Re: [PATCH] riscv: VMAP_STACK overflow detection thread-safe

From: Deepak Gupta
Date: Thu Nov 24 2022 - 02:27:30 EST


On Thu, Nov 24, 2022 at 02:31:25PM +0800, Guo Ren wrote:
On Thu, Nov 24, 2022 at 1:57 PM Deepak Gupta <debug@xxxxxxxxxxxx> wrote:

Inline.

On Wed, Nov 23, 2022 at 5:28 PM Guo Ren <guoren@xxxxxxxxxx> wrote:
>
>
>
> On Thu, Nov 24, 2022 at 8:50 AM Deepak Gupta <debug@xxxxxxxxxxxx> wrote:
>>
>> commit 31da94c25aea835ceac00575a9fd206c5a833fed added support for
>> CONFIG_VMAP_STACK. If overflow is detected, CPU switches to `shadow_stack`
>> temporarily before switching finally to per-cpu `overflow_stack`.
>>
>> If two CPUs/harts are racing and end up in over flowing kernel stack, one
>> or both will end up corrupting each other state because `shadow_stack` is
>> not per-cpu.
>>
>> Following are the changes in this patch
>>
>> - Defines an asm macro to obtain per-cpu symbols in destination
>> register.
>> - Computing per-cpu symbol requires a temporary register. When stack is
>> out of question, a place is needed to spill a register. `thread_info`
>> is good location to have spill register.
>> - In entry.S when overflow is detected x31 is spilled into thread_info.
>> x31 is used as temp reg for asm macro to locate per-cpu overflow stack
>>
>> Other relevant disccussion on this
>> https://lore.kernel.org/linux-riscv/Y347B0x4VUNOd6V7@xhacker/T/#t
>>
>> Tested by `echo EXHAUST_STACK > /sys/kernel/debug/provoke-crash/DIRECT`
>>
>> [ 286.223273] Insufficient stack space to handle exception!/debug/provoke-crash/DIRECT
>> [ 286.223878] Task stack: [0xff20000010a98000..0xff20000010a9c000]
>> [ 286.224411] Overflow stack: [0xff600001f7d98370..0xff600001f7d99370]
>> [ 286.226057] CPU: 1 PID: 205 Comm: bash Not tainted 6.1.0-rc2-00001-g328a1f96f7b9 #34
>> [ 286.227139] Hardware name: riscv-virtio,qemu (DT)
>> [ 286.228000] epc : __memset+0x60/0xfc
>> [ 286.229299] ra : recursive_loop+0x48/0xc6 [lkdtm]
>> [ 286.231457] epc : ffffffff808de0e4 ra : ffffffff0163a752 sp : ff20000010a97e80
>> [ 286.232207] gp : ffffffff815c0330 tp : ff600000820ea280 t0 : ff20000010a97e88
>> [ 286.233584] t1 : 000000000000002e t2 : 3233206874706564 s0 : ff20000010a982b0
>> [ 286.234293] s1 : 0000000000000012 a0 : ff20000010a97e88 a1 : 0000000000000000
>> [ 286.234998] a2 : 0000000000000400 a3 : ff20000010a98288 a4 : 0000000000000000
>> [ 286.235697] a5 : 0000000000000000 a6 : fffffffffffe43f0 a7 : 00007fffffffffff
>> [ 286.236384] s2 : ff20000010a97e88 s3 : ffffffff01644680 s4 : ff20000010a9be90
>> [ 286.237743] s5 : ff600000842ba6c0 s6 : 00aaaaaac29e42b0 s7 : 00fffffff0aa3684
>> [ 286.238691] s8 : 00aaaaaac2978040 s9 : 0000000000000065 s10: 00ffffff8a7cad10
>> [ 286.239591] s11: 00ffffff8a76a4e0 t3 : ffffffff815dbaf4 t4 : ffffffff815dbaf4
>> [ 286.240537] t5 : ffffffff815dbab8 t6 : ff20000010a9bb48
>> [ 286.241540] status: 0000000200000120 badaddr: ff20000010a97e88 cause: 000000000000000f
>> [ 286.242979] Kernel panic - not syncing: Kernel stack overflow
>> [ 286.244106] CPU: 1 PID: 205 Comm: bash Not tainted 6.1.0-rc2-00001-g328a1f96f7b9 #34
>> [ 286.245276] Hardware name: riscv-virtio,qemu (DT)
>> [ 286.245929] Call Trace:
>> [ 286.246954] [<ffffffff80006754>] dump_backtrace+0x30/0x38
>> [ 286.247813] [<ffffffff808de798>] show_stack+0x40/0x4c
>> [ 286.248429] [<ffffffff808ea2a8>] dump_stack_lvl+0x44/0x5c
>> [ 286.249439] [<ffffffff808ea2d8>] dump_stack+0x18/0x20
>> [ 286.250056] [<ffffffff808dec06>] panic+0x126/0x2fe
>> [ 286.250642] [<ffffffff800065ea>] walk_stackframe+0x0/0xf0
>> [ 286.251357] [<ffffffff0163a752>] recursive_loop+0x48/0xc6 [lkdtm]
>> [ 286.253321] SMP: stopping secondary CPUs
>> [ 286.256724] ---[ end Kernel panic - not syncing: Kernel stack overflow ]---
>>
>> Fixes: 31da94c25aea835ceac00575a9fd206c5a833fed
>
> The patch gives more significant change than the Fixes, and Fixes would expand to the previous stable versions. Please don't set it as a Fixes, but an improved OVERSTACK dead path performance feature.
>

Not a performance feature but more like correctness.
If kernel died and two CPUs raced to kernel stack overflow,
death post-mortem should be straightforward.
We already have had a fixup, and your patch likes a feature with a
significant change.
https://lore.kernel.org/linux-riscv/20221030124517.2370-1-jszhang@xxxxxxxxxx/
If it is for correctness, the simple lock is enough.

Sure lock is enough. It's different way to solve the problem. But I don't think it qualifies as significant change.





>>
>> Cc: Guo Ren <guoren@xxxxxxxxxx>
>> Cc: Jisheng Zhang <jszhang@xxxxxxxxxx>
>>
>> Signed-off-by: Deepak Gupta <debug@xxxxxxxxxxxx>
>> ---
>> arch/riscv/include/asm/asm.h | 11 ++++++
>> arch/riscv/include/asm/thread_info.h | 3 ++
>> arch/riscv/kernel/asm-offsets.c | 4 +++
>> arch/riscv/kernel/entry.S | 54 ++++------------------------
>> arch/riscv/kernel/traps.c | 12 +------
>> 5 files changed, 26 insertions(+), 58 deletions(-)
>>
>> diff --git a/arch/riscv/include/asm/asm.h b/arch/riscv/include/asm/asm.h
>> index 1b471ff73178..373eba843331 100644
>> --- a/arch/riscv/include/asm/asm.h
>> +++ b/arch/riscv/include/asm/asm.h
>> @@ -69,6 +69,7 @@
>>
>> #ifdef __ASSEMBLY__
>>
>> +#include <asm/asm-offsets.h>
>> /* Common assembly source macros */
>>
>> /*
>> @@ -80,6 +81,16 @@
>> .endr
>> .endm
>>
>> +.macro asm_per_cpu dst sym tmp
>> + REG_L \tmp, TASK_TI_CPU_NUM(tp)
>> + slli \tmp, \tmp, 0x3
>> + la \dst, __per_cpu_offset
>> + add \dst, \dst, \tmp
>> + REG_L \tmp, 0(\dst)
>> + la \dst, \sym
>> + add \dst, \dst, \tmp
>
> Another tricky asm code of using percpu, I don't know how the percpu maintenance guy thinks.

Pretty much the same exists for arm64 as well.
That's not the reason.

I don't see anything tricky here.
I prefer using standard percpu API, not hacking one.

`this_cpu_ptr` is not an API. It's also a C macro which expands similar as above written asm macro. First time that macro was written, it was a hack and now it's an API.



>
> Percpu is really complex to be used here. That's why I don't like Jisheng's previous idea, although I haven't given an explicitly opposed comment.
>

Why is per-cpu really complex here?
That's why per-cpu exists, right?
Compiler generates pretty much same code everywhere for C.

> Keep simple and direct.
>
>>
>> +.endm
>> +
>> #endif /* __ASSEMBLY__ */
>>
>> #endif /* _ASM_RISCV_ASM_H */
>> diff --git a/arch/riscv/include/asm/thread_info.h b/arch/riscv/include/asm/thread_info.h
>> index 67322f878e0d..7e17dc07cf11 100644
>> --- a/arch/riscv/include/asm/thread_info.h
>> +++ b/arch/riscv/include/asm/thread_info.h
>> @@ -65,6 +65,9 @@ struct thread_info {
>> */
>> long kernel_sp; /* Kernel stack pointer */
>> long user_sp; /* User stack pointer */
>> +#ifdef CONFIG_VMAP_STACK
>> + long spill_reg; /* per cpu scratch space to spill a single register */
>> +#endif
>
> Is it worth expanding "struct thread_info" with another reg size? A dead path performance issue affects living cost. May we give a cheap funeral for the OVERFLOW_STACK death?
>

Yes I had been thinking over it and your point is quite valid.
There is a death's charge (in a corner case) to living structures
(particularly to all threads)
Andrew Bresticker (abrestic@xxxxxxxxxxxx) suggested using `sscratch`
to spill into.
I think that makes sense.

I'll send a revised patch.
I think that's not a fix because your patch broke the previous author's design.


If that's the concern, I can send it out as "not a fix" but "alternative solution".

Below is the fixup style:
https://lore.kernel.org/linux-riscv/20221030124517.2370-1-jszhang@xxxxxxxxxx/
Just fix the previous author's design with a simple spinlock?


>>
>> int cpu;
>> };
>>
>> diff --git a/arch/riscv/kernel/asm-offsets.c b/arch/riscv/kernel/asm-offsets.c
>> index df9444397908..bed3c83bfb8f 100644
>> --- a/arch/riscv/kernel/asm-offsets.c
>> +++ b/arch/riscv/kernel/asm-offsets.c
>> @@ -38,6 +38,10 @@ void asm_offsets(void)
>> OFFSET(TASK_TI_KERNEL_SP, task_struct, thread_info.kernel_sp);
>> OFFSET(TASK_TI_USER_SP, task_struct, thread_info.user_sp);
>>
>> + OFFSET(TASK_TI_CPU_NUM, task_struct, thread_info.cpu);
>> +#ifdef CONFIG_VMAP_STACK
>> + OFFSET(TASK_TI_SPILL_REG, task_struct, thread_info.spill_reg);
>> +#endif
>> OFFSET(TASK_THREAD_F0, task_struct, thread.fstate.f[0]);
>> OFFSET(TASK_THREAD_F1, task_struct, thread.fstate.f[1]);
>> OFFSET(TASK_THREAD_F2, task_struct, thread.fstate.f[2]);
>> diff --git a/arch/riscv/kernel/entry.S b/arch/riscv/kernel/entry.S
>> index b9eda3fcbd6d..12f285cec136 100644
>> --- a/arch/riscv/kernel/entry.S
>> +++ b/arch/riscv/kernel/entry.S
>> @@ -10,9 +10,11 @@
>> #include <asm/asm.h>
>> #include <asm/csr.h>
>> #include <asm/unistd.h>
>> +#include <asm/page.h>
>> #include <asm/thread_info.h>
>> #include <asm/asm-offsets.h>
>> #include <asm/errata_list.h>
>> +#include <linux/sizes.h>
>>
>> #if !IS_ENABLED(CONFIG_PREEMPTION)
>> .set resume_kernel, restore_all
>> @@ -404,54 +406,12 @@ handle_syscall_trace_exit:
>>
>> #ifdef CONFIG_VMAP_STACK
>> handle_kernel_stack_overflow:
>> - la sp, shadow_stack
>> - addi sp, sp, SHADOW_OVERFLOW_STACK_SIZE
>> + REG_S x31, TASK_TI_SPILL_REG(tp)
>> + asm_per_cpu sp, overflow_stack, x31
>> + li x31, OVERFLOW_STACK_SIZE
>> + add sp, sp, x31
>> + REG_L x31, TASK_TI_SPILL_REG(tp)
>>
>> - //save caller register to shadow stack
>> - addi sp, sp, -(PT_SIZE_ON_STACK)
>> - REG_S x1, PT_RA(sp)
>> - REG_S x5, PT_T0(sp)
>> - REG_S x6, PT_T1(sp)
>> - REG_S x7, PT_T2(sp)
>> - REG_S x10, PT_A0(sp)
>> - REG_S x11, PT_A1(sp)
>> - REG_S x12, PT_A2(sp)
>> - REG_S x13, PT_A3(sp)
>> - REG_S x14, PT_A4(sp)
>> - REG_S x15, PT_A5(sp)
>> - REG_S x16, PT_A6(sp)
>> - REG_S x17, PT_A7(sp)
>> - REG_S x28, PT_T3(sp)
>> - REG_S x29, PT_T4(sp)
>> - REG_S x30, PT_T5(sp)
>> - REG_S x31, PT_T6(sp)
>> -
>> - la ra, restore_caller_reg
>> - tail get_overflow_stack
>> -
>> -restore_caller_reg:
>> - //save per-cpu overflow stack
>> - REG_S a0, -8(sp)
>> - //restore caller register from shadow_stack
>> - REG_L x1, PT_RA(sp)
>> - REG_L x5, PT_T0(sp)
>> - REG_L x6, PT_T1(sp)
>> - REG_L x7, PT_T2(sp)
>> - REG_L x10, PT_A0(sp)
>> - REG_L x11, PT_A1(sp)
>> - REG_L x12, PT_A2(sp)
>> - REG_L x13, PT_A3(sp)
>> - REG_L x14, PT_A4(sp)
>> - REG_L x15, PT_A5(sp)
>> - REG_L x16, PT_A6(sp)
>> - REG_L x17, PT_A7(sp)
>> - REG_L x28, PT_T3(sp)
>> - REG_L x29, PT_T4(sp)
>> - REG_L x30, PT_T5(sp)
>> - REG_L x31, PT_T6(sp)
>> -
>> - //load per-cpu overflow stack
>> - REG_L sp, -8(sp)
>
> Yes, your patch gives a good performance here. But what do we gain?
>
>>
>> addi sp, sp, -(PT_SIZE_ON_STACK)
>>
>> //save context to overflow stack
>> diff --git a/arch/riscv/kernel/traps.c b/arch/riscv/kernel/traps.c
>> index f3e96d60a2ff..eef3a87514c7 100644
>> --- a/arch/riscv/kernel/traps.c
>> +++ b/arch/riscv/kernel/traps.c
>> @@ -208,18 +208,8 @@ int is_valid_bugaddr(unsigned long pc)
>> #endif /* CONFIG_GENERIC_BUG */
>>
>> #ifdef CONFIG_VMAP_STACK
>> -static DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)],
>> +DEFINE_PER_CPU(unsigned long [OVERFLOW_STACK_SIZE/sizeof(long)],
>> overflow_stack)__aligned(16);
>> -/*
>> - * shadow stack, handled_ kernel_ stack_ overflow(in kernel/entry.S) is used
>> - * to get per-cpu overflow stack(get_overflow_stack).
>> - */
>> -long shadow_stack[SHADOW_OVERFLOW_STACK_SIZE/sizeof(long)];
>> -asmlinkage unsigned long get_overflow_stack(void)
>> -{
>> - return (unsigned long)this_cpu_ptr(overflow_stack) +
>> - OVERFLOW_STACK_SIZE;
>> -}
>>
>> asmlinkage void handle_bad_stack(struct pt_regs *regs)
>> {
>> --
>> 2.25.1
>>
>
>
> --
> Best Regards
> Guo Ren



--
Best Regards
Guo Ren