Donate to e Foundation | Murena handsets with /e/OS | Own a part of Murena! Learn more

Commit 554726d3 authored by Paolo Bonzini's avatar Paolo Bonzini
Browse files

Merge tag 'kvm-s390-next-20150728' of... 

Merge tag 'kvm-s390-next-20150728' of git://git.kernel.org/pub/scm/linux/kernel/git/kvms390/linux into kvm-next

KVM: s390: Fixes and features for kvm/next (4.3)

1. Rework logging infrastructure (s390dbf) to integrate feedback learned
   when debugging performance and test issues
2. Some cleanups and simplifications for CMMA handling
3. Fix gdb debugging and single stepping on some instructions
4. Error handling for storage key setup
parents 54928303 c92ea7b9

Documentation/s390/00-INDEX

+0 −2
Original line number Diff line number Diff line
@@ -16,8 +16,6 @@ Debugging390.txt 
	- hints for debugging on s390 systems.

driver-model.txt

	- information on s390 devices and the driver model.

kvm.txt

	- ioctl calls to /dev/kvm on s390.

monreader.txt

	- information on accessing the z/VM monitor stream from Linux.

qeth.txt

Documentation/s390/kvm.txt

deleted100644 → 0
+0 −125
Original line number Diff line number Diff line 
*** BIG FAT WARNING ***

The kvm module is currently in EXPERIMENTAL state for s390. This means that

the interface to the module is not yet considered to remain stable. Thus, be

prepared that we keep breaking your userspace application and guest

compatibility over and over again until we feel happy with the result. Make sure

your guest kernel, your host kernel, and your userspace launcher are in a

consistent state.



This Documentation describes the unique ioctl calls to /dev/kvm, the resulting

kvm-vm file descriptors, and the kvm-vcpu file descriptors that differ from x86.



1. ioctl calls to /dev/kvm

KVM does support the following ioctls on s390 that are common with other

architectures and do behave the same:

KVM_GET_API_VERSION

KVM_CREATE_VM		(*) see note

KVM_CHECK_EXTENSION

KVM_GET_VCPU_MMAP_SIZE



Notes:

* KVM_CREATE_VM may fail on s390, if the calling process has multiple

threads and has not called KVM_S390_ENABLE_SIE before.



In addition, on s390 the following architecture specific ioctls are supported:

ioctl:		KVM_S390_ENABLE_SIE

args:		none

see also:	include/linux/kvm.h

This call causes the kernel to switch on PGSTE in the user page table. This

operation is needed in order to run a virtual machine, and it requires the

calling process to be single-threaded. Note that the first call to KVM_CREATE_VM

will implicitly try to switch on PGSTE if the user process has not called

KVM_S390_ENABLE_SIE before. User processes that want to launch multiple threads

before creating a virtual machine have to call KVM_S390_ENABLE_SIE, or will

observe an error calling KVM_CREATE_VM. Switching on PGSTE is a one-time

operation, is not reversible, and will persist over the entire lifetime of

the calling process. It does not have any user-visible effect other than a small

performance penalty.



2. ioctl calls to the kvm-vm file descriptor

KVM does support the following ioctls on s390 that are common with other

architectures and do behave the same:

KVM_CREATE_VCPU

KVM_SET_USER_MEMORY_REGION      (*) see note

KVM_GET_DIRTY_LOG		(**) see note



Notes:

*  kvm does only allow exactly one memory slot on s390, which has to start

   at guest absolute address zero and at a user address that is aligned on any

   page boundary. This hardware "limitation" allows us to have a few unique

   optimizations. The memory slot doesn't have to be filled

   with memory actually, it may contain sparse holes. That said, with different

   user memory layout this does still allow a large flexibility when

   doing the guest memory setup.

** KVM_GET_DIRTY_LOG doesn't work properly yet. The user will receive an empty

log. This ioctl call is only needed for guest migration, and we intend to

implement this one in the future.



In addition, on s390 the following architecture specific ioctls for the kvm-vm

file descriptor are supported:

ioctl:		KVM_S390_INTERRUPT

args:		struct kvm_s390_interrupt *

see also:	include/linux/kvm.h

This ioctl is used to submit a floating interrupt for a virtual machine.

Floating interrupts may be delivered to any virtual cpu in the configuration.

Only some interrupt types defined in include/linux/kvm.h make sense when

submitted as floating interrupts. The following interrupts are not considered

to be useful as floating interrupts, and a call to inject them will result in

-EINVAL error code: program interrupts and interprocessor signals. Valid

floating interrupts are:

KVM_S390_INT_VIRTIO

KVM_S390_INT_SERVICE



3. ioctl calls to the kvm-vcpu file descriptor

KVM does support the following ioctls on s390 that are common with other

architectures and do behave the same:

KVM_RUN

KVM_GET_REGS

KVM_SET_REGS

KVM_GET_SREGS

KVM_SET_SREGS

KVM_GET_FPU

KVM_SET_FPU



In addition, on s390 the following architecture specific ioctls for the

kvm-vcpu file descriptor are supported:

ioctl:		KVM_S390_INTERRUPT

args:		struct kvm_s390_interrupt *

see also:	include/linux/kvm.h

This ioctl is used to submit an interrupt for a specific virtual cpu.

Only some interrupt types defined in include/linux/kvm.h make sense when

submitted for a specific cpu. The following interrupts are not considered

to be useful, and a call to inject them will result in -EINVAL error code:

service processor calls and virtio interrupts. Valid interrupt types are:

KVM_S390_PROGRAM_INT

KVM_S390_SIGP_STOP

KVM_S390_RESTART

KVM_S390_SIGP_SET_PREFIX

KVM_S390_INT_EMERGENCY



ioctl:		KVM_S390_STORE_STATUS

args:		unsigned long

see also:	include/linux/kvm.h

This ioctl stores the state of the cpu at the guest real address given as

argument, unless one of the following values defined in include/linux/kvm.h

is given as argument:

KVM_S390_STORE_STATUS_NOADDR - the CPU stores its status to the save area in

absolute lowcore as defined by the principles of operation

KVM_S390_STORE_STATUS_PREFIXED - the CPU stores its status to the save area in

its prefix page just like the dump tool that comes with zipl. This is useful

to create a system dump for use with lkcdutils or crash.



ioctl:		KVM_S390_SET_INITIAL_PSW

args:		struct kvm_s390_psw *

see also:	include/linux/kvm.h

This ioctl can be used to set the processor status word (psw) of a stopped cpu

prior to running it with KVM_RUN. Note that this call is not required to modify

the psw during sie intercepts that fall back to userspace because struct kvm_run

does contain the psw, and this value is evaluated during reentry of KVM_RUN

after the intercept exit was recognized.



ioctl:		KVM_S390_INITIAL_RESET

args:		none

see also:	include/linux/kvm.h

This ioctl can be used to perform an initial cpu reset as defined by the

principles of operation. The target cpu has to be in stopped state.

arch/s390/include/asm/kvm_host.h

+3 −1
Original line number Diff line number Diff line
@@ -258,6 +258,9 @@ struct kvm_vcpu_stat { 
	u32 diagnose_10;

	u32 diagnose_44;

	u32 diagnose_9c;

	u32 diagnose_258;

	u32 diagnose_308;

	u32 diagnose_500;

};



#define PGM_OPERATION			0x01
@@ -630,7 +633,6 @@ extern char sie_exit; 


static inline void kvm_arch_hardware_disable(void) {}

static inline void kvm_arch_check_processor_compat(void *rtn) {}

static inline void kvm_arch_exit(void) {}

static inline void kvm_arch_sync_events(struct kvm *kvm) {}

static inline void kvm_arch_vcpu_uninit(struct kvm_vcpu *vcpu) {}

static inline void kvm_arch_sched_in(struct kvm_vcpu *vcpu, int cpu) {}

arch/s390/kvm/diag.c

+11 −2
Original line number Diff line number Diff line
@@ -27,13 +27,13 @@ static int diag_release_pages(struct kvm_vcpu *vcpu) 


	start = vcpu->run->s.regs.gprs[(vcpu->arch.sie_block->ipa & 0xf0) >> 4];

	end = vcpu->run->s.regs.gprs[vcpu->arch.sie_block->ipa & 0xf] + 4096;

	vcpu->stat.diagnose_10++;



	if (start & ~PAGE_MASK || end & ~PAGE_MASK || start >= end

	    || start < 2 * PAGE_SIZE)

		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);



	VCPU_EVENT(vcpu, 5, "diag release pages %lX %lX", start, end);

	vcpu->stat.diagnose_10++;



	/*

	 * We checked for start >= end above, so lets check for the
@@ -75,6 +75,9 @@ static int __diag_page_ref_service(struct kvm_vcpu *vcpu) 
	u16 rx = (vcpu->arch.sie_block->ipa & 0xf0) >> 4;

	u16 ry = (vcpu->arch.sie_block->ipa & 0x0f);



	VCPU_EVENT(vcpu, 3, "diag page reference parameter block at 0x%llx",

		   vcpu->run->s.regs.gprs[rx]);

	vcpu->stat.diagnose_258++;

	if (vcpu->run->s.regs.gprs[rx] & 7)

		return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);

	rc = read_guest(vcpu, vcpu->run->s.regs.gprs[rx], rx, &parm, sizeof(parm));
@@ -85,6 +88,9 @@ static int __diag_page_ref_service(struct kvm_vcpu *vcpu) 


	switch (parm.subcode) {

	case 0: /* TOKEN */

		VCPU_EVENT(vcpu, 3, "pageref token addr 0x%llx "

			   "select mask 0x%llx compare mask 0x%llx",

			   parm.token_addr, parm.select_mask, parm.compare_mask);

		if (vcpu->arch.pfault_token != KVM_S390_PFAULT_TOKEN_INVALID) {

			/*

			 * If the pagefault handshake is already activated,
@@ -114,6 +120,7 @@ static int __diag_page_ref_service(struct kvm_vcpu *vcpu) 
		 * the cancel, therefore to reduce code complexity, we assume

		 * all outstanding tokens are already pending.

		 */

		VCPU_EVENT(vcpu, 3, "pageref cancel addr 0x%llx", parm.token_addr);

		if (parm.token_addr || parm.select_mask ||

		    parm.compare_mask || parm.zarch)

			return kvm_s390_inject_program_int(vcpu, PGM_SPECIFICATION);
@@ -174,7 +181,8 @@ static int __diag_ipl_functions(struct kvm_vcpu *vcpu) 
	unsigned int reg = vcpu->arch.sie_block->ipa & 0xf;

	unsigned long subcode = vcpu->run->s.regs.gprs[reg] & 0xffff;



	VCPU_EVENT(vcpu, 5, "diag ipl functions, subcode %lx", subcode);

	VCPU_EVENT(vcpu, 3, "diag ipl functions, subcode %lx", subcode);

	vcpu->stat.diagnose_308++;

	switch (subcode) {

	case 3:

		vcpu->run->s390_reset_flags = KVM_S390_RESET_CLEAR;
@@ -202,6 +210,7 @@ static int __diag_virtio_hypercall(struct kvm_vcpu *vcpu) 
{

	int ret;



	vcpu->stat.diagnose_500++;

	/* No virtio-ccw notification? Get out quickly. */

	if (!vcpu->kvm->arch.css_support ||

	    (vcpu->run->s.regs.gprs[1] != KVM_S390_VIRTIO_CCW_NOTIFY))

arch/s390/kvm/guestdbg.c

+35 −0
Original line number Diff line number Diff line
@@ -473,10 +473,45 @@ static void filter_guest_per_event(struct kvm_vcpu *vcpu) 
		vcpu->arch.sie_block->iprcc &= ~PGM_PER;

}



#define pssec(vcpu) (vcpu->arch.sie_block->gcr[1] & _ASCE_SPACE_SWITCH)

#define hssec(vcpu) (vcpu->arch.sie_block->gcr[13] & _ASCE_SPACE_SWITCH)

#define old_ssec(vcpu) ((vcpu->arch.sie_block->tecmc >> 31) & 0x1)

#define old_as_is_home(vcpu) !(vcpu->arch.sie_block->tecmc & 0xffff)



void kvm_s390_handle_per_event(struct kvm_vcpu *vcpu)

{

	int new_as;



	if (debug_exit_required(vcpu))

		vcpu->guest_debug |= KVM_GUESTDBG_EXIT_PENDING;



	filter_guest_per_event(vcpu);



	/*

	 * Only RP, SAC, SACF, PT, PTI, PR, PC instructions can trigger

	 * a space-switch event. PER events enforce space-switch events

	 * for these instructions. So if no PER event for the guest is left,

	 * we might have to filter the space-switch element out, too.

	 */

	if (vcpu->arch.sie_block->iprcc == PGM_SPACE_SWITCH) {

		vcpu->arch.sie_block->iprcc = 0;

		new_as = psw_bits(vcpu->arch.sie_block->gpsw).as;



		/*

		 * If the AS changed from / to home, we had RP, SAC or SACF

		 * instruction. Check primary and home space-switch-event

		 * controls. (theoretically home -> home produced no event)

		 */

		if (((new_as == PSW_AS_HOME) ^ old_as_is_home(vcpu)) &&

		     (pssec(vcpu) || hssec(vcpu)))

			vcpu->arch.sie_block->iprcc = PGM_SPACE_SWITCH;



		/*

		 * PT, PTI, PR, PC instruction operate on primary AS only. Check

		 * if the primary-space-switch-event control was or got set.

		 */

		if (new_as == PSW_AS_PRIMARY && !old_as_is_home(vcpu) &&

		    (pssec(vcpu) || old_ssec(vcpu)))

			vcpu->arch.sie_block->iprcc = PGM_SPACE_SWITCH;

	}

}