Re: [Minios-devel] [PATCH 25/40] arm64: set the stack for the arm_start_thread

[Huang Shijie <shijie.huang@xxxxxxx>]

On Wed, Nov 08, 2017 at 09:48:49AM +0000, Julien Grall wrote:
> Hi,
> 
> On 08/11/17 06:26, Huang Shijie wrote:
> > On Tue, Nov 07, 2017 at 10:22:02AM +0000, Julien Grall wrote:
> > > Hi Shijie,
> > > 
> > > On 07/11/17 08:25, Huang Shijie wrote:
> > > > On Mon, Nov 06, 2017 at 02:39:30PM +0000, Julien Grall wrote:
> > > > > Hi Shijie,
> > > > > 
> > > > > On 03/11/17 03:12, Huang Shijie wrote:
> > > > > > When a thread (which is picked by the scheduler) runs at the first 
> > > > > > time,
> > > > > > it needs a stack to store the data.
> > > > > > 
> > > > > > This patch set the stack for the arm_start_thread.
> > > > > > 
> > > > > > Change-Id: I4cfa18b199cb920e9ebe19bbbe2cf508264b0ba8
> > > > > > Jira: ENTOS-247
> > > > > > Signed-off-by: Huang Shijie <shijie.huang@xxxxxxx>
> > > > > > ---
> > > > > >     arch/arm/sched.c | 6 ++++++
> > > > > >     1 file changed, 6 insertions(+)
> > > > > > 
> > > > > > diff --git a/arch/arm/sched.c b/arch/arm/sched.c
> > > > > > index 448fd3e..f691d7d 100644
> > > > > > --- a/arch/arm/sched.c
> > > > > > +++ b/arch/arm/sched.c
> > > > > > @@ -32,6 +32,12 @@ struct thread* arch_create_thread(char *name, 
> > > > > > void (*function)(void *),
> > > > > >          * for the initial switch. */
> > > > > >         thread->sp = (unsigned long) sp - sizeof(unsigned long) * 
> > > > > > CALLEE_SAVED_REGISTERS;
> > > > > > +    /*
> > > > > > +     * Please refer to the stack layout in the comment for 
> > > > > > __arch_switch_threads.
> > > > > > +     * We need to set the sp register for arm_start_thread when a 
> > > > > > thread runs
> > > > > > +     * at the first time.
> > > > > > +     */
> > > > > > +    *(--sp) = thread->sp;
> > > > > 
> > > > > But this is common code between Arm64 and Arm32. How come you don't 
> > > > > update
> > > > > arm32?
> > > > > 
> > > > > Why do you need to save sp on Arm64 and not Arm32?
> > > > The arm32 also need to save sp.. this is bug for arm32..
> > > 
> > > How is it a bug given you said you were unable to test Arm32 guest? At the
> > > moment, it looks like a bit random...
> > You can simulate it in mind, and you will know it...
> 
> I am not in your mind neither have time to spend trying to figure out how
> you think that. So how about you explain your reasoning?
If you do not set it, you will get the wrong SP stack. and the program
will go wildly. 

Thanks
Huang Shijie

_______________________________________________
Minios-devel mailing list
Minios-devel@xxxxxxxxxxxxxxxxxxxx
https://lists.xenproject.org/cgi-bin/mailman/listinfo/minios-devel