| Message ID | 1403973928-9717-2-git-send-email-rth@twiddle.net |
|---|---|
| State | New |
| Headers | show |
> Am 28.06.2014 um 18:45 schrieb Richard Henderson <rth@twiddle.net>: > > The default, 970fx, doesn't support MSR_LE. So even though we set LE in > ppc_cpu_reset, it gets cleared again in hreg_store_msr. Error out if a > user-selected cpu model doesn't support LE. > > Cc: Aldy Hernandez <aldyh@redhat.com> > Signed-off-by: Richard Henderson <rth@twiddle.net> > --- > The warning one gets from the unimplemented insns from tcg > is perhaps unfortunate, but it's better than silently dropping > the MSR_LE bit and interpreting the first few insns with the > wrong endianness and generating SIGILL. > > One could, perhaps, simply return false from need_byteswap. > But then the value of MSR would still be wrong, and I can imagine > that would affect gdb's interpretation of the current mode. > > > r~ > --- > linux-user/main.c | 10 +++++++--- > target-ppc/translate_init.c | 4 ++++ > 2 files changed, 11 insertions(+), 3 deletions(-) > > diff --git a/linux-user/main.c b/linux-user/main.c > index 900a17f..058b08c 100644 > --- a/linux-user/main.c > +++ b/linux-user/main.c > @@ -3901,11 +3901,15 @@ int main(int argc, char **argv, char **envp) > #elif defined TARGET_OPENRISC > cpu_model = "or1200"; > #elif defined(TARGET_PPC) > -#ifdef TARGET_PPC64 > +# ifdef TARGET_PPC64 > +# ifdef TARGET_WORDS_BIGENDIAN > cpu_model = "970fx"; > -#else > +# else > + cpu_model = "POWER7"; How about we switch to p7 for BE top? Alex > +# endif > +# else > cpu_model = "750"; > -#endif > +# endif > #else > cpu_model = "any"; > #endif > diff --git a/target-ppc/translate_init.c b/target-ppc/translate_init.c > index 85581c9..72128d8 100644 > --- a/target-ppc/translate_init.c > +++ b/target-ppc/translate_init.c > @@ -9538,6 +9538,10 @@ static void ppc_cpu_reset(CPUState *s) > #endif > #if !defined(TARGET_WORDS_BIGENDIAN) > msr |= (target_ulong)1 << MSR_LE; /* Little-endian user mode */ > + if (!((env->msr_mask >> MSR_LE) & 1)) { > + fprintf(stderr, "Selected CPU does not support little-endian.\n"); > + exit(1); > + } > #endif > #endif > > -- > 1.9.3 >
On 06/28/2014 09:50 AM, Alexander Graf wrote:
> How about we switch to p7 for BE top?
Not ideal until we implement all of p7's insns in TCG:
Warning: Disabling some instructions which are not emulated by TCG (0x0, 0x4)
r~
On 6/28/2014 1:42 PM, Richard Henderson wrote: > On 06/28/2014 09:50 AM, Alexander Graf wrote: >> How about we switch to p7 for BE top? > > Not ideal until we implement all of p7's insns in TCG: > > Warning: Disabling some instructions which are not emulated by TCG (0x0, 0x4) > > > r~ > That was dealt with in this patch: http://lists.nongnu.org/archive/html/qemu-ppc/2014-06/msg00479.html and has now made its way into HEAD.
On 30.06.14 14:08, Tom Musta wrote: > On 6/28/2014 1:42 PM, Richard Henderson wrote: >> On 06/28/2014 09:50 AM, Alexander Graf wrote: >>> How about we switch to p7 for BE top? >> Not ideal until we implement all of p7's insns in TCG: >> >> Warning: Disabling some instructions which are not emulated by TCG (0x0, 0x4) >> >> >> r~ >> > That was dealt with in this patch: http://lists.nongnu.org/archive/html/qemu-ppc/2014-06/msg00479.html and has now made its way into HEAD. > Yup, I'll change the default for BE to POWER7 too while applying. Alex
diff --git a/linux-user/main.c b/linux-user/main.c index 900a17f..058b08c 100644 --- a/linux-user/main.c +++ b/linux-user/main.c @@ -3901,11 +3901,15 @@ int main(int argc, char **argv, char **envp) #elif defined TARGET_OPENRISC cpu_model = "or1200"; #elif defined(TARGET_PPC) -#ifdef TARGET_PPC64 +# ifdef TARGET_PPC64 +# ifdef TARGET_WORDS_BIGENDIAN cpu_model = "970fx"; -#else +# else + cpu_model = "POWER7"; +# endif +# else cpu_model = "750"; -#endif +# endif #else cpu_model = "any"; #endif diff --git a/target-ppc/translate_init.c b/target-ppc/translate_init.c index 85581c9..72128d8 100644 --- a/target-ppc/translate_init.c +++ b/target-ppc/translate_init.c @@ -9538,6 +9538,10 @@ static void ppc_cpu_reset(CPUState *s) #endif #if !defined(TARGET_WORDS_BIGENDIAN) msr |= (target_ulong)1 << MSR_LE; /* Little-endian user mode */ + if (!((env->msr_mask >> MSR_LE) & 1)) { + fprintf(stderr, "Selected CPU does not support little-endian.\n"); + exit(1); + } #endif #endif
The default, 970fx, doesn't support MSR_LE. So even though we set LE in ppc_cpu_reset, it gets cleared again in hreg_store_msr. Error out if a user-selected cpu model doesn't support LE. Cc: Aldy Hernandez <aldyh@redhat.com> Signed-off-by: Richard Henderson <rth@twiddle.net> --- The warning one gets from the unimplemented insns from tcg is perhaps unfortunate, but it's better than silently dropping the MSR_LE bit and interpreting the first few insns with the wrong endianness and generating SIGILL. One could, perhaps, simply return false from need_byteswap. But then the value of MSR would still be wrong, and I can imagine that would affect gdb's interpretation of the current mode. r~ --- linux-user/main.c | 10 +++++++--- target-ppc/translate_init.c | 4 ++++ 2 files changed, 11 insertions(+), 3 deletions(-)