On 09.12.2013, at 18:11, Greg Kurz <gk...@linux.vnet.ibm.com> wrote:
> On Mon, 9 Dec 2013 16:33:59 +0100
> Alexander Graf <ag...@suse.de> wrote:
>>
>> On 25.11.2013, at 16:35, Greg Kurz <gk...@linux.vnet.ibm.com> wrote:
>>
>>> We base it on the OS endian, as reflected by the endianness of the
>>> interrupt vectors (handled through the ILE bit in the LPCR register).
>>>
>>> This patch does two things:
>>> - make LPCR a KVM register
>>> - implement virtio_get_byteswap() over LPCR
>>>
>>> This patch requires to have the following defined in the linux headers:
>>>
>>> $ grep LPCR linux-headers/asm-powerpc/kvm.h
>>> #define KVM_REG_PPC_LPCR (KVM_REG_PPC | KVM_REG_SIZE_U32 | 0xb5)
>>>
>>> Suggested-by: Benjamin Herrenschmidt <b...@kernel.crashing.org>
>>> Signed-off-by: Rusty Russell <ru...@rustcorp.com.au>
>>> Signed-off-by: Greg Kurz <gk...@linux.vnet.ibm.com>
>>> ---
>>> target-ppc/kvm.c | 4 ++++
>>> target-ppc/misc_helper.c | 14 ++++++++++++++
>>> 2 files changed, 18 insertions(+)
>>>
>>> diff --git a/target-ppc/kvm.c b/target-ppc/kvm.c
>>> index 10d0cd9..b450a22 100644
>>> --- a/target-ppc/kvm.c
>>> +++ b/target-ppc/kvm.c
>>> @@ -869,6 +869,8 @@ int kvm_arch_put_registers(CPUState *cs, int level)
>>> DPRINTF("Warning: Unable to set VPA information to
>>> KVM\n"); }
>>> }
>>> +
>>> + kvm_put_one_spr(cs, KVM_REG_PPC_LPCR, SPR_LPCR);
>>> #endif /* TARGET_PPC64 */
>>> }
>>>
>>> @@ -1091,6 +1093,8 @@ int kvm_arch_get_registers(CPUState *cs)
>>> DPRINTF("Warning: Unable to get VPA information from
>>> KVM\n"); }
>>> }
>>> +
>>> + kvm_get_one_spr(cs, KVM_REG_PPC_LPCR, SPR_LPCR);
>>> #endif
>>> }
>>>
>>> diff --git a/target-ppc/misc_helper.c b/target-ppc/misc_helper.c
>>> index 616aab6..0e0743a 100644
>>> --- a/target-ppc/misc_helper.c
>>> +++ b/target-ppc/misc_helper.c
>>> @@ -20,6 +20,8 @@
>>> #include "helper.h"
>>>
>>> #include "helper_regs.h"
>>> +#include "hw/virtio/virtio.h"
>>> +#include "sysemu/kvm.h"
>>>
>>> /*****************************************************************************/
>>> /* SPR accesses */
>>> @@ -116,3 +118,15 @@ void ppc_store_msr(CPUPPCState *env, target_ulong
>>> value) {
>>> hreg_store_msr(env, value, 0);
>>> }
>>> +
>>> +bool virtio_get_byteswap(void)
>>> +{
>>> + PowerPCCPU *cp = POWERPC_CPU(first_cpu);
>>> + CPUPPCState *env = &cp->env;
>>> +
>>> + if (kvm_enabled()) {
>>> + kvm_arch_get_registers(first_cpu);
>>
>> This function is not defined when CONFIG_KVM is disabled.
>>
>> Please do
>>
>> cpu_synchronize_state(first_cpu);
>>
>> instead.
>>
>
> Oups... I'll fix that.
>
>> Also can't virtio_get_byteswap pass in the CPU pointer of the CPU that's
>> calling in this moment? I'm not sure how racy it is to synchronize the
>> first cpu while we're not in the first cpu's execution thread.
>>
>
> I kept the choices made by Rusty in the original serie. According to
> these, I am not sure if we can use current_cpu:
>
> https://lists.gnu.org/archive/html/qemu-devel/2013-08/msg01156.html
> https://lists.gnu.org/archive/html/qemu-devel/2013-08/msg01504.html
>
>> Either way, if we do this "right" we don't even have to jump through
>> these hoops, as little endian setting simply happens steered from QEMU,
>> so QEMU will have all knowledge about the guest's little endian mode
>> without the need to synchronize any state.
>>
>
> Sure. This is just primary work to get virtio working for cross-endian
> cases. :)
Yeah, but so far LE switching code is not upstream in KVM - and for TCG we
don't need the cpu_synchronize_state() call at all.
Alex
