[v2] arm64/debug: Fix registers on sleeping tasks

Commit Message

Doug Anderson March 8, 2018, 4:41 p.m. UTC

Hi,

On Thu, Mar 8, 2018 at 8:19 AM, Daniel Thompson
<[e-mail]> wrote:
> On 05/03/18 23:43, Douglas Anderson wrote:
>>
>> This is the equivalent of commit 001bf455d206 ("ARM: 8428/1: kgdb: Fix
>> registers on sleeping tasks") but for arm64.  Nuff said.
>>
>> ...well, perhaps I could also add that task_pt_regs are userspace
>> registers and that's not what kgdb is supposed to be reporting.  We're
>> supposed to be reporting kernel registers.
>>
>> Signed-off-by: Douglas Anderson <[e-mail]>
>
>
> I hacked together a (still very immature) kgdb test suite[1] around the turn
> of the year. Whilst its not quite solid enough for me to recommend others
> deploy it except out of curiosity... so I haven't yet started yelling about
> test suite failures except in the privacy of my own head.
>
> However I can confirm that this patch fixes one of the test suite failures I
> haven't had time to blame allocate yet!
>
> So...
> Tested-by: Daniel Thompson <[e-mail]>

Thanks for your testing!  ...I'll have to check out your test suite soon.


> BTW is this something that should Cc: stable?

It wouldn't hurt if this made it back to stable on a best-effort
approach.  The problem has been there since the beginning, so it's not
like it's fixing a regression that cropped up in a specific version.
...but it does fix a bug, so probably Cc stable makes sense.  I guess
I'd leave it up to the maintainer that applies the patch?


-Doug<div class="gmail_extra"><br><div class="gmail_quote">On Thu, Mar
8, 2018 at 8:19 AM, Daniel Thompson <span dir="ltr">&lt;<a
href="mailto:[e-mail]"
target="_blank">[e-mail]</a>&gt;</span>
wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex"><span class="">On
05/03/18 23:43, Douglas Anderson wrote:<br>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
This is the equivalent of commit 001bf455d206 ("ARM: 8428/1: kgdb: Fix<br>
registers on sleeping tasks") but for arm64.&nbsp; Nuff said.<br>
<br>
...well, perhaps I could also add that task_pt_regs are userspace<br>
registers and that's not what kgdb is supposed to be reporting.&nbsp; We're<br>
supposed to be reporting kernel registers.<br>
<br>
Signed-off-by: Douglas Anderson &lt;<a
href="mailto:[e-mail]"
target="_blank">[e-mail]</a>&gt;<br>
</blockquote>
<br></span>
I hacked together a (still very immature) kgdb test suite[1] around
the turn of the year. Whilst its not quite solid enough for me to
recommend others deploy it except out of curiosity... so I haven't yet
started yelling about test suite failures except in the privacy of my
own head.<br>
<br>
However I can confirm that this patch fixes one of the test suite
failures I haven't had time to blame allocate yet!<br>
<br>
So...<br>
Tested-by: Daniel Thompson &lt;<a
href="mailto:[e-mail]"
target="_blank">[e-mail]</a>&gt;<br>
<br>
BTW is this something that should Cc: stable?<br>
<br>
<br>
Daniel.<br>
<br>
[1] <a href="https://github.com/daniel-thompson/kgdbtest"
rel="noreferrer" target="_blank"
data-saferedirecturl="https://www.google.com/url?hl=en&amp;q=https://github.com/daniel-thompson/kgdbtest&amp;source=gmail&amp;ust=1520613443900000&amp;usg=AFQjCNHzh_tGdP65DdFEQo-ofF_LCI5XBQ">https://github.com/daniel-thom<wbr>pson/kgdbtest</a><div
class="HOEnZb"><div class="h5"><br>
<br>
<br>
<blockquote class="gmail_quote" style="margin:0 0 0
.8ex;border-left:1px #ccc solid;padding-left:1ex">
---<br>
<br>
Changes in v2:<br>
- Init cpu_context in one line<br>
- Explain that task_pt_regs are userspace<br>
<br>
&nbsp; arch/arm64/kernel/kgdb.c | 21 ++++++++++++++++-----<br>
&nbsp; 1 file changed, 16 insertions(+), 5 deletions(-)<br>
<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[31] = cpu_context-&gt;sp;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[32] = cpu_context-&gt;pc;<br>
&nbsp; }<br>
&nbsp; &nbsp; void kgdb_arch_set_pc(struct pt_regs *regs, unsigned long pc)<br>
<br>
</blockquote>
<br>
</div></div></blockquote></div><br></div>

Comments

Will Deacon March 8, 2018, 4:43 p.m. UTC | #1

On Thu, Mar 08, 2018 at 08:41:59AM -0800, Doug Anderson wrote:
> Hi,
> 
> On Thu, Mar 8, 2018 at 8:19 AM, Daniel Thompson
> <[e-mail]> wrote:
> > On 05/03/18 23:43, Douglas Anderson wrote:
> >>
> >> This is the equivalent of commit 001bf455d206 ("ARM: 8428/1: kgdb: Fix
> >> registers on sleeping tasks") but for arm64.  Nuff said.
> >>
> >> ...well, perhaps I could also add that task_pt_regs are userspace
> >> registers and that's not what kgdb is supposed to be reporting.  We're
> >> supposed to be reporting kernel registers.
> >>
> >> Signed-off-by: Douglas Anderson <[e-mail]>
> >
> >
> > I hacked together a (still very immature) kgdb test suite[1] around the turn
> > of the year. Whilst its not quite solid enough for me to recommend others
> > deploy it except out of curiosity... so I haven't yet started yelling about
> > test suite failures except in the privacy of my own head.
> >
> > However I can confirm that this patch fixes one of the test suite failures I
> > haven't had time to blame allocate yet!
> >
> > So...
> > Tested-by: Daniel Thompson <[e-mail]>
> 
> Thanks for your testing!  ...I'll have to check out your test suite soon.
> 
> 
> > BTW is this something that should Cc: stable?
> 
> It wouldn't hurt if this made it back to stable on a best-effort
> approach.  The problem has been there since the beginning, so it's not
> like it's fixing a regression that cropped up in a specific version.
> ...but it does fix a bug, so probably Cc stable makes sense.  I guess
> I'd leave it up to the maintainer that applies the patch?

I've already put this into -next, so I don't really want to rebase just for
this. If you think it's important, please send to [e-mail]
once it's landed in mainline.

Will

Patch

diff --git a/arch/arm64/kernel/kgdb.c b/arch/arm64/kernel/kgdb.c<br>
index 2122cd187f19..a20de58061a8 100644<br>
--- a/arch/arm64/kernel/kgdb.c<br>
+++ b/arch/arm64/kernel/kgdb.c<br>
@@ -138,14 +138,25 @@  int dbg_set_reg(int regno, void *mem, struct
pt_regs *regs)<br>
&nbsp; void<br>
&nbsp; sleeping_thread_to_gdb_regs(un<wbr>signed long *gdb_regs,
struct task_struct *task)<br>
&nbsp; {<br>
-&nbsp; &nbsp; &nbsp; &nbsp;struct pt_regs *thread_regs;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;struct cpu_context *cpu_context =
&amp;task-&gt;thread.cpu_context;<br>
&nbsp; &nbsp; &nbsp; &nbsp; /* Initialize to zero */<br>
&nbsp; &nbsp; &nbsp; &nbsp; memset((char *)gdb_regs, 0, NUMREGBYTES);<br>
-&nbsp; &nbsp; &nbsp; &nbsp;thread_regs = task_pt_regs(task);<br>
-&nbsp; &nbsp; &nbsp; &nbsp;memcpy((void *)gdb_regs, (void
*)thread_regs-&gt;regs, GP_REG_BYTES);<br>
-&nbsp; &nbsp; &nbsp; &nbsp;/* Special case for PSTATE (check comments
in asm/kgdb.h for details) */<br>
-&nbsp; &nbsp; &nbsp; &nbsp;dbg_get_reg(33, gdb_regs + GP_REG_BYTES,
thread_regs);<br>
+<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[19] = cpu_context-&gt;x19;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[20] = cpu_context-&gt;x20;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[21] = cpu_context-&gt;x21;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[22] = cpu_context-&gt;x22;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[23] = cpu_context-&gt;x23;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[24] = cpu_context-&gt;x24;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[25] = cpu_context-&gt;x25;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[26] = cpu_context-&gt;x26;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[27] = cpu_context-&gt;x27;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[28] = cpu_context-&gt;x28;<br>
+&nbsp; &nbsp; &nbsp; &nbsp;gdb_regs[29] = cpu_context-&gt;fp;<br>
+<br>