvier,
Alex Bennée
On 21/03/25, Philippe Mathieu-Daudé wrote:
> Now that TCG_GUEST_DEFAULT_MO is always defined,
> simplify the tcg_req_mo() macro.
>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> ---
> accel/tcg/internal-target.h | 9 +--------
> accel/tcg/tcg-all.c | 3 ---
> 2 files changed, 1 insertion(+), 11 deletions(-)
>
> diff --git a/accel/tcg/internal-target.h b/accel/tcg/internal-target.h
> index 2cdf11c905e..1cb35dba99e 100644
> --- a/accel/tcg/internal-target.h
> +++ b/accel/tcg/internal-target.h
> @@ -50,17 +50,10 @@ G_NORETURN void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr);
> * memory ordering vs the host memory ordering. A non-zero
> * result indicates that some barrier is required.
> *
> - * If TCG_GUEST_DEFAULT_MO is not defined, assume that the
> - * guest requires strict ordering.
> - *
> * This is a macro so that it's constant even without optimization.
> */
> -#ifdef TCG_GUEST_DEFAULT_MO
> -# define tcg_req_mo(type) \
> +#define tcg_req_mo(type) \
> ((type) & TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO)
> -#else
> -# define tcg_req_mo(type) ((type) & ~TCG_TARGET_DEFAULT_MO)
> -#endif
>
> /**
> * cpu_req_mo:
> diff --git a/accel/tcg/tcg-all.c b/accel/tcg/tcg-all.c
> index c1a30b01219..cb632cc8cc7 100644
> --- a/accel/tcg/tcg-all.c
> +++ b/accel/tcg/tcg-all.c
> @@ -77,9 +77,6 @@ static bool default_mttcg_enabled(void)
> return false;
> }
> #ifdef TARGET_SUPPORTS_MTTCG
> -# ifndef TCG_GUEST_DEFAULT_MO
> -# error "TARGET_SUPPORTS_MTTCG without TCG_GUEST_DEFAULT_MO"
> -# endif
> return true;
> #else
> return false;
> --
> 2.47.1
>
Reviewed-by: Anton Johansson <anjo@rev.ng>
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [RFC PATCH-for-10.1 3/4] tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo
2025-03-21 14:39 ` Anton Johansson via
@ 2025-03-21 14:46 ` Philippe Mathieu-Daudé
0 siblings, 0 replies; 19+ messages in thread
From: Philippe Mathieu-Daudé @ 2025-03-21 14:46 UTC (permalink / raw)
To: Anton Johansson
Cc: Richard Henderson, qemu-devel, Paolo Bonzini, Pierrick Bouvier,
Alex Bennée
On 21/3/25 15:39, Anton Johansson wrote:
> On 21/03/25, Philippe Mathieu-Daudé wrote:
>> In order to use TCG with multiple targets, replace the
>> compile time use of TCG_GUEST_DEFAULT_MO by a runtime
>> access to tcg_ctx->guest_mo.
>>
>> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
>> ---
>> accel/tcg/internal-target.h | 4 +---
>> 1 file changed, 1 insertion(+), 3 deletions(-)
>>
>> diff --git a/accel/tcg/internal-target.h b/accel/tcg/internal-target.h
>> index 1cb35dba99e..014ee756f9d 100644
>> --- a/accel/tcg/internal-target.h
>> +++ b/accel/tcg/internal-target.h
>> @@ -49,11 +49,9 @@ G_NORETURN void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr);
>> * Filter @type to the barrier that is required for the guest
>> * memory ordering vs the host memory ordering. A non-zero
>> * result indicates that some barrier is required.
>> - *
>> - * This is a macro so that it's constant even without optimization.
>> */
>> #define tcg_req_mo(type) \
>> - ((type) & TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO)
>> + ((type) & tcg_ctx->guest_mo & ~TCG_TARGET_DEFAULT_MO)
>>
>> /**
>> * cpu_req_mo:
>> --
>> 2.47.1
>>
>
> I'll parrot some older feedback by Richard
>
> https://lore.kernel.org/all/92cc9335-10bf-4a74-9eb4-249de5545dc5@linaro.org/
>
> but tcg_ctx->guest_mo can be undefined or bogus here. Could we not
> take cpu and access cpu->cc->tcg_ops->guest_default_memory_order
> directly instead?
Yes, cpu_req_mo() has access to @cpu.
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO
2025-03-21 12:57 ` [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO Philippe Mathieu-Daudé
2025-03-21 14:39 ` Anton Johansson via
@ 2025-03-21 15:14 ` Richard Henderson
2025-03-21 17:44 ` Pierrick Bouvier
2 siblings, 0 replies; 19+ messages in thread
From: Richard Henderson @ 2025-03-21 15:14 UTC (permalink / raw)
To: Philippe Mathieu-Daudé, qemu-devel
Cc: Paolo Bonzini, Pierrick Bouvier, Alex Bennée,
Anton Johansson
On 3/21/25 05:57, Philippe Mathieu-Daudé wrote:
> We only require the TCG_GUEST_DEFAULT_MO for MTTCG-enabled
> frontends, otherwise we use a default value of TCG_MO_ALL.
>
> In order to simplify, require the definition for all targets,
> defining it for hexagon, m68k, rx, sh4 and tricore.
>
> Signed-off-by: Philippe Mathieu-Daudé<philmd@linaro.org>
> ---
> target/hexagon/cpu-param.h | 3 +++
> target/m68k/cpu-param.h | 3 +++
> target/rx/cpu-param.h | 3 +++
> target/sh4/cpu-param.h | 3 +++
> target/tricore/cpu-param.h | 3 +++
> accel/tcg/translate-all.c | 4 ----
> 6 files changed, 15 insertions(+), 4 deletions(-)
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
r~
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [PATCH-for-10.1 2/4] tcg: Simplify tcg_req_mo() macro
2025-03-21 12:57 ` [PATCH-for-10.1 2/4] tcg: Simplify tcg_req_mo() macro Philippe Mathieu-Daudé
2025-03-21 14:40 ` Anton Johansson via
@ 2025-03-21 15:14 ` Richard Henderson
2025-03-21 17:44 ` Pierrick Bouvier
2 siblings, 0 replies; 19+ messages in thread
From: Richard Henderson @ 2025-03-21 15:14 UTC (permalink / raw)
To: Philippe Mathieu-Daudé, qemu-devel
Cc: Paolo Bonzini, Pierrick Bouvier, Alex Bennée,
Anton Johansson
On 3/21/25 05:57, Philippe Mathieu-Daudé wrote:
> Now that TCG_GUEST_DEFAULT_MO is always defined,
> simplify the tcg_req_mo() macro.
>
> Signed-off-by: Philippe Mathieu-Daudé<philmd@linaro.org>
> ---
> accel/tcg/internal-target.h | 9 +--------
> accel/tcg/tcg-all.c | 3 ---
> 2 files changed, 1 insertion(+), 11 deletions(-)
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
r~
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [RFC PATCH-for-10.1 3/4] tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo
2025-03-21 12:57 ` [RFC PATCH-for-10.1 3/4] tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo Philippe Mathieu-Daudé
2025-03-21 14:39 ` Anton Johansson via
@ 2025-03-21 15:16 ` Richard Henderson
1 sibling, 0 replies; 19+ messages in thread
From: Richard Henderson @ 2025-03-21 15:16 UTC (permalink / raw)
To: Philippe Mathieu-Daudé, qemu-devel
Cc: Paolo Bonzini, Pierrick Bouvier, Alex Bennée,
Anton Johansson
On 3/21/25 05:57, Philippe Mathieu-Daudé wrote:
> In order to use TCG with multiple targets, replace the
> compile time use of TCG_GUEST_DEFAULT_MO by a runtime
> access to tcg_ctx->guest_mo.
>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> ---
> accel/tcg/internal-target.h | 4 +---
> 1 file changed, 1 insertion(+), 3 deletions(-)
>
> diff --git a/accel/tcg/internal-target.h b/accel/tcg/internal-target.h
> index 1cb35dba99e..014ee756f9d 100644
> --- a/accel/tcg/internal-target.h
> +++ b/accel/tcg/internal-target.h
> @@ -49,11 +49,9 @@ G_NORETURN void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr);
> * Filter @type to the barrier that is required for the guest
> * memory ordering vs the host memory ordering. A non-zero
> * result indicates that some barrier is required.
> - *
> - * This is a macro so that it's constant even without optimization.
> */
> #define tcg_req_mo(type) \
> - ((type) & TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO)
> + ((type) & tcg_ctx->guest_mo & ~TCG_TARGET_DEFAULT_MO)
Incorrect. The setting in tcg_ctx is only valid during compilation.
This macro is used during execution, in cpu_req_mo(),
from the cpu memory access routines.
r~
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps
2025-03-21 12:57 ` [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps Philippe Mathieu-Daudé
2025-03-21 13:20 ` Philippe Mathieu-Daudé
2025-03-21 14:34 ` Anton Johansson via
@ 2025-03-21 15:25 ` Richard Henderson
2025-03-21 17:47 ` Pierrick Bouvier
3 siblings, 0 replies; 19+ messages in thread
From: Richard Henderson @ 2025-03-21 15:25 UTC (permalink / raw)
To: Philippe Mathieu-Daudé, qemu-devel
Cc: Paolo Bonzini, Pierrick Bouvier, Alex Bennée,
Anton Johansson
On 3/21/25 05:57, Philippe Mathieu-Daudé wrote:
> Add theTCGCPUOps::guest_default_memory_order field and have
> each target initialize it.
>
> Use it to setTCGContext::guest_mo in tb_gen_code(), removing
> the need for the TCG_GUEST_DEFAULT_MO definition.
>
> Signed-off-by: Philippe Mathieu-Daudé<philmd@linaro.org>
> ---
Reviewed-by: Richard Henderson <richard.henderson@linaro.org>
r~
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO
2025-03-21 12:57 ` [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO Philippe Mathieu-Daudé
2025-03-21 14:39 ` Anton Johansson via
2025-03-21 15:14 ` Richard Henderson
@ 2025-03-21 17:44 ` Pierrick Bouvier
2 siblings, 0 replies; 19+ messages in thread
From: Pierrick Bouvier @ 2025-03-21 17:44 UTC (permalink / raw)
To: Philippe Mathieu-Daudé, Richard Henderson, qemu-devel
Cc: Paolo Bonzini, Alex Bennée, Anton Johansson
On 3/21/25 05:57, Philippe Mathieu-Daudé wrote:
> We only require the TCG_GUEST_DEFAULT_MO for MTTCG-enabled
> frontends, otherwise we use a default value of TCG_MO_ALL.
>
> In order to simplify, require the definition for all targets,
> defining it for hexagon, m68k, rx, sh4 and tricore.
>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> ---
> target/hexagon/cpu-param.h | 3 +++
> target/m68k/cpu-param.h | 3 +++
> target/rx/cpu-param.h | 3 +++
> target/sh4/cpu-param.h | 3 +++
> target/tricore/cpu-param.h | 3 +++
> accel/tcg/translate-all.c | 4 ----
> 6 files changed, 15 insertions(+), 4 deletions(-)
>
> diff --git a/target/hexagon/cpu-param.h b/target/hexagon/cpu-param.h
> index 45ee7b46409..2d57ea6caf9 100644
> --- a/target/hexagon/cpu-param.h
> +++ b/target/hexagon/cpu-param.h
> @@ -23,4 +23,7 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 36
> #define TARGET_VIRT_ADDR_SPACE_BITS 32
>
> +/* MTTCG not yet supported: require strict ordering */
> +#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> +
> #endif
> diff --git a/target/m68k/cpu-param.h b/target/m68k/cpu-param.h
> index 7afbf6d302d..1a909eaa13e 100644
> --- a/target/m68k/cpu-param.h
> +++ b/target/m68k/cpu-param.h
> @@ -17,4 +17,7 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 32
> #define TARGET_VIRT_ADDR_SPACE_BITS 32
>
> +/* MTTCG not yet supported: require strict ordering */
> +#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> +
> #endif
> diff --git a/target/rx/cpu-param.h b/target/rx/cpu-param.h
> index ef1970a09e9..2ce199164d7 100644
> --- a/target/rx/cpu-param.h
> +++ b/target/rx/cpu-param.h
> @@ -24,4 +24,7 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 32
> #define TARGET_VIRT_ADDR_SPACE_BITS 32
>
> +/* MTTCG not yet supported: require strict ordering */
> +#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> +
> #endif
> diff --git a/target/sh4/cpu-param.h b/target/sh4/cpu-param.h
> index 2b6e11dd0ac..1bc90d4695e 100644
> --- a/target/sh4/cpu-param.h
> +++ b/target/sh4/cpu-param.h
> @@ -16,4 +16,7 @@
> # define TARGET_VIRT_ADDR_SPACE_BITS 32
> #endif
>
> +/* MTTCG not yet supported: require strict ordering */
> +#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> +
> #endif
> diff --git a/target/tricore/cpu-param.h b/target/tricore/cpu-param.h
> index 790242ef3d2..923459370cc 100644
> --- a/target/tricore/cpu-param.h
> +++ b/target/tricore/cpu-param.h
> @@ -12,4 +12,7 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 32
> #define TARGET_VIRT_ADDR_SPACE_BITS 32
>
> +/* MTTCG not yet supported: require strict ordering */
> +#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> +
> #endif
> diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
> index 82bc16bd535..fb9f83dbba3 100644
> --- a/accel/tcg/translate-all.c
> +++ b/accel/tcg/translate-all.c
> @@ -349,11 +349,7 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
> tcg_ctx->tlb_dyn_max_bits = CPU_TLB_DYN_MAX_BITS;
> #endif
> tcg_ctx->insn_start_words = TARGET_INSN_START_WORDS;
> -#ifdef TCG_GUEST_DEFAULT_MO
> tcg_ctx->guest_mo = TCG_GUEST_DEFAULT_MO;
> -#else
> - tcg_ctx->guest_mo = TCG_MO_ALL;
> -#endif
>
> restart_translate:
> trace_translate_block(tb, pc, tb->tc.ptr);
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [PATCH-for-10.1 2/4] tcg: Simplify tcg_req_mo() macro
2025-03-21 12:57 ` [PATCH-for-10.1 2/4] tcg: Simplify tcg_req_mo() macro Philippe Mathieu-Daudé
2025-03-21 14:40 ` Anton Johansson via
2025-03-21 15:14 ` Richard Henderson
@ 2025-03-21 17:44 ` Pierrick Bouvier
2 siblings, 0 replies; 19+ messages in thread
From: Pierrick Bouvier @ 2025-03-21 17:44 UTC (permalink / raw)
To: Philippe Mathieu-Daudé, Richard Henderson, qemu-devel
Cc: Paolo Bonzini, Alex Bennée, Anton Johansson
On 3/21/25 05:57, Philippe Mathieu-Daudé wrote:
> Now that TCG_GUEST_DEFAULT_MO is always defined,
> simplify the tcg_req_mo() macro.
>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> ---
> accel/tcg/internal-target.h | 9 +--------
> accel/tcg/tcg-all.c | 3 ---
> 2 files changed, 1 insertion(+), 11 deletions(-)
>
> diff --git a/accel/tcg/internal-target.h b/accel/tcg/internal-target.h
> index 2cdf11c905e..1cb35dba99e 100644
> --- a/accel/tcg/internal-target.h
> +++ b/accel/tcg/internal-target.h
> @@ -50,17 +50,10 @@ G_NORETURN void cpu_io_recompile(CPUState *cpu, uintptr_t retaddr);
> * memory ordering vs the host memory ordering. A non-zero
> * result indicates that some barrier is required.
> *
> - * If TCG_GUEST_DEFAULT_MO is not defined, assume that the
> - * guest requires strict ordering.
> - *
> * This is a macro so that it's constant even without optimization.
> */
> -#ifdef TCG_GUEST_DEFAULT_MO
> -# define tcg_req_mo(type) \
> +#define tcg_req_mo(type) \
> ((type) & TCG_GUEST_DEFAULT_MO & ~TCG_TARGET_DEFAULT_MO)
> -#else
> -# define tcg_req_mo(type) ((type) & ~TCG_TARGET_DEFAULT_MO)
> -#endif
>
> /**
> * cpu_req_mo:
> diff --git a/accel/tcg/tcg-all.c b/accel/tcg/tcg-all.c
> index c1a30b01219..cb632cc8cc7 100644
> --- a/accel/tcg/tcg-all.c
> +++ b/accel/tcg/tcg-all.c
> @@ -77,9 +77,6 @@ static bool default_mttcg_enabled(void)
> return false;
> }
> #ifdef TARGET_SUPPORTS_MTTCG
> -# ifndef TCG_GUEST_DEFAULT_MO
> -# error "TARGET_SUPPORTS_MTTCG without TCG_GUEST_DEFAULT_MO"
> -# endif
> return true;
> #else
> return false;
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
^ permalink raw reply [flat|nested] 19+ messages in thread
* Re: [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps
2025-03-21 12:57 ` [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps Philippe Mathieu-Daudé
` (2 preceding siblings ...)
2025-03-21 15:25 ` Richard Henderson
@ 2025-03-21 17:47 ` Pierrick Bouvier
3 siblings, 0 replies; 19+ messages in thread
From: Pierrick Bouvier @ 2025-03-21 17:47 UTC (permalink / raw)
To: Philippe Mathieu-Daudé, Richard Henderson, qemu-devel
Cc: Paolo Bonzini, Alex Bennée, Anton Johansson
On 3/21/25 05:57, Philippe Mathieu-Daudé wrote:
> Add the TCGCPUOps::guest_default_memory_order field and have
> each target initialize it.
>
> Use it to set TCGContext::guest_mo in tb_gen_code(), removing
> the need for the TCG_GUEST_DEFAULT_MO definition.
>
> Signed-off-by: Philippe Mathieu-Daudé <philmd@linaro.org>
> ---
> include/accel/tcg/cpu-ops.h | 8 ++++++++
> target/alpha/cpu-param.h | 3 ---
> target/arm/cpu-param.h | 3 ---
> target/avr/cpu-param.h | 2 --
> target/hexagon/cpu-param.h | 3 ---
> target/hppa/cpu-param.h | 8 --------
> target/i386/cpu-param.h | 3 ---
> target/loongarch/cpu-param.h | 2 --
> target/m68k/cpu-param.h | 3 ---
> target/microblaze/cpu-param.h | 3 ---
> target/mips/cpu-param.h | 2 --
> target/openrisc/cpu-param.h | 2 --
> target/ppc/cpu-param.h | 2 --
> target/riscv/cpu-param.h | 2 --
> target/rx/cpu-param.h | 3 ---
> target/s390x/cpu-param.h | 6 ------
> target/sh4/cpu-param.h | 3 ---
> target/sparc/cpu-param.h | 23 -----------------------
> target/tricore/cpu-param.h | 3 ---
> target/xtensa/cpu-param.h | 3 ---
> accel/tcg/translate-all.c | 2 +-
> target/alpha/cpu.c | 3 +++
> target/arm/cpu.c | 3 +++
> target/arm/tcg/cpu-v7m.c | 3 +++
> target/avr/cpu.c | 1 +
> target/hexagon/cpu.c | 2 ++
> target/hppa/cpu.c | 8 ++++++++
> target/i386/tcg/tcg-cpu.c | 5 +++++
> target/loongarch/cpu.c | 2 ++
> target/m68k/cpu.c | 3 +++
> target/microblaze/cpu.c | 3 +++
> target/mips/cpu.c | 2 ++
> target/openrisc/cpu.c | 2 ++
> target/ppc/cpu_init.c | 2 ++
> target/riscv/tcg/tcg-cpu.c | 2 ++
> target/rx/cpu.c | 3 +++
> target/s390x/cpu.c | 6 ++++++
> target/sh4/cpu.c | 3 +++
> target/sparc/cpu.c | 23 +++++++++++++++++++++++
> target/tricore/cpu.c | 2 ++
> target/xtensa/cpu.c | 3 +++
> 41 files changed, 90 insertions(+), 80 deletions(-)
>
> diff --git a/include/accel/tcg/cpu-ops.h b/include/accel/tcg/cpu-ops.h
> index f60e5303f21..5fd299cefb6 100644
> --- a/include/accel/tcg/cpu-ops.h
> +++ b/include/accel/tcg/cpu-ops.h
> @@ -16,8 +16,16 @@
> #include "exec/memop.h"
> #include "exec/mmu-access-type.h"
> #include "exec/vaddr.h"
> +#include "tcg/tcg-mo.h"
>
> struct TCGCPUOps {
> +
> + /**
> + * @guest_default_memory_order: default barrier that is required
> + * for the guest memory ordering.
> + */
> + TCGBar guest_default_memory_order;
> +
> /**
> * @initialize: Initialize TCG state
> *
> diff --git a/target/alpha/cpu-param.h b/target/alpha/cpu-param.h
> index ff06e41497a..c74556d2667 100644
> --- a/target/alpha/cpu-param.h
> +++ b/target/alpha/cpu-param.h
> @@ -25,7 +25,4 @@
> # define TARGET_VIRT_ADDR_SPACE_BITS (30 + TARGET_PAGE_BITS)
> #endif
>
> -/* Alpha processors have a weak memory model */
> -#define TCG_GUEST_DEFAULT_MO (0)
> -
> #endif
> diff --git a/target/arm/cpu-param.h b/target/arm/cpu-param.h
> index 896b35bd6d5..55de4d54544 100644
> --- a/target/arm/cpu-param.h
> +++ b/target/arm/cpu-param.h
> @@ -38,7 +38,4 @@
> # define TARGET_PAGE_BITS_MIN 10
> #endif /* !CONFIG_USER_ONLY */
>
> -/* ARM processors have a weak memory model */
> -#define TCG_GUEST_DEFAULT_MO (0)
> -
> #endif
> diff --git a/target/avr/cpu-param.h b/target/avr/cpu-param.h
> index 81f3f49ee1f..11e827109c0 100644
> --- a/target/avr/cpu-param.h
> +++ b/target/avr/cpu-param.h
> @@ -31,6 +31,4 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 24
> #define TARGET_VIRT_ADDR_SPACE_BITS 24
>
> -#define TCG_GUEST_DEFAULT_MO 0
> -
> #endif
> diff --git a/target/hexagon/cpu-param.h b/target/hexagon/cpu-param.h
> index 2d57ea6caf9..45ee7b46409 100644
> --- a/target/hexagon/cpu-param.h
> +++ b/target/hexagon/cpu-param.h
> @@ -23,7 +23,4 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 36
> #define TARGET_VIRT_ADDR_SPACE_BITS 32
>
> -/* MTTCG not yet supported: require strict ordering */
> -#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> -
> #endif
> diff --git a/target/hppa/cpu-param.h b/target/hppa/cpu-param.h
> index 7ed6b5741e7..e0b2c7c9157 100644
> --- a/target/hppa/cpu-param.h
> +++ b/target/hppa/cpu-param.h
> @@ -19,12 +19,4 @@
>
> #define TARGET_PAGE_BITS 12
>
> -/* PA-RISC 1.x processors have a strong memory model. */
> -/*
> - * ??? While we do not yet implement PA-RISC 2.0, those processors have
> - * a weak memory model, but with TLB bits that force ordering on a per-page
> - * basis. It's probably easier to fall back to a strong memory model.
> - */
> -#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> -
> #endif
> diff --git a/target/i386/cpu-param.h b/target/i386/cpu-param.h
> index b0e884c5d70..909bc027923 100644
> --- a/target/i386/cpu-param.h
> +++ b/target/i386/cpu-param.h
> @@ -22,7 +22,4 @@
> #endif
> #define TARGET_PAGE_BITS 12
>
> -/* The x86 has a strong memory model with some store-after-load re-ordering */
> -#define TCG_GUEST_DEFAULT_MO (TCG_MO_ALL & ~TCG_MO_ST_LD)
> -
> #endif
> diff --git a/target/loongarch/cpu-param.h b/target/loongarch/cpu-param.h
> index 52437946e56..071567712b3 100644
> --- a/target/loongarch/cpu-param.h
> +++ b/target/loongarch/cpu-param.h
> @@ -13,6 +13,4 @@
>
> #define TARGET_PAGE_BITS 12
>
> -#define TCG_GUEST_DEFAULT_MO (0)
> -
> #endif
> diff --git a/target/m68k/cpu-param.h b/target/m68k/cpu-param.h
> index 1a909eaa13e..7afbf6d302d 100644
> --- a/target/m68k/cpu-param.h
> +++ b/target/m68k/cpu-param.h
> @@ -17,7 +17,4 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 32
> #define TARGET_VIRT_ADDR_SPACE_BITS 32
>
> -/* MTTCG not yet supported: require strict ordering */
> -#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> -
> #endif
> diff --git a/target/microblaze/cpu-param.h b/target/microblaze/cpu-param.h
> index c866ec6c149..6a0714bb3d7 100644
> --- a/target/microblaze/cpu-param.h
> +++ b/target/microblaze/cpu-param.h
> @@ -27,7 +27,4 @@
> /* FIXME: MB uses variable pages down to 1K but linux only uses 4k. */
> #define TARGET_PAGE_BITS 12
>
> -/* MicroBlaze is always in-order. */
> -#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> -
> #endif
> diff --git a/target/mips/cpu-param.h b/target/mips/cpu-param.h
> index 11b3ac0ac63..35fb6ea7243 100644
> --- a/target/mips/cpu-param.h
> +++ b/target/mips/cpu-param.h
> @@ -25,6 +25,4 @@
> #define TARGET_PAGE_BITS_MIN 12
> #endif
>
> -#define TCG_GUEST_DEFAULT_MO (0)
> -
> #endif
> diff --git a/target/openrisc/cpu-param.h b/target/openrisc/cpu-param.h
> index 37627f2c394..3011bf5fcca 100644
> --- a/target/openrisc/cpu-param.h
> +++ b/target/openrisc/cpu-param.h
> @@ -12,6 +12,4 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 32
> #define TARGET_VIRT_ADDR_SPACE_BITS 32
>
> -#define TCG_GUEST_DEFAULT_MO (0)
> -
> #endif
> diff --git a/target/ppc/cpu-param.h b/target/ppc/cpu-param.h
> index 6c4525fdf3c..2cee113ddd3 100644
> --- a/target/ppc/cpu-param.h
> +++ b/target/ppc/cpu-param.h
> @@ -38,6 +38,4 @@
> # define TARGET_PAGE_BITS 12
> #endif
>
> -#define TCG_GUEST_DEFAULT_MO 0
> -
> #endif
> diff --git a/target/riscv/cpu-param.h b/target/riscv/cpu-param.h
> index fba30e966a8..a80310ef2c5 100644
> --- a/target/riscv/cpu-param.h
> +++ b/target/riscv/cpu-param.h
> @@ -26,6 +26,4 @@
> * - M mode HLV/HLVX/HSV 0b111
> */
>
> -#define TCG_GUEST_DEFAULT_MO 0
> -
> #endif
> diff --git a/target/rx/cpu-param.h b/target/rx/cpu-param.h
> index 2ce199164d7..ef1970a09e9 100644
> --- a/target/rx/cpu-param.h
> +++ b/target/rx/cpu-param.h
> @@ -24,7 +24,4 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 32
> #define TARGET_VIRT_ADDR_SPACE_BITS 32
>
> -/* MTTCG not yet supported: require strict ordering */
> -#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> -
> #endif
> diff --git a/target/s390x/cpu-param.h b/target/s390x/cpu-param.h
> index 5c331ec424c..a5f798eeae7 100644
> --- a/target/s390x/cpu-param.h
> +++ b/target/s390x/cpu-param.h
> @@ -12,10 +12,4 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 64
> #define TARGET_VIRT_ADDR_SPACE_BITS 64
>
> -/*
> - * The z/Architecture has a strong memory model with some
> - * store-after-load re-ordering.
> - */
> -#define TCG_GUEST_DEFAULT_MO (TCG_MO_ALL & ~TCG_MO_ST_LD)
> -
> #endif
> diff --git a/target/sh4/cpu-param.h b/target/sh4/cpu-param.h
> index 1bc90d4695e..2b6e11dd0ac 100644
> --- a/target/sh4/cpu-param.h
> +++ b/target/sh4/cpu-param.h
> @@ -16,7 +16,4 @@
> # define TARGET_VIRT_ADDR_SPACE_BITS 32
> #endif
>
> -/* MTTCG not yet supported: require strict ordering */
> -#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> -
> #endif
> diff --git a/target/sparc/cpu-param.h b/target/sparc/cpu-param.h
> index 6952ee2b826..6e8e2a51469 100644
> --- a/target/sparc/cpu-param.h
> +++ b/target/sparc/cpu-param.h
> @@ -21,27 +21,4 @@
> # define TARGET_VIRT_ADDR_SPACE_BITS 32
> #endif
>
> -/*
> - * From Oracle SPARC Architecture 2015:
> - *
> - * Compatibility notes: The PSO memory model described in SPARC V8 and
> - * SPARC V9 compatibility architecture specifications was never implemented
> - * in a SPARC V9 implementation and is not included in the Oracle SPARC
> - * Architecture specification.
> - *
> - * The RMO memory model described in the SPARC V9 specification was
> - * implemented in some non-Sun SPARC V9 implementations, but is not
> - * directly supported in Oracle SPARC Architecture 2015 implementations.
> - *
> - * Therefore always use TSO in QEMU.
> - *
> - * D.5 Specification of Partial Store Order (PSO)
> - * ... [loads] are followed by an implied MEMBAR #LoadLoad | #LoadStore.
> - *
> - * D.6 Specification of Total Store Order (TSO)
> - * ... PSO with the additional requirement that all [stores] are followed
> - * by an implied MEMBAR #StoreStore.
> - */
> -#define TCG_GUEST_DEFAULT_MO (TCG_MO_LD_LD | TCG_MO_LD_ST | TCG_MO_ST_ST)
> -
> #endif
> diff --git a/target/tricore/cpu-param.h b/target/tricore/cpu-param.h
> index 923459370cc..790242ef3d2 100644
> --- a/target/tricore/cpu-param.h
> +++ b/target/tricore/cpu-param.h
> @@ -12,7 +12,4 @@
> #define TARGET_PHYS_ADDR_SPACE_BITS 32
> #define TARGET_VIRT_ADDR_SPACE_BITS 32
>
> -/* MTTCG not yet supported: require strict ordering */
> -#define TCG_GUEST_DEFAULT_MO TCG_MO_ALL
> -
> #endif
> diff --git a/target/xtensa/cpu-param.h b/target/xtensa/cpu-param.h
> index 5e4848ad059..06d85218b84 100644
> --- a/target/xtensa/cpu-param.h
> +++ b/target/xtensa/cpu-param.h
> @@ -16,7 +16,4 @@
> #define TARGET_VIRT_ADDR_SPACE_BITS 32
> #endif
>
> -/* Xtensa processors have a weak memory model */
> -#define TCG_GUEST_DEFAULT_MO (0)
> -
> #endif
> diff --git a/accel/tcg/translate-all.c b/accel/tcg/translate-all.c
> index fb9f83dbba3..26442e83776 100644
> --- a/accel/tcg/translate-all.c
> +++ b/accel/tcg/translate-all.c
> @@ -349,7 +349,7 @@ TranslationBlock *tb_gen_code(CPUState *cpu,
> tcg_ctx->tlb_dyn_max_bits = CPU_TLB_DYN_MAX_BITS;
> #endif
> tcg_ctx->insn_start_words = TARGET_INSN_START_WORDS;
> - tcg_ctx->guest_mo = TCG_GUEST_DEFAULT_MO;
> + tcg_ctx->guest_mo = cpu->cc->tcg_ops->guest_default_memory_order;
>
> restart_translate:
> trace_translate_block(tb, pc, tb->tc.ptr);
> diff --git a/target/alpha/cpu.c b/target/alpha/cpu.c
> index 584c2aa76bd..e5e14976f51 100644
> --- a/target/alpha/cpu.c
> +++ b/target/alpha/cpu.c
> @@ -239,6 +239,9 @@ static const TCGCPUOps alpha_tcg_ops = {
> .synchronize_from_tb = alpha_cpu_synchronize_from_tb,
> .restore_state_to_opc = alpha_restore_state_to_opc,
>
> + /* Alpha processors have a weak memory model */
> + .guest_default_memory_order = 0,
> +
> #ifdef CONFIG_USER_ONLY
> .record_sigsegv = alpha_cpu_record_sigsegv,
> .record_sigbus = alpha_cpu_record_sigbus,
> diff --git a/target/arm/cpu.c b/target/arm/cpu.c
> index 01786ac7879..8b9f2acf82b 100644
> --- a/target/arm/cpu.c
> +++ b/target/arm/cpu.c
> @@ -2675,6 +2675,9 @@ static const TCGCPUOps arm_tcg_ops = {
> .debug_excp_handler = arm_debug_excp_handler,
> .restore_state_to_opc = arm_restore_state_to_opc,
>
> + /* ARM processors have a weak memory model */
> + .guest_default_memory_order = 0,
> +
> #ifdef CONFIG_USER_ONLY
> .record_sigsegv = arm_cpu_record_sigsegv,
> .record_sigbus = arm_cpu_record_sigbus,
> diff --git a/target/arm/tcg/cpu-v7m.c b/target/arm/tcg/cpu-v7m.c
> index c4dd3092726..df6b7198944 100644
> --- a/target/arm/tcg/cpu-v7m.c
> +++ b/target/arm/tcg/cpu-v7m.c
> @@ -238,6 +238,9 @@ static const TCGCPUOps arm_v7m_tcg_ops = {
> .debug_excp_handler = arm_debug_excp_handler,
> .restore_state_to_opc = arm_restore_state_to_opc,
>
> + /* ARM processors have a weak memory model */
> + .guest_default_memory_order = 0,
> +
> #ifdef CONFIG_USER_ONLY
> .record_sigsegv = arm_cpu_record_sigsegv,
> .record_sigbus = arm_cpu_record_sigbus,
> diff --git a/target/avr/cpu.c b/target/avr/cpu.c
> index 834c7082aa7..24e52e28f44 100644
> --- a/target/avr/cpu.c
> +++ b/target/avr/cpu.c
> @@ -216,6 +216,7 @@ static const TCGCPUOps avr_tcg_ops = {
> .cpu_exec_halt = avr_cpu_has_work,
> .tlb_fill = avr_cpu_tlb_fill,
> .do_interrupt = avr_cpu_do_interrupt,
> + .guest_default_memory_order = 0,
> };
>
> static void avr_cpu_class_init(ObjectClass *oc, void *data)
> diff --git a/target/hexagon/cpu.c b/target/hexagon/cpu.c
> index 766b6786511..34734b0edb0 100644
> --- a/target/hexagon/cpu.c
> +++ b/target/hexagon/cpu.c
> @@ -324,6 +324,8 @@ static const TCGCPUOps hexagon_tcg_ops = {
> .translate_code = hexagon_translate_code,
> .synchronize_from_tb = hexagon_cpu_synchronize_from_tb,
> .restore_state_to_opc = hexagon_restore_state_to_opc,
> + /* MTTCG not yet supported: require strict ordering */
> + .guest_default_memory_order = TCG_MO_ALL,
> };
>
> static void hexagon_cpu_class_init(ObjectClass *c, void *data)
> diff --git a/target/hppa/cpu.c b/target/hppa/cpu.c
> index 2a85495d02f..997bd69db19 100644
> --- a/target/hppa/cpu.c
> +++ b/target/hppa/cpu.c
> @@ -257,6 +257,14 @@ static const TCGCPUOps hppa_tcg_ops = {
> .synchronize_from_tb = hppa_cpu_synchronize_from_tb,
> .restore_state_to_opc = hppa_restore_state_to_opc,
>
> + /* PA-RISC 1.x processors have a strong memory model. */
> + /*
> + * ??? While we do not yet implement PA-RISC 2.0, those processors have
> + * a weak memory model, but with TLB bits that force ordering on a per-page
> + * basis. It's probably easier to fall back to a strong memory model.
> + */
> + .guest_default_memory_order = TCG_MO_ALL,
> +
> #ifndef CONFIG_USER_ONLY
> .tlb_fill_align = hppa_cpu_tlb_fill_align,
> .cpu_exec_interrupt = hppa_cpu_exec_interrupt,
> diff --git a/target/i386/tcg/tcg-cpu.c b/target/i386/tcg/tcg-cpu.c
> index b8aff825eec..4a76c475971 100644
> --- a/target/i386/tcg/tcg-cpu.c
> +++ b/target/i386/tcg/tcg-cpu.c
> @@ -128,6 +128,11 @@ static const TCGCPUOps x86_tcg_ops = {
> .debug_check_breakpoint = x86_debug_check_breakpoint,
> .need_replay_interrupt = x86_need_replay_interrupt,
> #endif /* !CONFIG_USER_ONLY */
> +
> + /*
> + * The x86 has a strong memory model with some store-after-load re-ordering
> + */
> + .guest_default_memory_order = TCG_MO_ALL & ~TCG_MO_ST_LD,
> };
>
> static void x86_tcg_cpu_init_ops(AccelCPUClass *accel_cpu, CPUClass *cc)
> diff --git a/target/loongarch/cpu.c b/target/loongarch/cpu.c
> index ea1665e2705..c39ff056157 100644
> --- a/target/loongarch/cpu.c
> +++ b/target/loongarch/cpu.c
> @@ -869,6 +869,8 @@ static const TCGCPUOps loongarch_tcg_ops = {
> .synchronize_from_tb = loongarch_cpu_synchronize_from_tb,
> .restore_state_to_opc = loongarch_restore_state_to_opc,
>
> + .guest_default_memory_order = 0,
> +
> #ifndef CONFIG_USER_ONLY
> .tlb_fill = loongarch_cpu_tlb_fill,
> .cpu_exec_interrupt = loongarch_cpu_exec_interrupt,
> diff --git a/target/m68k/cpu.c b/target/m68k/cpu.c
> index 0065e1c1ca5..e96b379e266 100644
> --- a/target/m68k/cpu.c
> +++ b/target/m68k/cpu.c
> @@ -593,6 +593,9 @@ static const TCGCPUOps m68k_tcg_ops = {
> .translate_code = m68k_translate_code,
> .restore_state_to_opc = m68k_restore_state_to_opc,
>
> + /* MTTCG not yet supported: require strict ordering */
> + .guest_default_memory_order = TCG_MO_ALL,
> +
> #ifndef CONFIG_USER_ONLY
> .tlb_fill = m68k_cpu_tlb_fill,
> .cpu_exec_interrupt = m68k_cpu_exec_interrupt,
> diff --git a/target/microblaze/cpu.c b/target/microblaze/cpu.c
> index f3bebea856e..4b9ef6e52c4 100644
> --- a/target/microblaze/cpu.c
> +++ b/target/microblaze/cpu.c
> @@ -432,6 +432,9 @@ static const TCGCPUOps mb_tcg_ops = {
> .synchronize_from_tb = mb_cpu_synchronize_from_tb,
> .restore_state_to_opc = mb_restore_state_to_opc,
>
> + /* MicroBlaze is always in-order. */
> + .guest_default_memory_order = TCG_MO_ALL,
> +
> #ifndef CONFIG_USER_ONLY
> .tlb_fill = mb_cpu_tlb_fill,
> .cpu_exec_interrupt = mb_cpu_exec_interrupt,
> diff --git a/target/mips/cpu.c b/target/mips/cpu.c
> index b207106dd79..5ddc9bbb829 100644
> --- a/target/mips/cpu.c
> +++ b/target/mips/cpu.c
> @@ -554,6 +554,8 @@ static const TCGCPUOps mips_tcg_ops = {
> .synchronize_from_tb = mips_cpu_synchronize_from_tb,
> .restore_state_to_opc = mips_restore_state_to_opc,
>
> + .guest_default_memory_order = 0,
> +
> #if !defined(CONFIG_USER_ONLY)
> .tlb_fill = mips_cpu_tlb_fill,
> .cpu_exec_interrupt = mips_cpu_exec_interrupt,
> diff --git a/target/openrisc/cpu.c b/target/openrisc/cpu.c
> index e8abf1f8b5c..6a878aaadd8 100644
> --- a/target/openrisc/cpu.c
> +++ b/target/openrisc/cpu.c
> @@ -248,6 +248,8 @@ static const TCGCPUOps openrisc_tcg_ops = {
> .synchronize_from_tb = openrisc_cpu_synchronize_from_tb,
> .restore_state_to_opc = openrisc_restore_state_to_opc,
>
> + .guest_default_memory_order = 0,
> +
> #ifndef CONFIG_USER_ONLY
> .tlb_fill = openrisc_cpu_tlb_fill,
> .cpu_exec_interrupt = openrisc_cpu_exec_interrupt,
> diff --git a/target/ppc/cpu_init.c b/target/ppc/cpu_init.c
> index 8b590e7f17c..28fbbb8d3c1 100644
> --- a/target/ppc/cpu_init.c
> +++ b/target/ppc/cpu_init.c
> @@ -7490,6 +7490,8 @@ static const TCGCPUOps ppc_tcg_ops = {
> .translate_code = ppc_translate_code,
> .restore_state_to_opc = ppc_restore_state_to_opc,
>
> + .guest_default_memory_order = 0,
> +
> #ifdef CONFIG_USER_ONLY
> .record_sigsegv = ppc_cpu_record_sigsegv,
> #else
> diff --git a/target/riscv/tcg/tcg-cpu.c b/target/riscv/tcg/tcg-cpu.c
> index 5aef9eef366..fb903992faa 100644
> --- a/target/riscv/tcg/tcg-cpu.c
> +++ b/target/riscv/tcg/tcg-cpu.c
> @@ -139,6 +139,8 @@ static const TCGCPUOps riscv_tcg_ops = {
> .synchronize_from_tb = riscv_cpu_synchronize_from_tb,
> .restore_state_to_opc = riscv_restore_state_to_opc,
>
> + .guest_default_memory_order = 0,
> +
> #ifndef CONFIG_USER_ONLY
> .tlb_fill = riscv_cpu_tlb_fill,
> .cpu_exec_interrupt = riscv_cpu_exec_interrupt,
> diff --git a/target/rx/cpu.c b/target/rx/cpu.c
> index 0ba0d55ab5b..6a24e7e9136 100644
> --- a/target/rx/cpu.c
> +++ b/target/rx/cpu.c
> @@ -212,6 +212,9 @@ static const TCGCPUOps rx_tcg_ops = {
> .cpu_exec_interrupt = rx_cpu_exec_interrupt,
> .cpu_exec_halt = rx_cpu_has_work,
> .do_interrupt = rx_cpu_do_interrupt,
> +
> + /* MTTCG not yet supported: require strict ordering */
> + .guest_default_memory_order = TCG_MO_ALL,
> };
>
> static void rx_cpu_class_init(ObjectClass *klass, void *data)
> diff --git a/target/s390x/cpu.c b/target/s390x/cpu.c
> index d73142600bf..12fd853c00a 100644
> --- a/target/s390x/cpu.c
> +++ b/target/s390x/cpu.c
> @@ -360,6 +360,12 @@ static const TCGCPUOps s390_tcg_ops = {
> .debug_excp_handler = s390x_cpu_debug_excp_handler,
> .do_unaligned_access = s390x_cpu_do_unaligned_access,
> #endif /* !CONFIG_USER_ONLY */
> +
> + /*
> + * The z/Architecture has a strong memory model with some
> + * store-after-load re-ordering.
> + */
> + .guest_default_memory_order = TCG_MO_ALL & ~TCG_MO_ST_LD,
> };
> #endif /* CONFIG_TCG */
>
> diff --git a/target/sh4/cpu.c b/target/sh4/cpu.c
> index ce84bdf539a..ce9ed75107a 100644
> --- a/target/sh4/cpu.c
> +++ b/target/sh4/cpu.c
> @@ -267,6 +267,9 @@ static const TCGCPUOps superh_tcg_ops = {
> .synchronize_from_tb = superh_cpu_synchronize_from_tb,
> .restore_state_to_opc = superh_restore_state_to_opc,
>
> + /* MTTCG not yet supported: require strict ordering */
> + .guest_default_memory_order = TCG_MO_ALL,
> +
> #ifndef CONFIG_USER_ONLY
> .tlb_fill = superh_cpu_tlb_fill,
> .cpu_exec_interrupt = superh_cpu_exec_interrupt,
> diff --git a/target/sparc/cpu.c b/target/sparc/cpu.c
> index 57161201173..39bd0c42855 100644
> --- a/target/sparc/cpu.c
> +++ b/target/sparc/cpu.c
> @@ -1005,6 +1005,29 @@ static const TCGCPUOps sparc_tcg_ops = {
> .synchronize_from_tb = sparc_cpu_synchronize_from_tb,
> .restore_state_to_opc = sparc_restore_state_to_opc,
>
> + /*
> + * From Oracle SPARC Architecture 2015:
> + *
> + * Compatibility notes: The PSO memory model described in SPARC V8 and
> + * SPARC V9 compatibility architecture specifications was never
> + * implemented in a SPARC V9 implementation and is not included in the
> + * Oracle SPARC Architecture specification.
> + *
> + * The RMO memory model described in the SPARC V9 specification was
> + * implemented in some non-Sun SPARC V9 implementations, but is not
> + * directly supported in Oracle SPARC Architecture 2015 implementations.
> + *
> + * Therefore always use TSO in QEMU.
> + *
> + * D.5 Specification of Partial Store Order (PSO)
> + * ... [loads] are followed by an implied MEMBAR #LoadLoad | #LoadStore.
> + *
> + * D.6 Specification of Total Store Order (TSO)
> + * ... PSO with the additional requirement that all [stores] are followed
> + * by an implied MEMBAR #StoreStore.
> + */
> + .guest_default_memory_order = TCG_MO_LD_LD | TCG_MO_LD_ST | TCG_MO_ST_ST,
> +
> #ifndef CONFIG_USER_ONLY
> .tlb_fill = sparc_cpu_tlb_fill,
> .cpu_exec_interrupt = sparc_cpu_exec_interrupt,
> diff --git a/target/tricore/cpu.c b/target/tricore/cpu.c
> index 16acc4ecb92..e0a48065948 100644
> --- a/target/tricore/cpu.c
> +++ b/target/tricore/cpu.c
> @@ -179,6 +179,8 @@ static const TCGCPUOps tricore_tcg_ops = {
> .tlb_fill = tricore_cpu_tlb_fill,
> .cpu_exec_interrupt = tricore_cpu_exec_interrupt,
> .cpu_exec_halt = tricore_cpu_has_work,
> + /* MTTCG not yet supported: require strict ordering */
> + .guest_default_memory_order = TCG_MO_ALL,
> };
>
> static void tricore_cpu_class_init(ObjectClass *c, void *data)
> diff --git a/target/xtensa/cpu.c b/target/xtensa/cpu.c
> index 7663b62d01e..dd9061ba469 100644
> --- a/target/xtensa/cpu.c
> +++ b/target/xtensa/cpu.c
> @@ -237,6 +237,9 @@ static const TCGCPUOps xtensa_tcg_ops = {
> .debug_excp_handler = xtensa_breakpoint_handler,
> .restore_state_to_opc = xtensa_restore_state_to_opc,
>
> + /* Xtensa processors have a weak memory model */
> + .guest_default_memory_order = 0,
> +
> #ifndef CONFIG_USER_ONLY
> .tlb_fill = xtensa_cpu_tlb_fill,
> .cpu_exec_interrupt = xtensa_cpu_exec_interrupt,
Excellent.
It's definitely a nice paradigm to move those things here instead of
having them optionally defined in headers. I've been bitten by that on
the latest series remove cpu-all.h.
Reviewed-by: Pierrick Bouvier <pierrick.bouvier@linaro.org>
^ permalink raw reply [flat|nested] 19+ messages in thread
end of thread, other threads:[~2025-03-21 17:47 UTC | newest]
Thread overview: 19+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-03-21 12:57 [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order Philippe Mathieu-Daudé
2025-03-21 12:57 ` [PATCH-for-10.1 1/4] tcg: Always define TCG_GUEST_DEFAULT_MO Philippe Mathieu-Daudé
2025-03-21 14:39 ` Anton Johansson via
2025-03-21 15:14 ` Richard Henderson
2025-03-21 17:44 ` Pierrick Bouvier
2025-03-21 12:57 ` [PATCH-for-10.1 2/4] tcg: Simplify tcg_req_mo() macro Philippe Mathieu-Daudé
2025-03-21 14:40 ` Anton Johansson via
2025-03-21 15:14 ` Richard Henderson
2025-03-21 17:44 ` Pierrick Bouvier
2025-03-21 12:57 ` [RFC PATCH-for-10.1 3/4] tcg: Have tcg_req_mo() use runtime TCGContext::guest_mo Philippe Mathieu-Daudé
2025-03-21 14:39 ` Anton Johansson via
2025-03-21 14:46 ` Philippe Mathieu-Daudé
2025-03-21 15:16 ` Richard Henderson
2025-03-21 12:57 ` [RFC PATCH-for-10.1 4/4] tcg: Define guest_default_memory_order in TCGCPUOps Philippe Mathieu-Daudé
2025-03-21 13:20 ` Philippe Mathieu-Daudé
2025-03-21 14:34 ` Anton Johansson via
2025-03-21 15:25 ` Richard Henderson
2025-03-21 17:47 ` Pierrick Bouvier
2025-03-21 14:40 ` [RFC PATCH-for-10.1 0/4] tcg: Move TCG_GUEST_DEFAULT_MO -> TCGCPUOps::guest_default_memory_order Anton Johansson via
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).