Re: [PATCH RFCv2 06/18] uprobes: Add orig argument to uprobe_write and uprobe_write_opcode

[Jiri Olsa <olsajiri@gmail.com>]

On Fri, Feb 28, 2025 at 11:07:38AM -0800, Andrii Nakryiko wrote:
> On Mon, Feb 24, 2025 at 6:03 AM Jiri Olsa <jolsa@kernel.org> wrote:
> >
> > The uprobe_write has special path to restore the original page when
> > we write original instruction back.
> >
> > This happens when uprobe_write detects that we want to write anything
> > else but breakpoint instruction.
> >
> > In following changes we want to use uprobe_write function for multiple
> > updates, so adding new function argument to denote that this is the
> > original instruction update. This way uprobe_write can make appropriate
> > checks and restore the original page when possible.
> >
> > Signed-off-by: Jiri Olsa <jolsa@kernel.org>
> > ---
> >  arch/arm/probes/uprobes/core.c |  2 +-
> >  include/linux/uprobes.h        |  5 +++--
> >  kernel/events/uprobes.c        | 22 ++++++++++------------
> >  3 files changed, 14 insertions(+), 15 deletions(-)
> >
> 
> [...]
> 
> > diff --git a/kernel/events/uprobes.c b/kernel/events/uprobes.c
> > index ad5879fc2d26..2b542043089e 100644
> > --- a/kernel/events/uprobes.c
> > +++ b/kernel/events/uprobes.c
> > @@ -471,25 +471,23 @@ static int update_ref_ctr(struct uprobe *uprobe, struct mm_struct *mm,
> >   * Return 0 (success) or a negative errno.
> >   */
> >  int uprobe_write_opcode(struct arch_uprobe *auprobe, struct mm_struct *mm,
> > -                       unsigned long vaddr, uprobe_opcode_t opcode)
> > +                       unsigned long vaddr, uprobe_opcode_t opcode, bool orig)
> >  {
> > -       return uprobe_write(auprobe, mm, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE, verify_opcode);
> > +       return uprobe_write(auprobe, mm, vaddr, &opcode, UPROBE_SWBP_INSN_SIZE, verify_opcode, orig);
> >  }
> >
> >  int uprobe_write(struct arch_uprobe *auprobe, struct mm_struct *mm,
> >                  unsigned long vaddr, uprobe_opcode_t *insn,
> > -                int nbytes, uprobe_write_verify_t verify)
> > +                int nbytes, uprobe_write_verify_t verify, bool orig)
> 
> why not call orig -> is_register and avoid a bunch of code churn?...
> (and while "is_register" is not awesome name, still a bit more clear
> compared to "orig", IMO)

I see the logic in the function the other way around: if you want to try
and load the original page as part of your update, then you pass true to
orig argument

also the is_register makes sense to me in the old code where you have
just 2 states: breakpoint or original instruction ... now when we added
call instruction the is_register would need to cover that as well

jirka

> 
> >  {
> >         struct page *old_page, *new_page;
> >         struct vm_area_struct *vma;
> > -       int ret, is_register;
> > +       int ret;
> >         bool orig_page_huge = false;
> >         unsigned int gup_flags = FOLL_FORCE;
> >
> > -       is_register = is_swbp_insn(insn);
> > -
> 
> [...]