Re: [kvm-unit-tests PATCH v3 11/17] powerpc/ppc64: add hcall support and putchar

[Thomas Huth <thuth@redhat.com>]

[-- Attachment #1: Type: text/plain, Size: 4325 bytes --]

On 15.02.2016 00:29, David Gibson wrote:
> On Sat, Feb 13, 2016 at 05:28:24PM +0100, Andrew Jones wrote:
>> Add broken sc1 detection and patching and an hcall for putchar,
>> to use in puts. That, along with a couple more lines in start to
>> prepare for C code, and a branch to main(), gets us "hello world".
>> Run with
>>
>> qemu-system-ppc64 -M pseries            \
>>         -bios powerpc/boot_rom.bin      \
>>         -display none -serial stdio     \
>>         -kernel powerpc/selftest.elf
>>
>> (We're still not relocating yet, that comes in a later patch. Thus,
>>  testing hello-world at this point requires a hacked QEMU and linking
>>  the unit test at QEMU's kernel load address.)
>>
>> Signed-off-by: Andrew Jones <drjones@redhat.com>
>> ---
>>  lib/powerpc/asm/hcall.h | 43 +++++++++++++++++++++++++++
>>  lib/powerpc/hcall.c     | 79 +++++++++++++++++++++++++++++++++++++++++++++++++
>>  lib/powerpc/io.c        | 15 ++++++++--
>>  lib/ppc64/asm/hcall.h   |  1 +
>>  powerpc/Makefile.common |  2 ++
>>  powerpc/cstart64.S      | 21 +++++++++++++
>>  6 files changed, 159 insertions(+), 2 deletions(-)
>>  create mode 100644 lib/powerpc/asm/hcall.h
>>  create mode 100644 lib/powerpc/hcall.c
>>  create mode 100644 lib/ppc64/asm/hcall.h
>>
>> diff --git a/lib/powerpc/asm/hcall.h b/lib/powerpc/asm/hcall.h
>> new file mode 100644
>> index 0000000000000..75eac456c49eb
>> --- /dev/null
>> +++ b/lib/powerpc/asm/hcall.h
>> @@ -0,0 +1,43 @@
>> +#ifndef _ASMPOWERPC_HCALL_H_
>> +#define _ASMPOWERPC_HCALL_H_
>> +/*
>> + * Copyright (C) 2016, Red Hat Inc, Andrew Jones <drjones@redhat.com>
>> + *
>> + * This work is licensed under the terms of the GNU LGPL, version 2.
>> + */
>> +
>> +#define SC1			0x44000022
>> +#define SC1_REPLACEMENT		0x7c000268
>> +
>> +#define H_SUCCESS		0
>> +#define H_HARDWARE		-1
>> +
>> +#define H_SET_DABR		0x28
>> +#define H_PUT_TERM_CHAR		0x58
>> +
>> +#ifndef __ASSEMBLY__
>> +#include <libcflat.h>
>> +
>> +/*
>> + * hcall_have_broken_sc1 checks if we're on a host with a broken sc1.
>> + * Returns true if we are.
>> + */
>> +extern bool hcall_have_broken_sc1(void);
>> +
>> +/*
>> + * hcall_patch_broken_sc1 patches hcall's sc1 instruction, if needed,
>> + * allowing all hypercalls built on it to work.
>> + */
>> +extern void hcall_patch_broken_sc1(void);
>> +
>> +/*
>> + * hcall is the hypercall wrapper function. unittests may do what
>> + * they like, but the framework should make all hypercalls through
>> + * here to ensure they use a working sc1 instruction, and properly
>> + * handle clobbered registers. @nr is the hypercall number.
>> + */
>> +extern unsigned long
>> +hcall(unsigned long nr, unsigned long *in, unsigned long *out);
>> +
>> +#endif /* !__ASSEMBLY__ */
>> +#endif /* _ASMPOWERPC_HCALL_H_ */
>> diff --git a/lib/powerpc/hcall.c b/lib/powerpc/hcall.c
>> new file mode 100644
>> index 0000000000000..4c42b18ef9901
>> --- /dev/null
>> +++ b/lib/powerpc/hcall.c
>> @@ -0,0 +1,79 @@
>> +/*
>> + * Hypercall helpers
>> + *
>> + * broken_sc1 probing/patching inspired by SLOF, see
>> + *   SLOF:lib/libhvcall/brokensc1.c
>> + *
>> + * hcall() implementation inspired by Linux's epapr_hypercall, see
>> + *   arch/powerpc/include/asm/epapr_hcalls.h
>> + *
>> + * Copyright (C) 2016, Red Hat Inc, Andrew Jones <drjones@redhat.com>
>> + *
>> + * This work is licensed under the terms of the GNU LGPL, version 2.
>> + */
>> +#include <libcflat.h>
>> +#include <asm/hcall.h>
>> +
>> +static u32 sc1[];
>> +
>> +bool hcall_have_broken_sc1(void)
>> +{
>> +	register unsigned long r3 asm("r3") = H_SET_DABR;
>> +	register unsigned long r4 asm("r4") = 0;
>> +
>> +	asm volatile("sc 1"
>> +	: "=r" (r3)
>> +	: "r" (r3), "r" (r4)
>> +	: "r5", "r6", "r7", "r8", "r9", "r10", "r11", "r12");
>> +
>> +	return !(r3 == H_SUCCESS || r3 == (unsigned long)H_HARDWARE);
> 
> It might be safer to explicitly check for H_PRIVILEGE, which is what
> you'll get in the case of a broken sc 1 under nested virtualization.
> That way you won't get false positives if H_SET_DABR ever gets any new
> error codes it can return.

Right - actually, H_SET_DABR already might return H_RESERVED_DABR now if
the host does not support the DABRX register.

 Thomas



[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 836 bytes --]