riscv: asid: Fixup stale TLB entry cause application crash
PATCH 1/1 riscv: asid: Fixup stale TLB entry cause application crash
From: Guo Ren <[email protected]>
When use_asid_allocator is enabled, the userspace application would crash by stale tlb entry. Because only using cpumask_clear_cpu without local_flush_tlb_all couldn't guarantee CPU's tlb contains fresh mapping entry. Then set_mm_asid would cause user space application get a stale value by stale tlb entry, but set_mm_noasid is okay.
Here is the symptom of the bug:
unhandled signal 11 code 0x1 (coredump)
0x0000003fd6d22524 <+4>: auipc s0,0x70 0x0000003fd6d22528 <+8>: ld s0,-148(s0) # 0x3fd6d92490 => 0x0000003fd6d2252c <+12>: ld a5,0(s0)
(gdb) i r s0
s0 0x8082ed1cc3198b21 0x8082ed1cc3198b21
(gdb) x/16 0x3fd6d924900x3fd6d92490: 0xd80ac8a8 0x0000003f
The core dump file show us the value of register s0 is wrong, but the value in memory is right.
When task run on CPU0, the task loaded/speculative-loaded the value of address-0x3fd6d92490, the first version tlb mapping enter in CPU0's tlb. When the task switched from CPU0 to CPU1 without local_tlb_flush_all (because of asid), the task happened to write a value on address:0x3fd6d92490 that caused do_page_fault -> wp_page_copy -> ptep_clear_flush -> ptep_get_and_clear & flush_tlb_page.
The flush_tlb_page would use mm_cpumask(mm) to determine which CPUs need tlb flush, but CPU0 cleared the CPU0's mm_cpumask in previous switch_mm. So we only flushed the CPU1's tlb entry, and setted second version mapping of the pte. When the task switch from CPU1 to CPU0 again, it still used a stale tlb entry on CPU0 which contained a wrong target physical address. When the task happened to read that value, the bug would be raised.
--- arch/riscv/mm/context.c | 1 - 1 file changed, 1 deletion(-)
arch/riscv/mm/context.c +0 -1
@@ -317,7 +317,6 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,*/ cpu =smp_processor_id (); -cpumask_clear_cpu (cpu,mm_cpumask (prev));cpumask_set_cpu (cpu,mm_cpumask (next));set_mm (next, cpu);
Signature · 1 line
2.36.1
Quoted 48 lines · On Tue, Nov 8, 2022 at 5:02 PM <[email protected]> wrote:
From: Guo Ren <[email protected]>
When
use_asid_allocatoris enabled, the userspace application would crash by stale tlb entry. Because only usingcpumask_clear_cpuwithoutlocal_flush_tlb_allcouldn't guarantee CPU's tlb contains fresh mapping entry. Thenset_mm_asidwould cause user space application get a stale value by stale tlb entry, butset_mm_noasidis okay.Here is the symptom of the bug:
unhandled signal 11 code0x1(coredump)0x0000003fd6d22524 <+4>: auipc s0,0x70 0x0000003fd6d22528 <+8>: ld s0,-148(s0) # 0x3fd6d92490 => 0x0000003fd6d2252c <+12>: ld a5,0(s0)(gdb) i r s0
s00x8082ed1cc3198b210x8082ed1cc3198b21
(gdb) x/160x3fd6d924900x3fd6d92490:0xd80ac8a80x0000003f
The core dump file show us the value of register s0 is wrong, but the value in memory is right.When task run on CPU0, the task loaded/speculative-loaded the value of address-0x3fd6d92490, the first version tlb mapping enter in CPU0's tlb. When the task switched from CPU0 to CPU1 without
local_tlb_flush_all(because of asid), the task happened to write a value on address:0x3fd6d92490that causeddo_page_fault->wp_page_copy->ptep_clear_flush->ptep_get_and_clear&flush_tlb_page.
Theflush_tlb_pagewould usemm_cpumask(mm)to determine which CPUs need tlb flush, but CPU0 cleared the CPU0'smm_cpumaskin previousswitch_mm. So we only flushed the CPU1's tlb entry, and setted second version mapping of the pte. When the task switch from CPU1 to CPU0 again, it still used a stale tlb entry on CPU0 which contained a wrong target physical address. When the task happened to read that value, the bug would be raised.Fixes 65d4b9c53017 ("RISC-V: Implement ASID allocator")Signed-off-by Guo Ren <[email protected]>Signed-off-by Guo Ren <[email protected]>Cc Anup Patel <[email protected]>Cc Palmer Dabbelt <[email protected]>--- arch/riscv/mm/context.c | 1 - 1 file changed, 1 deletion(-)arch/riscv/mm/context.c +0
-0
arch/riscv/mm/context.c void switch_mm(struct mm_struct *prev, struct mm_struct *next, · line 320
Quoted 4 lines
arch/riscv/mm/context.c +0
-1@@ -317,7 +317,6 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,*/ cpu =smp_processor_id (); -cpumask_clear_cpu (cpu,mm_cpumask (prev));
It should be:
- cpumask_clear_cpu(cpu, mm_cpumask(prev));
+ if (static_branch_unlikely(&use_asid_allocator))
+ cpumask_clear_cpu(cpu, mm_cpumask(prev));
+
No bug on noasid part.
Quoted 6 lines
arch/riscv/mm/context.c +0
-1cpumask_set_cpu (cpu,mm_cpumask (next));set_mm (next, cpu); --2.36.1
Signature · 2 lines
Best Regards Guo Ren
Quoted 56 lines · Guo Ren
Quoted 48 lines · On Tue, Nov 8, 2022 at 5:02 PM <[email protected]> wrote:
From: Guo Ren <[email protected]>
When
use_asid_allocatoris enabled, the userspace application would crash by stale tlb entry. Because only usingcpumask_clear_cpuwithoutlocal_flush_tlb_allcouldn't guarantee CPU's tlb contains fresh mapping entry. Thenset_mm_asidwould cause user space application get a stale value by stale tlb entry, butset_mm_noasidis okay.Here is the symptom of the bug:
unhandled signal 11 code0x1(coredump)0x0000003fd6d22524 <+4>: auipc s0,0x70 0x0000003fd6d22528 <+8>: ld s0,-148(s0) # 0x3fd6d92490 => 0x0000003fd6d2252c <+12>: ld a5,0(s0)(gdb) i r s0
s00x8082ed1cc3198b210x8082ed1cc3198b21
(gdb) x/160x3fd6d924900x3fd6d92490:0xd80ac8a80x0000003f
The core dump file show us the value of register s0 is wrong, but the value in memory is right.When task run on CPU0, the task loaded/speculative-loaded the value of address-0x3fd6d92490, the first version tlb mapping enter in CPU0's tlb. When the task switched from CPU0 to CPU1 without
local_tlb_flush_all(because of asid), the task happened to write a value on address:0x3fd6d92490that causeddo_page_fault->wp_page_copy->ptep_clear_flush->ptep_get_and_clear&flush_tlb_page.
Theflush_tlb_pagewould usemm_cpumask(mm)to determine which CPUs need tlb flush, but CPU0 cleared the CPU0'smm_cpumaskin previousswitch_mm. So we only flushed the CPU1's tlb entry, and setted second version mapping of the pte. When the task switch from CPU1 to CPU0 again, it still used a stale tlb entry on CPU0 which contained a wrong target physical address. When the task happened to read that value, the bug would be raised.Fixes 65d4b9c53017 ("RISC-V: Implement ASID allocator")Signed-off-by Guo Ren <[email protected]>Signed-off-by Guo Ren <[email protected]>Cc Anup Patel <[email protected]>Cc Palmer Dabbelt <[email protected]>--- arch/riscv/mm/context.c | 1 - 1 file changed, 1 deletion(-)arch/riscv/mm/context.c +0
-0arch/riscv/mm/context.c void switch_mm(struct mm_struct *prev, struct mm_struct *next, · line 320
Quoted 4 lines
arch/riscv/mm/context.c +0
-1@@ -317,7 +317,6 @@ void switch_mm(struct mm_struct *prev, struct mm_struct *next,*/ cpu =smp_processor_id (); -cpumask_clear_cpu (cpu,mm_cpumask (prev));It should be:
Review
Quoted 2 lines
(quoted diff) +1
-1- cpumask_clear_cpu(cpu, mm_cpumask(prev)); + if (static_branch_unlikely(&use_asid_allocator))
Sorry,
if (!static_branch_unlikely(&use_asid_allocator))
I would RESEND the patch.
Quoted 16 lines
(quoted diff) +2
-0+ cpumask_clear_cpu(cpu, mm_cpumask(prev)); +No bug on noasid part.
Quoted 6 lines
arch/riscv/mm/context.c +0
-1cpumask_set_cpu (cpu,mm_cpumask (next));set_mm (next, cpu); --2.36.1
--
Best Regards
Guo Ren
Signature · 2 lines
Best Regards Guo Ren