Date: Tue, 05 Mar 2013 20:33:18 +1300 From: Andrew Turner <andrew@fubar.geek.nz> To: Ganbold Tsagaankhuu <ganbold@gmail.com> Cc: freebsd-arm@freebsd.org Subject: Re: ARM EABI test image Message-ID: <20130305203318.24f244c9@bender> In-Reply-To: <20130304221205.7d427b38@bender> References: <20130302172556.5b59e122@bender> <CAGtf9xMcSLpin9oR9eDZdzVnfsT0N6jQowzNq23TAsO9kTqpYA@mail.gmail.com> <20130304221205.7d427b38@bender>
next in thread | previous in thread | raw e-mail | index | archive | help
--MP_/KpoxgBcMnzg5T6o_qvFqvMh
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit
Content-Disposition: inline
On Mon, 04 Mar 2013 22:12:05 +1300
Andrew Turner <andrew@fubar.geek.nz> wrote:
> On Sun, 3 Mar 2013 01:17:01 +0800
> Ganbold Tsagaankhuu <ganbold@gmail.com> wrote:
>
> > Andrew,
> >
> > On Sat, Mar 2, 2013 at 12:25 PM, Andrew Turner
> > <andrew@fubar.geek.nz> wrote:
> > > Hello,
> > >
> > > I have built an updated ARM EABI test image for Raspberry Pi [1].
> > >
> > > The only known issue is c++ exception handling is broken when
> > > using in a dynamically linked executable. Static executables
> > > should work with c++ exceptions.
> > >
> > > To test it you will have to extract it using unxz and dd it to an
> > > sd card, for example, with a USB to SD adapter on /dev/da0:
> > > $ unxz bsd-pi-eabi-r247609.img.xz
> > > $ dd if=bsd-pi-eabi-r247609.img of=/dev/da0
> > >
> > > If you don't have a Raspberry Pi but would like to try it on your
> > > board you can add -DWITH_ARM_EABI to the make commands you use to
> > > build and install world and the kernel.
> > >
> > > Can people try this as I would like to know if anything else is
> > > broken as this will become the default ABI for 10.
> > >
> >
> > Just tried the image. Seems work but observed for instance gpart
> > shows big numbers for 2GB SD:
>
> I've confirmed this is a bug where the stack is incorrectly aligned. I
> have a fix for this and will post a patch for review when I've cleaned
> it up.
Can you try this patch. It fixes the alignment of the stack in the
kernel to be on an 8 byte boundary.
Andrew
--MP_/KpoxgBcMnzg5T6o_qvFqvMh
Content-Type: text/x-patch
Content-Transfer-Encoding: 7bit
Content-Disposition: attachment; filename=stack_align.diff
Index: arm/vm_machdep.c
===================================================================
--- arm/vm_machdep.c (revision 247608)
+++ arm/vm_machdep.c (working copy)
@@ -73,6 +73,12 @@
#include <machine/md_var.h>
+/*
+ * struct switchframe must be a multiple of 8 for correct stack alignment
+ */
+CTASSERT(sizeof(struct switchframe) == 24);
+CTASSERT(sizeof(struct trapframe) == 76);
+
#ifndef NSFBUFS
#define NSFBUFS (512 + maxusers * 16)
#endif
@@ -131,8 +137,8 @@
pcb2->un_32.pcb32_sp = td2->td_kstack +
USPACE_SVC_STACK_TOP - sizeof(*pcb2);
pmap_activate(td2);
- td2->td_frame = tf =
- (struct trapframe *)pcb2->un_32.pcb32_sp - 1;
+ td2->td_frame = tf = (struct trapframe *)STACKALIGN(
+ pcb2->un_32.pcb32_sp - sizeof(struct trapframe));
*tf = *td1->td_frame;
sf = (struct switchframe *)tf - 1;
sf->sf_r4 = (u_int)fork_return;
@@ -142,6 +148,8 @@
tf->tf_r0 = 0;
tf->tf_r1 = 0;
pcb2->un_32.pcb32_sp = (u_int)sf;
+ KASSERT((pcb2->un_32.pcb32_sp & 7) == 0,
+ ("cpu_fork: Incorrect stack alignment"));
/* Setup to release spin count in fork_exit(). */
td2->td_md.md_spinlock_count = 1;
@@ -345,6 +353,8 @@
tf->tf_r0 = 0;
td->td_pcb->un_32.pcb32_sp = (u_int)sf;
td->td_pcb->un_32.pcb32_und_sp = td->td_kstack + USPACE_UNDEF_STACK_TOP;
+ KASSERT((td->td_pcb->un_32.pcb32_sp & 7) == 0,
+ ("cpu_set_upcall: Incorrect stack alignment"));
/* Setup to release spin count in fork_exit(). */
td->td_md.md_spinlock_count = 1;
@@ -438,6 +448,8 @@
sf->sf_r4 = (u_int)func;
sf->sf_r5 = (u_int)arg;
td->td_pcb->un_32.pcb32_sp = (u_int)sf;
+ KASSERT((td->td_pcb->un_32.pcb32_sp & 7) == 0,
+ ("cpu_set_fork_handler: Incorrect stack alignment"));
}
/*
Index: arm/swtch.S
===================================================================
--- arm/swtch.S (revision 247608)
+++ arm/swtch.S (working copy)
@@ -211,10 +211,12 @@
GET_PCPU(r6)
str r7, [r6, #PC_CURPCB]
+ add sp, sp, #4;
ldmfd sp!, {r4-r7, pc}
ENTRY(cpu_switch)
stmfd sp!, {r4-r7, lr}
+ sub sp, sp, #4;
mov r6, r2 /* Save the mutex */
.Lswitch_resume:
@@ -488,6 +490,7 @@
* Pull the registers that got pushed when either savectx() or
* cpu_switch() was called and return.
*/
+ add sp, sp, #4;
ldmfd sp!, {r4-r7, pc}
#ifdef DIAGNOSTIC
.Lswitch_bogons:
@@ -501,6 +504,7 @@
#endif
ENTRY(savectx)
stmfd sp!, {r4-r7, lr}
+ sub sp, sp, #4
/*
* r0 = pcb
*/
@@ -528,6 +532,7 @@
bl _C_LABEL(vfp_store)
1:
#endif /* ARM_VFP_SUPPORT */
+ add sp, sp, #4;
ldmfd sp!, {r4-r7, pc}
ENTRY(fork_trampoline)
Index: include/frame.h
===================================================================
--- include/frame.h (revision 247608)
+++ include/frame.h (working copy)
@@ -138,10 +138,14 @@
} irqframe_t;
/*
- * Switch frame
+ * Switch frame.
+ *
+ * It is important this is a multiple of 8 bytes so the stack is correctly
+ * aligned when we create new threads.
*/
struct switchframe {
+ u_int pad; /* Used to pad the struct to a multiple of 8-bytes */
u_int sf_r4;
u_int sf_r5;
u_int sf_r6;
--MP_/KpoxgBcMnzg5T6o_qvFqvMh--
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20130305203318.24f244c9>