[PATCH v3] ppc: use rfid when running under a CPU from the 970 family.

[Cédric Le Goater]

A recent attempt to restrict the use of rfi on 64bit cpus in qemu
broke 32bit OpenBIOS when run under a 970.

This patches memory to replace rfi instructions with rfid in the
vector code.

Signed-off-by: Cédric Le Goater 
Suggested-by: Alexander Graf 
---

Tested on qemu.

Changes since v2 :

- make it nicer

arch/ppc/qemu/ofmem.c |   26 ++++++++++++++++++++++++++
 1 file changed, 26 insertions(+)

Index: openbios.git/arch/ppc/qemu/ofmem.c
===================================================================
--- openbios.git.orig/arch/ppc/qemu/ofmem.c
+++ openbios.git/arch/ppc/qemu/ofmem.c
@@ -478,6 +478,30 @@ isi_exception(void)
     hash_page(nip, phys, mode);
 }

+/* When running on ppc64, we cannot use rfi anymore. Let's patch the
+ * vectors to use rfid instead.
+ */
+#define RFI     0x4c000064
+#define RFID    0x4c000024
+
+extern char __vectors[];
+extern char __vectors_end[];
+
+static void patch_rfi(void)
+{
+    uint32_t* ptr;
+    uint32_t* vec_start = (uint32_t*) 0x100UL;
+    uint32_t* vec_end = (uint32_t*) (__vectors_end - __vectors);
+
+    if (!is_ppc64())
+        return;
+
+    for (ptr = vec_start; ptr < vec_end; ptr++)  {
+        if (*ptr == RFI)
+            *ptr = RFID;
+    }
+    flush_icache_range((char*) vec_start , (char*) vec_end);
+}

/************************************************************************/
 /*	init / cleanup							*/
@@ -532,6 +556,8 @@ setup_mmu(unsigned long ramsize)

memcpy((void *)get_rom_base(), (void *)OF_CODE_START, OF_CODE_SIZE);

+    patch_rfi();
+
     /* Enable MMU */

mtmsr(mfmsr() | MSR_IR | MSR_DR);

[Alexander Graf]

> Am 22.06.2016 um 09:29 schrieb Cédric Le Goater :
>
> A recent attempt to restrict the use of rfi on 64bit cpus in qemu
> broke 32bit OpenBIOS when run under a 970.
>
> This patches memory to replace rfi instructions with rfid in the
> vector code.
>
> Signed-off-by: Cédric Le Goater 
> Suggested-by: Alexander Graf
>
> Tested on qemu.
>
> Changes since v2 :
> make it nicer
>
> arch/ppc/qemu/ofmem.c |   26 ++++++++++++++++++++++++++
> 1 file changed, 26 insertions(+)
> Index: openbios.git/arch/ppc/qemu/ofmem.c
>
> --- openbios.git.orig/arch/ppc/qemu/ofmem.c
> +++ openbios.git/arch/ppc/qemu/ofmem.c
> @@ -478,6 +478,30 @@ isi_exception(void)
>     hash_page(nip, phys, mode);
> }
>
> +/* When running on ppc64, we cannot use rfi anymore. Let's patch the
> vectors to use rfid instead.

Please use

/*
 * Comments like this
 */

and explain exactly why "we can not use rfi anymore", so that someone looking at the code in 5 years will still know what's going on.
> */
>
> +#define RFI     0x4c000064
> +#define RFID    0x4c000024
>
> +extern char __vectors[];
> +extern char __vectors_end[];
>
> +static void patch_rfi(void)
> +{
> uint32_t* ptr;
> uint32_t* vec_start = (uint32_t*) 0x100UL;
> uint32_t* vec_end = (uint32_t*) (__vectors_end - __vectors);
> if (!is_ppc64())
>    return;
> for (ptr = vec_start; ptr < vec_end; ptr++)  {

Last time I asked < was not really defined for pointers. You can only match (==, !=) or convert to numbers (uintptr_t) them. The smaller comparison might work for now, but I can't guarantee it won't break in gcc7 or so.

In short, please just make it != :).

Alex
>    if (*ptr == RFI)
>        *ptr = RFID;
> }
> flush_icache_range((char*) vec_start , (char*) vec_end);
>
> +}
>
> /************************************************************************/
> /*    init / cleanup                            */
> @@ -532,6 +556,8 @@ setup_mmu(unsigned long ramsize)
> memcpy((void *)get_rom_base(), (void *)OF_CODE_START, OF_CODE_SIZE);
>
> patch_rfi();
>
> /* Enable MMU */
>
> mtmsr(mfmsr() | MSR_IR | MSR_DR);

[Cédric Le Goater]

On 06/22/2016 09:41 AM, Alexander Graf wrote:
> > Am 22.06.2016 um 09:29 schrieb Cédric Le Goater :
> >
> > A recent attempt to restrict the use of rfi on 64bit cpus in qemu
> > broke 32bit OpenBIOS when run under a 970.
> >
> > This patches memory to replace rfi instructions with rfid in the
> > vector code.
> >
> > Signed-off-by: Cédric Le Goater 
> > Suggested-by: Alexander Graf
> >
> > Tested on qemu.
> >
> > Changes since v2 :
> > make it nicer
> >
> > arch/ppc/qemu/ofmem.c |   26 ++++++++++++++++++++++++++
> > 1 file changed, 26 insertions(+)
> > Index: openbios.git/arch/ppc/qemu/ofmem.c
> >
> > --- openbios.git.orig/arch/ppc/qemu/ofmem.c
> > +++ openbios.git/arch/ppc/qemu/ofmem.c
> > @@ -478,6 +478,30 @@ isi_exception(void)
> >     hash_page(nip, phys, mode);
> > }
> >
> > +/* When running on ppc64, we cannot use rfi anymore. Let's patch the
> > vectors to use rfid instead.
>
> Please use
>
> /*
> Comments like this
>
> */
>
> and explain exactly why "we can not use rfi anymore", so that someone 
> looking at the code in 5 years will still know what's going on.

yep.
> > */
> >
> > +#define RFI     0x4c000064
> > +#define RFID    0x4c000024
> >
> > +extern char __vectors[];
> > +extern char __vectors_end[];
> >
> > +static void patch_rfi(void)
> > +{
> > uint32_t* ptr;
> > uint32_t* vec_start = (uint32_t*) 0x100UL;
> > uint32_t* vec_end = (uint32_t*) (__vectors_end - __vectors);
> > if (!is_ppc64())
> >    return;
> > for (ptr = vec_start; ptr < vec_end; ptr++)  {
>
> Last time I asked < was not really defined for pointers. You can only
> match (==, !=) or convert to numbers (uintptr_t) them. The smaller 
> comparison might work for now, but I can't guarantee it won't break 
> in gcc7 or so.

oh yes. that is true.
> In short, please just make it != :).

Thanks,

C.
> Alex
> >    if (*ptr == RFI)
> >        *ptr = RFID;
> > }
> > flush_icache_range((char*) vec_start , (char*) vec_end);
> >
> > +}
> >
> > /************************************************************************/
> > /*    init / cleanup                            */
> > @@ -532,6 +556,8 @@ setup_mmu(unsigned long ramsize)
> > memcpy((void *)get_rom_base(), (void *)OF_CODE_START, OF_CODE_SIZE);
> >
> > patch_rfi();
> >
> > /* Enable MMU */
> >
> > mtmsr(mfmsr() | MSR_IR | MSR_DR);