| Message ID | 20250217120812.396522-1-pbonzini@redhat.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | physmem: replace assertion with error | expand |
On Mon, Feb 17, 2025 at 01:08:12PM +0100, Paolo Bonzini wrote: > It is possible to start QEMU with a confidential-guest-support object > even in TCG mode. While there is already a check in qemu_machine_creation_done: > > if (machine->cgs && !machine->cgs->ready) { > error_setg(errp, "accelerator does not support confidential guest %s", > object_get_typename(OBJECT(machine->cgs))); > exit(1); > } > > the creation of RAMBlocks happens earlier, in qemu_init_board(), if > the command line does not override the default memory backend with > -M memdev. Then the RAMBlock will try to use guest_memfd (because > machine_require_guest_memfd correctly returns true; at least correctly > according to the current implementation) and trigger the assertion > failure for kvm_enabled(). This happend with a command line as > simple as the following: > > qemu-system-x86_64 -m 512 -nographic -object sev-snp-guest,reduced-phys-bits=48,id=sev0 \ > -M q35,kernel-irqchip=split,confidential-guest-support=sev0 > qemu-system-x86_64: ../system/physmem.c:1871: ram_block_add: Assertion `kvm_enabled()' failed. > > Cc: Xiaoyao Li <xiaoyao.li@intel.com> > Cc: qemu-stable@nongnu.org > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> > --- > system/physmem.c | 6 +++++- > 1 file changed, 5 insertions(+), 1 deletion(-) Reviewed-by: Daniel P. Berrangé <berrange@redhat.com> With regards, Daniel
Hi Paolo, On 17/2/25 13:08, Paolo Bonzini wrote: > It is possible to start QEMU with a confidential-guest-support object > even in TCG mode. While there is already a check in qemu_machine_creation_done: > > if (machine->cgs && !machine->cgs->ready) { > error_setg(errp, "accelerator does not support confidential guest %s", > object_get_typename(OBJECT(machine->cgs))); > exit(1); > } > > the creation of RAMBlocks happens earlier, in qemu_init_board(), if > the command line does not override the default memory backend with > -M memdev. Then the RAMBlock will try to use guest_memfd (because > machine_require_guest_memfd correctly returns true; at least correctly > according to the current implementation) and trigger the assertion > failure for kvm_enabled(). This happend with a command line as > simple as the following: > > qemu-system-x86_64 -m 512 -nographic -object sev-snp-guest,reduced-phys-bits=48,id=sev0 \ > -M q35,kernel-irqchip=split,confidential-guest-support=sev0 > qemu-system-x86_64: ../system/physmem.c:1871: ram_block_add: Assertion `kvm_enabled()' failed. I'd expect sev-snp-guest to bail out early enough. Is a KVM-enabled check in sev_snp_guest_instance_init() missing? > Cc: Xiaoyao Li <xiaoyao.li@intel.com> > Cc: qemu-stable@nongnu.org > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> > --- > system/physmem.c | 6 +++++- > 1 file changed, 5 insertions(+), 1 deletion(-) > > diff --git a/system/physmem.c b/system/physmem.c > index 67c9db9daad..1ddf9fb10d0 100644 > --- a/system/physmem.c > +++ b/system/physmem.c > @@ -1882,7 +1882,11 @@ static void ram_block_add(RAMBlock *new_block, Error **errp) > if (new_block->flags & RAM_GUEST_MEMFD) { > int ret; > > - assert(kvm_enabled()); > + if (!kvm_enabled()) { > + error_setg(errp, "cannot set up private guest memory for %s: KVM required", > + object_get_typename(OBJECT(current_machine->cgs))); Common code should be SW vs HW accel, so IMHO your check should go within sev_snp_guest_instance_init(), removing the assertion here. That said I have no clue about SEV. > + goto out_free; > + } > assert(new_block->guest_memfd < 0); > > ret = ram_block_discard_require(true);
On Mon, Feb 17, 2025 at 1:28 PM Philippe Mathieu-Daudé <philmd@linaro.org> wrote: > > qemu-system-x86_64 -m 512 -nographic -object sev-snp-guest,reduced-phys-bits=48,id=sev0 \ > > -M q35,kernel-irqchip=split,confidential-guest-support=sev0 > > qemu-system-x86_64: ../system/physmem.c:1871: ram_block_add: Assertion `kvm_enabled()' failed. > > I'd expect sev-snp-guest to bail out early enough. > > Is a KVM-enabled check in sev_snp_guest_instance_init() missing? instance_init cannot do any check (it cannot fail). There is a check for whether sev_common_kvm_init succeeded: if (machine->cgs && !machine->cgs->ready) { error_setg(errp, "accelerator does not support confidential guest %s", object_get_typename(OBJECT(machine->cgs))); exit(1); } but that doesn't help if the system/physmem.c code is reached before qemu_machine_creation_done(), for example in qemu_init_board(). Likewise, you cannot fal in host_memory_backend_init() because that's also an instance_init callback. > > - assert(kvm_enabled()); > > + if (!kvm_enabled()) { > > + error_setg(errp, "cannot set up private guest memory for %s: KVM required", > > + object_get_typename(OBJECT(current_machine->cgs))); > > Common code should be SW vs HW accel, so IMHO your check > should go within sev_snp_guest_instance_init(), removing > the assertion here. That said I have no clue about SEV. This code is outside accel/kvm or target/*/kvn, but it is not quite common code. A few lines below you have: new_block->guest_memfd = kvm_create_guest_memfd(new_block->max_length, 0, errp); Paolo
On 2/17/2025 8:08 PM, Paolo Bonzini wrote: > It is possible to start QEMU with a confidential-guest-support object > even in TCG mode. While there is already a check in qemu_machine_creation_done: > > if (machine->cgs && !machine->cgs->ready) { > error_setg(errp, "accelerator does not support confidential guest %s", > object_get_typename(OBJECT(machine->cgs))); > exit(1); > } > > the creation of RAMBlocks happens earlier, in qemu_init_board(), if > the command line does not override the default memory backend with > -M memdev. Then the RAMBlock will try to use guest_memfd (because > machine_require_guest_memfd correctly returns true; at least correctly > according to the current implementation) and trigger the assertion > failure for kvm_enabled(). This happend with a command line as > simple as the following: > > qemu-system-x86_64 -m 512 -nographic -object sev-snp-guest,reduced-phys-bits=48,id=sev0 \ > -M q35,kernel-irqchip=split,confidential-guest-support=sev0 > qemu-system-x86_64: ../system/physmem.c:1871: ram_block_add: Assertion `kvm_enabled()' failed. > > Cc: Xiaoyao Li <xiaoyao.li@intel.com> > Cc: qemu-stable@nongnu.org > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Reviewed-by: Xiaoyao Li <xiaoyao.li@intel.com> > --- > system/physmem.c | 6 +++++- > 1 file changed, 5 insertions(+), 1 deletion(-) > > diff --git a/system/physmem.c b/system/physmem.c > index 67c9db9daad..1ddf9fb10d0 100644 > --- a/system/physmem.c > +++ b/system/physmem.c > @@ -1882,7 +1882,11 @@ static void ram_block_add(RAMBlock *new_block, Error **errp) > if (new_block->flags & RAM_GUEST_MEMFD) { > int ret; > > - assert(kvm_enabled()); > + if (!kvm_enabled()) { > + error_setg(errp, "cannot set up private guest memory for %s: KVM required", > + object_get_typename(OBJECT(current_machine->cgs))); > + goto out_free; > + } > assert(new_block->guest_memfd < 0); > > ret = ram_block_discard_require(true);
On 17.02.25 13:08, Paolo Bonzini wrote: > It is possible to start QEMU with a confidential-guest-support object > even in TCG mode. While there is already a check in qemu_machine_creation_done: > > if (machine->cgs && !machine->cgs->ready) { > error_setg(errp, "accelerator does not support confidential guest %s", > object_get_typename(OBJECT(machine->cgs))); > exit(1); > } > > the creation of RAMBlocks happens earlier, in qemu_init_board(), if > the command line does not override the default memory backend with > -M memdev. Then the RAMBlock will try to use guest_memfd (because > machine_require_guest_memfd correctly returns true; at least correctly > according to the current implementation) and trigger the assertion > failure for kvm_enabled(). This happend with a command line as > simple as the following: > > qemu-system-x86_64 -m 512 -nographic -object sev-snp-guest,reduced-phys-bits=48,id=sev0 \ > -M q35,kernel-irqchip=split,confidential-guest-support=sev0 > qemu-system-x86_64: ../system/physmem.c:1871: ram_block_add: Assertion `kvm_enabled()' failed. > > Cc: Xiaoyao Li <xiaoyao.li@intel.com> > Cc: qemu-stable@nongnu.org > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> > --- > system/physmem.c | 6 +++++- > 1 file changed, 5 insertions(+), 1 deletion(-) > > diff --git a/system/physmem.c b/system/physmem.c > index 67c9db9daad..1ddf9fb10d0 100644 > --- a/system/physmem.c > +++ b/system/physmem.c > @@ -1882,7 +1882,11 @@ static void ram_block_add(RAMBlock *new_block, Error **errp) > if (new_block->flags & RAM_GUEST_MEMFD) { > int ret; > > - assert(kvm_enabled()); > + if (!kvm_enabled()) { > + error_setg(errp, "cannot set up private guest memory for %s: KVM required", > + object_get_typename(OBJECT(current_machine->cgs))); > + goto out_free; > + } > assert(new_block->guest_memfd < 0); > > ret = ram_block_discard_require(true); Reviewed-by: David Hildenbrand <david@redhat.com>
On 2/17/2025 1:08 PM, Paolo Bonzini wrote: > It is possible to start QEMU with a confidential-guest-support object > even in TCG mode. While there is already a check in qemu_machine_creation_done: > > if (machine->cgs && !machine->cgs->ready) { > error_setg(errp, "accelerator does not support confidential guest %s", > object_get_typename(OBJECT(machine->cgs))); > exit(1); > } > > the creation of RAMBlocks happens earlier, in qemu_init_board(), if > the command line does not override the default memory backend with > -M memdev. Then the RAMBlock will try to use guest_memfd (because > machine_require_guest_memfd correctly returns true; at least correctly > according to the current implementation) and trigger the assertion > failure for kvm_enabled(). This happend with a command line as > simple as the following: > > qemu-system-x86_64 -m 512 -nographic -object sev-snp-guest,reduced-phys-bits=48,id=sev0 \ > -M q35,kernel-irqchip=split,confidential-guest-support=sev0 > qemu-system-x86_64: ../system/physmem.c:1871: ram_block_add: Assertion `kvm_enabled()' failed. > > Cc: Xiaoyao Li <xiaoyao.li@intel.com> > Cc: qemu-stable@nongnu.org > Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> Reviewed-by: Pankaj Gupta <pankaj.gupta@amd.com> > --- > system/physmem.c | 6 +++++- > 1 file changed, 5 insertions(+), 1 deletion(-) > > diff --git a/system/physmem.c b/system/physmem.c > index 67c9db9daad..1ddf9fb10d0 100644 > --- a/system/physmem.c > +++ b/system/physmem.c > @@ -1882,7 +1882,11 @@ static void ram_block_add(RAMBlock *new_block, Error **errp) > if (new_block->flags & RAM_GUEST_MEMFD) { > int ret; > > - assert(kvm_enabled()); > + if (!kvm_enabled()) { > + error_setg(errp, "cannot set up private guest memory for %s: KVM required", > + object_get_typename(OBJECT(current_machine->cgs))); > + goto out_free; > + } > assert(new_block->guest_memfd < 0); > > ret = ram_block_discard_require(true);
diff --git a/system/physmem.c b/system/physmem.c index 67c9db9daad..1ddf9fb10d0 100644 --- a/system/physmem.c +++ b/system/physmem.c @@ -1882,7 +1882,11 @@ static void ram_block_add(RAMBlock *new_block, Error **errp) if (new_block->flags & RAM_GUEST_MEMFD) { int ret; - assert(kvm_enabled()); + if (!kvm_enabled()) { + error_setg(errp, "cannot set up private guest memory for %s: KVM required", + object_get_typename(OBJECT(current_machine->cgs))); + goto out_free; + } assert(new_block->guest_memfd < 0); ret = ram_block_discard_require(true);
It is possible to start QEMU with a confidential-guest-support object even in TCG mode. While there is already a check in qemu_machine_creation_done: if (machine->cgs && !machine->cgs->ready) { error_setg(errp, "accelerator does not support confidential guest %s", object_get_typename(OBJECT(machine->cgs))); exit(1); } the creation of RAMBlocks happens earlier, in qemu_init_board(), if the command line does not override the default memory backend with -M memdev. Then the RAMBlock will try to use guest_memfd (because machine_require_guest_memfd correctly returns true; at least correctly according to the current implementation) and trigger the assertion failure for kvm_enabled(). This happend with a command line as simple as the following: qemu-system-x86_64 -m 512 -nographic -object sev-snp-guest,reduced-phys-bits=48,id=sev0 \ -M q35,kernel-irqchip=split,confidential-guest-support=sev0 qemu-system-x86_64: ../system/physmem.c:1871: ram_block_add: Assertion `kvm_enabled()' failed. Cc: Xiaoyao Li <xiaoyao.li@intel.com> Cc: qemu-stable@nongnu.org Signed-off-by: Paolo Bonzini <pbonzini@redhat.com> --- system/physmem.c | 6 +++++- 1 file changed, 5 insertions(+), 1 deletion(-)