- 27 Mar, 2020 1 commit
-
-
Christian Borntraeger authored
If a signal is pending we might return -ENOMEM instead of -EINTR. We should propagate the proper error during KSM unsharing. unmerge_ksm_pages returns -ERESTARTSYS on signal_pending. This gets translated by entry.S to -EINTR. It is important to get this error code so that userspace can retry. To make this clearer we also add -EINTR to the documentation of the PV_ENABLE call, which calls unmerge_ksm_pages. Fixes: 3ac8e380 ("s390/mm: disable KSM for storage key enabled pages") Reviewed-by:
Janosch Frank <frankja@linux.vnet.ibm.com> Reported-by:
Marc Hartmayer <mhartmay@linux.ibm.com> Tested-by:
David Hildenbrand <david@redhat.com> Reviewed-by:
Cornelia Huck <cohuck@redhat.com> Signed-off-by:
Christian Borntraeger <borntraeger@de.ibm.com>
-
- 23 Mar, 2020 2 commits
-
-
Christian Borntraeger authored
The sie block must be aligned to 512 bytes. Mark it as such. Signed-off-by:
-
Joe Perches authored
Convert the various uses of fallthrough comments to fallthrough; Done via script Link: https://lore.kernel.org/lkml/b56602fcf79f849e733e7b521bb0e17895d390fa.1582230379.git.joe@perches.comSigned-off-by:
Joe Perches <joe@perches.com> Link: https://lore.kernel.org/r/d63c86429f3e5aa806aa3e185c97d213904924a5.1583896348.git.joe@perches.com [borntrager@de.ibm.com: Fix link to tool and subject] Signed-off-by:
-
- 27 Feb, 2020 37 commits
-
-
Michael Mueller authored
The boolean module parameter "kvm.use_gisa" controls if newly created guests will use the GISA facility if provided by the host system. The default is yes. # cat /sys/module/kvm/parameters/use_gisa Y The parameter can be changed on the fly. # echo N > /sys/module/kvm/parameters/use_gisa Already running guests are not affected by this change. The kvm s390 debug feature shows if a guest is running with GISA. # grep gisa /sys/kernel/debug/s390dbf/kvm-$pid/sprintf 00 01582725059:843303 3 - 08 00000000e119bc01 gisa 0x00000000c9ac2642 initialized 00 01582725059:903840 3 - 11 000000004391ee22 00[0000000000000000-0000000000000000]: AIV gisa format-1 enabled for cpu 000 ... 00 01582725059:916847 3 - 08 0000000094fff572 gisa 0x00000000c9ac2642 cleared In general, that value should not be changed as the GISA facility enhances interruption delivery performance. A reason to switch the GISA facility off might be a performance comparison run or debugging. Signed-off-by:
Michael Mueller <mimu@linux.ibm.com> Link: https://lore.kernel.org/r/20200227091031.102993-1-mimu@linux.ibm.comReviewed-by:
-
Janosch Frank authored
Add documentation for KVM_CAP_S390_PROTECTED capability and the KVM_S390_PV_COMMAND ioctl. Signed-off-by:
Janosch Frank <frankja@linux.ibm.com> Reviewed-by:
-
Christian Borntraeger authored
Now that everything is in place, we can announce the feature. Signed-off-by:
-
Janosch Frank authored
Add documentation about protected KVM guests and description of changes that are necessary to move a KVM VM into Protected Virtualization mode. Signed-off-by:
-
Janosch Frank authored
For protected VMs, the VCPU resets are done by the Ultravisor, as KVM has no access to the VCPU registers. Note that the ultravisor will only accept a call for the exact reset that has been requested. Signed-off-by:
Thomas Huth <thuth@redhat.com> Reviewed-by:
-
Christian Borntraeger authored
As PSW restart is handled by the ultravisor (and we only get a start notification) we must re-check the PSW after a start before injecting interrupts. Signed-off-by:
-
Janosch Frank authored
We're not allowed to inject interrupts on intercepts that leave the guest state in an "in-between" state where the next SIE entry will do a continuation, namely secure instruction interception (104) and secure prefix interception (112). As our PSW is just a copy of the real one that will be replaced on the next exit, we can mask out the interrupt bits in the PSW to make sure that we do not inject anything. Signed-off-by:
-
Janosch Frank authored
Code 5 for the set cpu state UV call tells the UV to load a PSW from the SE header (first IPL) or from guest location 0x0 (diag 308 subcode 0/1). Also it sets the cpu into operating state afterwards, so we can start it. Signed-off-by:
-
Janosch Frank authored
VCPU states have to be reported to the ultravisor for SIGP interpretation, kdump, kexec and reboot. Signed-off-by:
-
Janosch Frank authored
diag 308 subcode 0 and 1 require several KVM and Ultravisor interactions. Specific to these "soft" reboots are * The "unshare all" UVC * The "prepare for reset" UVC Signed-off-by:
-
Janosch Frank authored
Only two program exceptions can be injected for a protected guest: specification and operand. For both, a code needs to be specified in the interrupt injection control of the state description, as the guest prefix page is not accessible to KVM for such guests. Signed-off-by:
-
Janosch Frank authored
A lot of the registers are controlled by the Ultravisor and never visible to KVM. Also some registers are overlayed, like gbea is with sidad, which might leak data to userspace. Hence we sync a minimal set of registers for both SIE formats and then check and sync format 2 registers if necessary. Signed-off-by:
-
Janosch Frank authored
For protected VMs the hypervisor can not access guest breaking event address, program parameter, bpbc and todpr. Do not reset those fields as the control block does not provide access to these fields. Signed-off-by:
-
Janosch Frank authored
A lot of the registers are controlled by the Ultravisor and never visible to KVM. Some fields in the sie control block are overlayed, like gbea. As no known userspace uses the ONE_REG interface on s390 if sync regs are available, no functionality is lost if it is disabled for protected guests. Signed-off-by:
-
Janosch Frank authored
Save response to sidad and disable address checking for protected guests. Signed-off-by:
-
Janosch Frank authored
STHYI data has to go through the bounce buffer. Signed-off-by:
-
Claudio Imbrenda authored
The current code tries to first pin shared pages, if that fails (e.g. because the page is not shared) it will export them. For shared pages this means that we get a new intercept telling us that the guest is unsharing that page. We will unpin the page at that point in time, following the same rules as for making a page secure (i.e. waiting for writeback, no elevated page references, etc.) Signed-off-by:
Claudio Imbrenda <imbrenda@linux.ibm.com> Acked-by:
-
Janosch Frank authored
The SPX instruction is handled by the ultravisor. We do get a notification intercept, though. Let us update our internal view. In addition to that, when the guest prefix page is not secure, an intercept 112 (0x70) is indicated. Let us make the prefix pages secure again. Signed-off-by:
-
Janosch Frank authored
Now that we can't access guest memory anymore, we have a dedicated satellite block that's a bounce buffer for instruction data. We re-use the memop interface to copy the instruction data to / from userspace. This lets us re-use a lot of QEMU code which used that interface to make logical guest memory accesses which are not possible anymore in protected mode anyway. Signed-off-by:
-
Janosch Frank authored
Guest registers for protected guests are stored at offset 0x380. We will copy those to the usual places. Long term we could refactor this or use register access functions. Signed-off-by:
-
Janosch Frank authored
SIE intercept code 8 is used only on exception loops for protected guests. That means we need to stop the guest when we see it. This is done by userspace. Signed-off-by:
-
Christian Borntraeger authored
The sclp interrupt is kind of special. The ultravisor polices that we do not inject an sclp interrupt with payload if no sccb is outstanding. On the other hand we have "asynchronous" event interrupts, e.g. for console input. We separate both variants into sclp interrupt and sclp event interrupt. The sclp interrupt is masked until a previous servc instruction has finished (sie exit 108). [frankja@linux.ibm.com: factoring out write_sclp] Signed-off-by:
-
Michael Mueller authored
This defines the necessary data structures in the SIE control block to inject machine checks,external and I/O interrupts. We first define the the interrupt injection control, which defines the next interrupt to inject. Then we define the fields that contain the payload for machine checks,external and I/O interrupts. This is then used to implement interruption injection for the following list of interruption types: - I/O (uses inject io interruption) __deliver_io - External (uses inject external interruption) __deliver_cpu_timer __deliver_ckc __deliver_emergency_signal __deliver_external_call - cpu restart (uses inject restart interruption) __deliver_restart - machine checks (uses mcic, failing address and external damage) __write_machine_check Please note that posted interrupts (GISA) are not used for protected guests as of today. The service interrupt is handled in a followup patch. Signed-off-by: -
Janosch Frank authored
We have two new SIE exit codes dealing with instructions. 104 (0x68) for a secure instruction interception, on which the SIE needs hypervisor action to complete the instruction. We can piggy-back on the existing instruction handlers. 108 which is merely a notification and provides data for tracking and management. For example this is used to tell the host about a new value for the prefix register. As there will be several special case handlers in later patches, we handle this in a separate function. Signed-off-by:
-
Janosch Frank authored
Since there is no interception for load control and load psw instruction in the protected mode, we need a new way to get notified whenever we can inject an IRQ right after the guest has just enabled the possibility for receiving them. The new interception codes solve that problem by providing a notification for changes to IRQ enablement relevant bits in CRs 0, 6 and 14, as well a the machine check mask bit in the PSW. No special handling is needed for these interception codes, the KVM pre-run code will consult all necessary CRs and PSW bits and inject IRQs the guest is enabled for. Signed-off-by:
-
Christian Borntraeger authored
Before we destroy the secure configuration, we better make all pages accessible again. This also happens during reboot, where we reboot into a non-secure guest that then can go again into secure mode. As this "new" secure guest will have a new ID we cannot reuse the old page state. Signed-off-by:
-
Janosch Frank authored
KSM will not work on secure pages, because when the kernel reads a secure page, it will be encrypted and hence no two pages will look the same. Let's mark the guest pages as unmergeable when we transition to secure mode. Signed-off-by:
-
Janosch Frank authored
This contains 3 main changes: 1. changes in SIE control block handling for secure guests 2. helper functions for create/destroy/unpack secure guests 3. KVM_S390_PV_COMMAND ioctl to allow userspace dealing with secure machines Signed-off-by:
-
Janosch Frank authored
This adds two new helper functions for doing UV CALLs. The first variant handles UV CALLs that might have longer busy conditions or just need longer when doing partial completion. We should schedule when necessary. The second variant handles UV CALLs that only need the handle but have no payload (e.g. destroying a VM). We can provide a simple wrapper for those. Signed-off-by:
-
Janosch Frank authored
Let's have some debug traces which stay around for longer than the guest. Signed-off-by:
-
Ulrich Weigand authored
The adapter interrupt page containing the indicator bits is currently pinned. That means that a guest with many devices can pin a lot of memory pages in the host. This also complicates the reference tracking which is needed for memory management handling of protected virtual machines. It might also have some strange side effects for madvise MADV_DONTNEED and other things. We can simply try to get the userspace page set the bits and free the page. By storing the userspace address in the irq routing entry instead of the guest address we can actually avoid many lookups and list walks so that this variant is very likely not slower. If userspace messes around with the memory slots the worst thing that can happen is that we write to some other memory within that process. As we get the the page with FOLL_WRITE this can also not be used to write to shared read-only pages. Signed-off-by:
Ulrich Weigand <Ulrich.Weigand@de.ibm.com> Acked-by:
-
git://git.kernel.org/pub/scm/linux/kernel/git/kvms390/linuxChristian Borntraeger authored
s390 base parts (non kvm) for protvirt
-
Janosch Frank authored
That information, e.g. the maximum number of guests or installed Ultravisor facilities, is interesting for QEMU, Libvirt and administrators. Let's provide an easily parsable API to get that information. Signed-off-by:
-
Vasily Gorbik authored
Add exceptions handlers performing transparent transition of non-secure pages to secure (import) upon guest access and secure pages to non-secure (export) upon hypervisor access. Signed-off-by:
Vasily Gorbik <gor@linux.ibm.com> [frankja@linux.ibm.com: adding checks for failures] Signed-off-by:
-
Claudio Imbrenda authored
This provides the basic ultravisor calls and page table handling to cope with secure guests: - provide arch_make_page_accessible - make pages accessible after unmapping of secure guests - provide the ultravisor commands convert to/from secure - provide the ultravisor commands pin/unpin shared - provide callbacks to make pages secure (inacccessible) - we check for the expected pin count to only make pages secure if the host is not accessing them - we fence hugetlbfs for secure pages - add missing radix-tree include into gmap.h The basic idea is that a page can have 3 states: secure, normal or shared. The hypervisor can call into a firmware function called ultravisor that allows to change the state of a page: convert from/to secure. The convert from secure will encrypt the page and make it available to the host and host I/O. The convert to secure will remove the host capability to access this page. The design is that on convert to secure we will wait until writeback and page refs are indicating no host usage. At the same time the convert from secure (export to host) will be called in common code when the refcount or the writeback bit is already set. This avoids races between convert from and to secure. Then there is also the concept of shared pages. Those are kind of secure where the host can still access those pages. We need to be notified when the guest "unshares" such a page, basically doing a convert to secure by then. There is a call "pin shared page" that we use instead of convert from secure when possible. We do use PG_arch_1 as an optimization to minimize the convert from secure/pin shared. Several comments have been added in the code to explain the logic in the relevant places. Co-developed-by:
-
Vasily Gorbik authored
Before being able to host protected virtual machines, donate some of the memory to the ultravisor. Besides that the ultravisor might impose addressing limitations for memory used to back protected VM storage. Treat that limit as protected virtualization host's virtual memory limit. Signed-off-by:
-
Vasily Gorbik authored
Add "prot_virt" command line option which controls if the kernel protected VMs support is enabled at early boot time. This has to be done early, because it needs large amounts of memory and will disable some features like STP time sync for the lpar. Extend ultravisor info definitions and expose it via uv_info struct filled in during startup. Signed-off-by:
-
