Re: [RFC PATCH 1/9] arm64/entry: Bound certain debug exception paths in instrumentation windows
Hongyan Xia <[email protected]>
| Newsgroups | org.infradead.lists.linux-arm-kernel,org.kernel.vger.linux-kernel |
|---|---|
| Message-ID | <[email protected]> |
On 7/31/2026 10:41 PM, Mark Rutland wrote: > On Mon, Jul 27, 2026 at 12:25:34PM +0000, Hongyan Xia wrote: >> From: Hongyan Xia <[email protected]> >> >> The el1 debug exception handlers are noinstr, but their payload calls >> (do_breakpoint(), do_watchpoint(), try_step_suspended_breakpoints() >> and the Cortex-A76 erratum handler) are ordinary instrumentable code. >> >> Mark the boundaries explicitly with instrumentation_begin()/end(), in >> the same style as x86 exception entries. >> >> The BRK handlers (do_el1_brk64(), do_el1_softstep()) will be dealt with >> in later patches. >> >> Signed-off-by: Hongyan Xia <[email protected]> > > Please don't do this. > > The instrumentation_{begin,end}() markers don't do anything useful on > arm64 (since we don't use objtoo, etc), and adding them to a subset of > arm64's exception entry code gives the misleading impression that these > cases are different from other exception entry cases. Okay. I think it's a bit unclear what the scope of this series is. This series started out trying to make the whole debug_exception() path noinstr, not just Kprobe. With my recent involvement in Kprobe, the Kprobe part is handled carefully while other paths are hit with a giant instructiontation_{begin,end}() hammer. I can see this is confusing. I'll limit the series to just Kprobe noinstr and make the commit messages clear. This should also drop a few patches in this series. > > Mark. > >> --- >> arch/arm64/kernel/entry-common.c | 19 +++++++++++++++++-- >> 1 file changed, 17 insertions(+), 2 deletions(-) >> >> diff --git a/arch/arm64/kernel/entry-common.c b/arch/arm64/kernel/entry-common.c >> index ceb4eb11232a..466529cce1c3 100644 >> --- a/arch/arm64/kernel/entry-common.c >> +++ b/arch/arm64/kernel/entry-common.c >> @@ -6,6 +6,7 @@ >> */ >> >> #include <linux/context_tracking.h> >> +#include <linux/instrumentation.h> >> #include <linux/irq-entry-common.h> >> #include <linux/kasan.h> >> #include <linux/linkage.h> >> @@ -380,7 +381,9 @@ static void noinstr el1_breakpt(struct pt_regs *regs, unsigned long esr) >> >> state = arm64_enter_el1_dbg(regs); >> debug_exception_enter(regs); >> + instrumentation_begin(); >> do_breakpoint(esr, regs); >> + instrumentation_end(); >> debug_exception_exit(regs); >> arm64_exit_el1_dbg(regs, state); >> } >> @@ -388,9 +391,15 @@ static void noinstr el1_breakpt(struct pt_regs *regs, unsigned long esr) >> static void noinstr el1_softstp(struct pt_regs *regs, unsigned long esr) >> { >> irqentry_state_t state; >> + bool handled; >> >> state = arm64_enter_el1_dbg(regs); >> - if (!cortex_a76_erratum_1463225_debug_handler(regs)) { >> + >> + instrumentation_begin(); >> + handled = cortex_a76_erratum_1463225_debug_handler(regs); >> + instrumentation_end(); >> + >> + if (!handled) { >> debug_exception_enter(regs); >> /* >> * After handling a breakpoint, we suspend the breakpoint >> @@ -398,7 +407,11 @@ static void noinstr el1_softstp(struct pt_regs *regs, unsigned long esr) >> * If we are stepping a suspended breakpoint there's nothing more to do: >> * the single-step is complete. >> */ >> - if (!try_step_suspended_breakpoints(regs)) >> + instrumentation_begin(); >> + handled = try_step_suspended_breakpoints(regs); >> + instrumentation_end(); >> + >> + if (!handled) >> do_el1_softstep(esr, regs); >> debug_exception_exit(regs); >> } >> @@ -413,7 +426,9 @@ static void noinstr el1_watchpt(struct pt_regs *regs, unsigned long esr) >> >> state = arm64_enter_el1_dbg(regs); >> debug_exception_enter(regs); >> + instrumentation_begin(); >> do_watchpoint(far, esr, regs); >> + instrumentation_end(); >> debug_exception_exit(regs); >> arm64_exit_el1_dbg(regs, state); >> } >> -- >> 2.47.3 >>