On 4/29/25 21:57, Will Deacon wrote:
> On Wed, Apr 16, 2025 at 11:57:06AM +0530, Anshuman Khandual wrote:
>> Add an entry for nodebugmon which has been used on arm64 platforms.
>>
>> Cc: Jonathan Corbet <corbet@xxxxxxx>
>> Cc: Andrew Morton <akpm@xxxxxxxxxxxxxxxxxxxx>
>> Cc: Catalin Marinas <catalin.marinas@xxxxxxx>
>> Cc: Will Deacon <will@xxxxxxxxxx>
>> Cc: Mark Rutland <mark.rutland@xxxxxxx>
>> Cc: linux-doc@xxxxxxxxxxxxxxx
>> Cc: linux-kernel@xxxxxxxxxxxxxxx
>> Cc: linux-arm-kernel@xxxxxxxxxxxxxxxxxxx
>> Signed-off-by: Anshuman Khandual <anshuman.khandual@xxxxxxx>
>> ---
>> This patch applies on v6.15-rc2
>>
>> Documentation/admin-guide/kernel-parameters.txt | 7 +++++++
>> 1 file changed, 7 insertions(+)
>>
>> diff --git a/Documentation/admin-guide/kernel-parameters.txt b/Documentation/admin-guide/kernel-parameters.txt
>> index d9fd26b95b34..f4a313d6c0ab 100644
>> --- a/Documentation/admin-guide/kernel-parameters.txt
>> +++ b/Documentation/admin-guide/kernel-parameters.txt
>> @@ -4085,6 +4085,13 @@
>> /sys/module/printk/parameters/console_suspend) to
>> turn on/off it dynamically.
>>
>> + nodebugmon
>> + [HW,ARM64] Disable debug monitor
>> + Disables the debug monitor exception handling on the platform
>> + regardless whether breakpoints and watchpoints are programmed
>> + or not. This prevents debug exceptions from being enabled via
>> + MDSCR_EL1 register.
>
> Hmm. I appreciate the effort to document this, but I'm not sure that the
> text above is really going to help anybody.
>
> Firstly, this option goes hand-in-hand with a debugfs control
> ("debug_enabled") and so the two would ideally be documented together.
Completely agree on this, but could find an appropriate Documentation/
file listing arm64 platform relate debugfs interfaces. Just wondering
if we need to have such a file created in Documentation/arch/arm64 ?
>
> Secondly, I think the documentation should talk about the user-visible
> effects that the control has... and that's where I get stuck! This has
> been there since day 1 but I'm not really sure what it's useful for. I
Hmm, alright. I tried to reason about what is being done in the kernel
but as you mentioned above, user visible effects were not very clearly
apparent. But having a kernel command line option without documenting
on what it really does seemed problematic as well.
> _do_ remember needing it to use the DS-5 debugger back in the day, but
> looking at what the option does (it mostly prevents the kernel from
> touching MDSCR_EL1.{MDE,KDE} which effectively disables hardware
> breakpoints and watchpoints for kernel/user and single-step for kernel)
Right, it is an system wide override regardless whether there are any
current users for these break/watch points.
> I don't see why that's relevant for halting debug. The architecture says
> that EDSCR.HDE can be used by the external debugger to enter debug state
> and I don't see anything to suggest that MDSCR_EL1 settings interfere
> with this.
ARM ARM DDI 0487 L.A states that EDSCR.HDE can be used to enable/disable
debug halting with OSLSR_EL1.OSLK = 1. Hence MDSCR_EL1.{MDE,KDE} controls
are not required to disable debug state in kernel.
>
> Maybe the problem was that the external debugger left a load of live
> breakpoint registers around after detaching, so this was a way to
> suppress those? I don't remember :/
But that would be a bug in the HW, not an expected behaviour - is not it ?
>
> I wonder if we can remove this?
Yes, we could probably remove both 'nodebugmon' kernel command line option
and 'debug_enabled' debug fs control if EDSCR.HDE is the controller of the
debug state for external debuggers.
