On 11.06.2018 15:56, David Hildenbrand wrote: > On 11.06.2018 15:52, Christian Borntraeger wrote: >> Right now the IPL device always starts from address 0x10000 (the usual >> Linux entry point). To run other guests (e.g. test programs) it is >> useful to use the IPL PSW from address 0. We can use the Linux magic >> at 0x10008 to decide. >> >> Signed-off-by: Christian Borntraeger <[email protected]> >> --- >> v1->v2: >> - use LINUX_MAGIC_ADDR define >> - use assert for valid iplpsw pointer >> - add endianess conversion >> hw/s390x/ipl.c | 15 ++++++++++++++- >> 1 file changed, 14 insertions(+), 1 deletion(-) >> >> diff --git a/hw/s390x/ipl.c b/hw/s390x/ipl.c >> index 04245b5258..3790153fa9 100644 >> --- a/hw/s390x/ipl.c >> +++ b/hw/s390x/ipl.c >> @@ -29,6 +29,7 @@ >> #include "exec/exec-all.h" >> >> #define KERN_IMAGE_START 0x010000UL >> +#define LINUX_MAGIC_ADDR 0x010008UL >> #define KERN_PARM_AREA 0x010480UL >> #define INITRD_START 0x800000UL >> #define INITRD_PARM_START 0x010408UL >> @@ -105,7 +106,9 @@ static uint64_t bios_translate_addr(void *opaque, >> uint64_t srcaddr) >> static void s390_ipl_realize(DeviceState *dev, Error **errp) >> { >> S390IPLState *ipl = S390_IPL(dev); >> - uint64_t pentry = KERN_IMAGE_START; >> + uint64_t *iplpsw; >> + uint64_t pentry; >> + char *magic; >> int kernel_size; >> Error *err = NULL; >> >> @@ -157,6 +160,16 @@ static void s390_ipl_realize(DeviceState *dev, Error >> **errp) >> NULL, 1, EM_S390, 0, 0); >> if (kernel_size < 0) { >> kernel_size = load_image_targphys(ipl->kernel, 0, ram_size); >> + /* if this is Linux use KERN_IMAGE_START */ >> + magic = rom_ptr(LINUX_MAGIC_ADDR); >> + if (magic && !memcmp(magic, "S390EP", 6)) { >> + pentry = KERN_IMAGE_START; >> + } else { >> + /* if not Linux use the IPL PSW */ >> + iplpsw = rom_ptr(0); >> + assert(iplpsw); >> + pentry = be64_to_cpu(*iplpsw) & 0x7fffffffUL; >> + } >> } >> if (kernel_size < 0) { >> error_setg(&err, "could not load kernel '%s'", ipl->kernel); >> > > Have you tried this with kvm-unit-tests? (no magic but we rely on 0x10000)
kvm-unit-tests are elf files which should be handled by the load_elf() some lines earlier already, so I think we should be fine there, shouldn't we? Thomas
