Test-by: Ma Wupeng <mawupeng1@xxxxxxxxxx> I have test this on arm64 with following steps: 1. make memory failure return EBUSY 2. force a UCE with einj Without this patchset, user task will not be kill since memory_failure can not handle this UCE properly and user task is in D state. The stack can be found in the end. With this patchset, user task can be killed even memory_failure return -EBUSY without doing anything. Here is the stack of user task with D state: # cat /proc/7001/stack [<0>] __flush_work.isra.0+0x80/0xa8 [<0>] __cancel_work_timer+0x144/0x1c8 [<0>] cancel_work_sync+0x1c/0x30 [<0>] memory_failure_queue_kick+0x3c/0x88 [<0>] ghes_kick_task_work+0x28/0x78 [<0>] task_work_run+0xb8/0x188 [<0>] do_notify_resume+0x1e0/0x280 [<0>] el0_da+0x130/0x138 [<0>] el0t_64_sync_handler+0x68/0xc0 [<0>] el0t_64_sync+0x188/0x190 On 2023/3/17 15:24, Shuai Xue wrote: > changes since v2 by addressing comments from Naoya: > - rename mce_task_work to sync_task_work > - drop ACPI_HEST_NOTIFY_MCE case in is_hest_sync_notify() > - add steps to reproduce this problem in cover letter > - Link: https://lore.kernel.org/lkml/1aa0ca90-d44c-aa99-1e2d-bd2ae610b088@xxxxxxxxxxxxxxxxx/T/#mb3dede6b7a6d189dc8de3cf9310071e38a192f8e > > changes since v1: > - synchronous events by notify type > - Link: https://lore.kernel.org/lkml/20221206153354.92394-3-xueshuai@xxxxxxxxxxxxxxxxx/ > > Currently, both synchronous and asynchronous error are queued and handled > by a dedicated kthread in workqueue. And Memory failure for synchronous > error is synced by a cancel_work_sync trick which ensures that the > corrupted page is unmapped and poisoned. And after returning to user-space, > the task starts at current instruction which triggering a page fault in > which kernel will send SIGBUS to current process due to VM_FAULT_HWPOISON. > > However, the memory failure recovery for hwpoison-aware mechanisms does not > work as expected. For example, hwpoison-aware user-space processes like > QEMU register their customized SIGBUS handler and enable early kill mode by > seting PF_MCE_EARLY at initialization. Then the kernel will directy notify > the process by sending a SIGBUS signal in memory failure with wrong > si_code: BUS_MCEERR_AO si_code to the actual user-space process instead of > BUS_MCEERR_AR. > > To address this problem: > > - PATCH 1 sets mf_flags as MF_ACTION_REQUIRED on synchronous events which > indicates error happened in current execution context > - PATCH 2 separates synchronous error handling into task work so that the > current context in memory failure is exactly belongs to the task > consuming poison data. > > Then, kernel will send SIGBUS with proper si_code in kill_proc(). > > Lv Ying and XiuQi also proposed to address similar problem and we discussed > about new solution to add a new flag(acpi_hest_generic_data::flags bit 8) to > distinguish synchronous event. [2][3] The UEFI community still has no response. > After a deep dive into the SDEI TRM, the SDEI notification should be used for > asynchronous error. As SDEI TRM[1] describes "the dispatcher can simulate an > exception-like entry into the client, **with the client providing an additional > asynchronous entry point similar to an interrupt entry point**". The client > (kernel) lacks complete synchronous context, e.g. systeam register (ELR, ESR, > etc). So notify type is enough to distinguish synchronous event. > > 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. > > 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. > > [1] https://developer.arm.com/documentation/den0054/latest/ > [2] https://lore.kernel.org/linux-arm-kernel/20221205160043.57465-4-xiexiuqi@xxxxxxxxxx/T/ > [3] https://lore.kernel.org/lkml/20221209095407.383211-1-lvying6@xxxxxxxxxx/ > > Shuai Xue (2): > ACPI: APEI: set memory failure flags as MF_ACTION_REQUIRED on > synchronous events > ACPI: APEI: handle synchronous exceptions in task work > > drivers/acpi/apei/ghes.c | 135 ++++++++++++++++++++++++--------------- > include/acpi/ghes.h | 3 - > mm/memory-failure.c | 13 ---- > 3 files changed, 83 insertions(+), 68 deletions(-) >
