The prior commit introduced VM_DROPPABLE, but in a limited form where the faulting instruction was retried instead of skipped. Finish that up with the platform-specific aspect of skipping the actual instruction. This works by copying userspace's %rip to a stack buffer of size MAX_INSN_SIZE, decoding it, and then adding the length of the decoded instruction to userspace's %rip. In the event any of these fail, just fallback to not advancing %rip and trying again. Cc: linux-mm@xxxxxxxxx Signed-off-by: Jason A. Donenfeld <Jason@xxxxxxxxx> --- arch/x86/mm/fault.c | 19 +++++++++++++++++++ include/linux/mm_types.h | 5 ++++- mm/memory.c | 4 +++- 3 files changed, 26 insertions(+), 2 deletions(-) diff --git a/arch/x86/mm/fault.c b/arch/x86/mm/fault.c index 7b0d4ab894c8..76ca99ab6eb7 100644 --- a/arch/x86/mm/fault.c +++ b/arch/x86/mm/fault.c @@ -33,6 +33,8 @@ #include <asm/kvm_para.h> /* kvm_handle_async_pf */ #include <asm/vdso.h> /* fixup_vdso_exception() */ #include <asm/irq_stack.h> +#include <asm/insn.h> /* struct insn */ +#include <asm/insn-eval.h> /* insn_fetch_from_user(), ... */ #define CREATE_TRACE_POINTS #include <asm/trace/exceptions.h> @@ -1454,6 +1456,23 @@ void do_user_addr_fault(struct pt_regs *regs, } mmap_read_unlock(mm); + + if (fault & VM_FAULT_SKIP_INSN) { + u8 buf[MAX_INSN_SIZE]; + struct insn insn; + int nr_copied; + + nr_copied = insn_fetch_from_user(regs, buf); + if (nr_copied <= 0) + return; + + if (!insn_decode_from_regs(&insn, regs, buf, nr_copied)) + return; + + regs->ip += insn.length; + return; + } + if (likely(!(fault & VM_FAULT_ERROR))) return; diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h index 3b8475007734..e76ab9ad555c 100644 --- a/include/linux/mm_types.h +++ b/include/linux/mm_types.h @@ -945,6 +945,7 @@ typedef __bitwise unsigned int vm_fault_t; * fsync() to complete (for synchronous page faults * in DAX) * @VM_FAULT_COMPLETED: ->fault completed, meanwhile mmap lock released + * @VM_FAULT_SKIP_INSN: ->handle the fault by skipping faulting instruction * @VM_FAULT_HINDEX_MASK: mask HINDEX value * */ @@ -962,6 +963,7 @@ enum vm_fault_reason { VM_FAULT_DONE_COW = (__force vm_fault_t)0x001000, VM_FAULT_NEEDDSYNC = (__force vm_fault_t)0x002000, VM_FAULT_COMPLETED = (__force vm_fault_t)0x004000, + VM_FAULT_SKIP_INSN = (__force vm_fault_t)0x008000, VM_FAULT_HINDEX_MASK = (__force vm_fault_t)0x0f0000, }; @@ -985,7 +987,8 @@ enum vm_fault_reason { { VM_FAULT_RETRY, "RETRY" }, \ { VM_FAULT_FALLBACK, "FALLBACK" }, \ { VM_FAULT_DONE_COW, "DONE_COW" }, \ - { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" } + { VM_FAULT_NEEDDSYNC, "NEEDDSYNC" }, \ + { VM_FAULT_SKIP_INSN, "SKIP_INSN" } struct vm_special_mapping { const char *name; /* The name, e.g. "[vdso]". */ diff --git a/mm/memory.c b/mm/memory.c index 1ade407ccbf9..62ba9b7b713e 100644 --- a/mm/memory.c +++ b/mm/memory.c @@ -5221,8 +5221,10 @@ vm_fault_t handle_mm_fault(struct vm_area_struct *vma, unsigned long address, lru_gen_exit_fault(); /* If the mapping is droppable, then errors due to OOM aren't fatal. */ - if (vma->vm_flags & VM_DROPPABLE) + if ((ret & VM_FAULT_OOM) && (vma->vm_flags & VM_DROPPABLE)) { ret &= ~VM_FAULT_OOM; + ret |= VM_FAULT_SKIP_INSN; + } if (flags & FAULT_FLAG_USER) { mem_cgroup_exit_user_fault(); -- 2.39.0