Hi, Oleg
在 2024/6/19 22:38, Oleg Nesterov 写道:
> On 06/19, Liao Chang wrote:
>>
>> When the new uretprobe system call was added [1], the xol slots reserved
>> for the uretprobe trampoline might be insufficient on some architecture.
>
> Confused... this doesn't depend on the change above?
You are right, the uretprobe syscall is specifc to x86_64. This patch wouldn't
address that issue. However, when i asm porting uretprobe trampoline to arm64
to explore its benefits on that architecture, i discovered the problem that
single slot is not large enought for trampoline code.
>
>> For example, on arm64, the trampoline is consist of three instructions
>> at least. So it should mark enough bits in area->bitmaps and
>> and area->slot_count for the reserved slots.
>
> Do you mean that on arm64 UPROBE_SWBP_INSN_SIZE > UPROBE_XOL_SLOT_BYTES ?
>
>>From arch/arm64/include/asm/uprobes.h
>
> #define MAX_UINSN_BYTES AARCH64_INSN_SIZE
>
> #define UPROBE_SWBP_INSN cpu_to_le32(BRK64_OPCODE_UPROBES)
> #define UPROBE_SWBP_INSN_SIZE AARCH64_INSN_SIZE
> #define UPROBE_XOL_SLOT_BYTES MAX_UINSN_BYTES
>
> typedef __le32 uprobe_opcode_t;
>
> struct arch_uprobe_task {
> };
>
> struct arch_uprobe {
> union {
> u8 insn[MAX_UINSN_BYTES];
> u8 ixol[MAX_UINSN_BYTES];
>
> So it seems that UPROBE_SWBP_INSN_SIZE == MAX_UINSN_BYTES and it must
> be less than UPROBE_XOL_SLOT_BYTES, otherwise
>
> arch_uprobe_copy_ixol(..., uprobe->arch.ixol, sizeof(uprobe->arch.ixol))
> in xol_get_insn_slot() won't fit the slot as well?
This real reason is that the current implmentation seems to assume the
ixol slot has enough space for the uretprobe trampoline. This assumption
is true for x86_64, since the ixol slot is size of 128 bytes.
