riscv: setup per-hart stack earlier
PATCH RESEND v2 1/1 riscv: setup per-hart stack earlier
Harts need to use per-hart stack before any function call, even if that function is a simple one. When the callee uses stack for register save/ restore, especially RA, if nested call, concurrent access by multiple harts on the same stack will cause data-race.
This patch sets up SP before board_init_f_alloc_reserve. A side effect of this is that the memory layout has changed as the following:
+----------------+ +----------------+ <----- SPL_STACK/ | ...... | | hart 0 stack | SYS_INIT_SP_ADDR | malloc_base | +----------------+ +----------------+ | hart 1 stack | | GD | +----------------+ If not SMP, N=1 +----------------+ | ...... | | hart 0 stack | +----------------+ +----------------+ ==> | hart N-1 stack| | hart 1 stack | +----------------+ +----------------+ | ...... | | ...... | | malloc_base | +----------------+ +----------------+ | hart N-1 stack| | GD | +----------------+ +----------------+ | | | |
---
v2:
- Fixed macro CONFIG_SYS_INIT_SP_ADDR -> SYS_INIT_SP_ADDR
- Tested SPL with VisionFive 2 board
--- arch/riscv/cpu/start.S | 37 ++++++++++++++++++++++++------------- 1 file changed, 24 insertions(+), 13 deletions(-)
arch/riscv/cpu/start.S +24 -13
@@ -91,16 +91,35 @@ _start: * Set stackpointer in internal/ex RAM to call board_init_f */ call_board_init_f: - li t0, -16 #if defined(CONFIG_SPL_BUILD) && defined(CONFIG_SPL_STACK) - li t1, CONFIG_SPL_STACK + li t0, CONFIG_SPL_STACK #else - li t1, SYS_INIT_SP_ADDR + li t0, SYS_INIT_SP_ADDR #endif - and sp, t1, t0 /* force 16 byte alignment */ + and t0, t0, -16 /* force 16 byte alignment */ + + /* setup stack */ +#if CONFIG_IS_ENABLED(SMP) + /* tp: hart id */ + slli t1, tp, CONFIG_STACK_SIZE_SHIFT + sub sp, t0, t1 +#else + mv sp, t0 +#endif +/* + * Now sp points to the right stack belonging to current CPU. + * It's essential before any function call, otherwise, we get data-race. + */ call_board_init_f_0: - mv a0, sp + /* find top of reserve space */ +#if CONFIG_IS_ENABLED(SMP) + li t1, CONFIG_NR_CPUS +#else + li t1, 1 +#endif + slli t1, t1, CONFIG_STACK_SIZE_SHIFT + sub a0, t0, t1 /* t1 -> size of all CPU stacks */ jal board_init_f_alloc_reserve /*
@@ -109,14 +128,6 @@ call_board_init_f_0: */ mv s0, a0 - /* setup stack */ -#if CONFIG_IS_ENABLED(SMP) - /* tp: hart id */ - slli t0, tp, CONFIG_STACK_SIZE_SHIFT - sub sp, a0, t0 -#else - mv sp, a0 -#endif /* Configure proprietary settings and customized CSRs of harts */ call_harts_early_init:
Signature · 1 line
2.7.4
A message that is not in the archive (sent off-list or not yet received): SEZPR03MB8064071D92E63D7A4C631359C154A@SEZPR03MB8064.apcprd03.prod.outlook.com
Rick Chen
Reviewed-byQuoted 32 lines
From: Bo Gan <[email protected]>
Sent: Monday, June 12, 2023 7:54 AM
To: [email protected]
Cc: Bo Gan <[email protected]>; Rick Jian-Zhi Chen(陳建志) <[email protected]>; Leo Yu-Chi Liang(梁育齊) <[email protected]>; Sean Anderson <[email protected]>; Bin Meng <[email protected]>; Lukas Auer <[email protected]>
Subject: [RESEND PATCH v2] riscv: setup per-hart stack earlierHarts need to use per-hart stack before any function call, even if that function is a simple one. When the callee uses stack for register save/ restore, especially RA, if nested call, concurrent access by multiple harts on the same stack will cause data-race.
This patch sets up SP before
board_init_f_alloc_reserve. A side effect of this is that the memory layout has changed as the following:+----------------+ +----------------+ <----- SPL_STACK/ | ...... | | hart 0 stack | SYS_INIT_SP_ADDR | malloc_base | +----------------+ +----------------+ | hart 1 stack | | GD | +----------------+ If not SMP, N=1 +----------------+ | ...... | | hart 0 stack | +----------------+ +----------------+ ==> | hart N-1 stack| | hart 1 stack | +----------------+ +----------------+ | ...... | | ...... | | malloc_base | +----------------+ +----------------+ | hart N-1 stack| | GD | +----------------+ +----------------+ | | | |Signed-off-by Bo Gan <[email protected]>Cc Rick Chen <[email protected]>Cc Leo <[email protected]>Cc Sean Anderson <[email protected]>Cc Bin Meng <[email protected]>Cc Lukas Auer <[email protected]>
Quoted 8 lines
---
v2:
- Fixed macroCONFIG_SYS_INIT_SP_ADDR->SYS_INIT_SP_ADDR
- Tested SPL with VisionFive 2 board--- arch/riscv/cpu/start.S | 37 ++++++++++++++++++++++++------------- 1 file changed, 24 insertions(+), 13 deletions(-)
Quoted 31 lines · Bo Gan
Harts need to use per-hart stack before any function call, even if that function is a simple one. When the callee uses stack for register save/ restore, especially RA, if nested call, concurrent access by multiple harts on the same stack will cause data-race.
This patch sets up SP before
board_init_f_alloc_reserve. A side effect of this is that the memory layout has changed as the following:+----------------+ +----------------+ <----- SPL_STACK/ | ...... | | hart 0 stack | SYS_INIT_SP_ADDR | malloc_base | +----------------+ +----------------+ | hart 1 stack | | GD | +----------------+ If not SMP, N=1 +----------------+ | ...... | | hart 0 stack | +----------------+ +----------------+ ==> | hart N-1 stack| | hart 1 stack | +----------------+ +----------------+ | ...... | | ...... | | malloc_base | +----------------+ +----------------+ | hart N-1 stack| | GD | +----------------+ +----------------+ | | | |Signed-off-by Bo Gan <[email protected]>Cc Rick Chen <[email protected]>Cc Leo <[email protected]>Cc Sean Anderson <[email protected]>Cc Bin Meng <[email protected]>Cc Lukas Auer <[email protected]>Reviewed-by Rick Chen <[email protected]>
Quoted 8 lines
---
v2:
- Fixed macroCONFIG_SYS_INIT_SP_ADDR->SYS_INIT_SP_ADDR
- Tested SPL with VisionFive 2 board--- arch/riscv/cpu/start.S | 37 ++++++++++++++++++++++++------------- 1 file changed, 24 insertions(+), 13 deletions(-)