There are two major types of uncorrected recoverable (UCR) errors :
- Synchronous error: The error is detected and raised at the point of the consumption in the execution flow, e.g. when a CPU tries to access a poisoned cache line. The CPU will take a synchronous error exception such as Synchronous External Abort (SEA) on Arm64 and Machine Check Exception (MCE) on X86. OS requires to take action (for example, offline failure page/kill failure thread) to recover this uncorrectable error.
- Asynchronous error: The error is detected out of processor execution context, e.g. when an error is detected by a background scrubber. Some data in the memory are corrupted. But the data have not been consumed. OS is optional to take action to recover this uncorrectable error.
Currently, both synchronous and asynchronous error use memory_failure_queue() to schedule memory_failure() exectute in kworker context. As a result, when a user-space process is accessing a poisoned data, a data abort is taken and the memory_failure() is executed in the kworker context:
- will send wrong si_code by SIGBUS signal in early_kill mode, and - can not kill the user-space in some cases resulting a synchronous error infinite loop
Issue 1: send wrong si_code in early_kill mode
Since commit a70297d22132 ("ACPI: APEI: set memory failure flags as MF_ACTION_REQUIRED on synchronous events")', the flag MF_ACTION_REQUIRED could be used to determine whether a synchronous exception occurs on ARM64 platform. When a synchronous exception is detected, the kernel is expected to terminate the current process which has accessed poisoned page. This is done by sending a SIGBUS signal with an error code BUS_MCEERR_AR, indicating an action-required machine check error on read.
However, when kill_proc() is called to terminate the processes who have the poisoned page mapped, it sends the incorrect SIGBUS error code BUS_MCEERR_AO because the context in which it operates is not the one where the error was triggered.
To reproduce this problem:
# STEP1: enable early kill mode #sysctl -w vm.memory_failure_early_kill=1 vm.memory_failure_early_kill = 1
# STEP2: inject an UCE error and consume it to trigger a synchronous error #einj_mem_uc single 0: single vaddr = 0xffffb0d75400 paddr = 4092d55b400 injecting ... triggering ... signal 7 code 5 addr 0xffffb0d75000 page not present Test passed
The si_code (code 5) from einj_mem_uc indicates that it is BUS_MCEERR_AO error and it is not fact.
To fix it, queue memory_failure() as a task_work so that it runs in the context of the process that is actually consuming the poisoned data.
After this patch set:
# STEP1: enable early kill mode #sysctl -w vm.memory_failure_early_kill=1 vm.memory_failure_early_kill = 1
# STEP2: inject an UCE error and consume it to trigger a synchronous error #einj_mem_uc single 0: single vaddr = 0xffffb0d75400 paddr = 4092d55b400 injecting ... triggering ... signal 7 code 4 addr 0xffffb0d75000 page not present Test passed
The si_code (code 4) from einj_mem_uc indicates that it is BUS_MCEERR_AR error as we expected.
Issue 2: a synchronous error infinite loop due to memory_failure() failed
If a user-space process, e.g. devmem, a poisoned page which has been set HWPosion flag, kill_accessing_process() is called to send SIGBUS to the current processs with error info. Because the memory_failure() is executed in the kworker contex, it will just do nothing but return EFAULT. So, devmem will access the posioned page and trigger an excepction again, resulting in a synchronous error infinite loop. Such loop may cause platform firmware to exceed some threshold and reboot when Linux could have recovered from this error.
To reproduce this problem:
# STEP 1: inject an UCE error, and kernel will set HWPosion flag for related page #einj_mem_uc single 0: single vaddr = 0xffffb0d75400 paddr = 4092d55b400 injecting ... triggering ... signal 7 code 4 addr 0xffffb0d75000 page not present Test passed
# STEP 2: access the same page and it will trigger a synchronous error infinite loop devmem 0x4092d55b400
To fix it, if memory_failure() failed, perform a force kill to current process.
Issue 3: a synchronous error infinite loop due to no memory_failure() queued
No memory_failure() work is queued unless all bellow preconditions check passed:
- `if (!(mem_err->validation_bits & CPER_MEM_VALID_PA))` in ghes_handle_memory_failure() - `if (flags == -1)` in ghes_handle_memory_failure() - `if (!IS_ENABLED(CONFIG_ACPI_APEI_MEMORY_FAILURE))` in ghes_do_memory_failure() - `if (!pfn_valid(pfn) && !arch_is_platform_page(physical_addr)) ` in ghes_do_memory_failure()
If the preconditions are not passed, the user-space process will trigger SEA again. This loop can potentially exceed the platform firmware threshold or even trigger a kernel hard lockup, leading to a system reboot.
To fix it, if no memory_failure() queued, perform a force kill to current process.
And the the memory errors triggered in kernel-mode[5], also relies on this patchset to kill the failure thread.
Lv Ying and XiuQi from Huawei also proposed to address similar problem[2][4]. Acknowledge to discussion with them.
[1] Add ARMv8 RAS virtualization support in QEMU https://patchew.org/QEMU/20200512030609.19593-1-gengdongjiu@huawei.com/ [2] https://lore.kernel.org/lkml/20221205115111.131568-3-lvying6@huawei.com/ [3] https://lkml.kernel.org/r/20220914064935.7851-1-xueshuai@linux.alibaba.com [4] https://lore.kernel.org/lkml/20221209095407.383211-1-lvying6@huawei.com/ [5] https://patchwork.kernel.org/project/linux-arm-kernel/cover/20240528085915.1...
Shuai Xue (4): ACPI: APEI: set memory failure flags as MF_ACTION_REQUIRED on synchronous events ACPI: APEI: send SIGBUS to current task if synchronous memory error not recovered mm: memory-failure: move return value documentation to function declaration ACPI: APEI: handle synchronous exceptions in task work
arch/x86/kernel/cpu/mce/core.c | 5 -- drivers/acpi/apei/ghes.c | 114 ++++++++++++++++++++++----------- include/acpi/ghes.h | 3 - include/linux/mm.h | 1 - mm/memory-failure.c | 19 ++---- 5 files changed, 82 insertions(+), 60 deletions(-)