Re: [Qemu-devel] [PATCH v3 0/6] target-i386: Make most CPU models work with enforce out of the box
Am 04.10.2014 um 00:16 schrieb Paolo Bonzini: Il 03/10/2014 21:39, Eduardo Habkost ha scritto: Changes v2 - v3: * None. This is just a rebase against latest qemu.git master (commit b00a0dd) Changes v1 - v2: * Commit message and comment changes. * Update compat code to change pc-*-2.1, not pc-*-2.0. * Added patch to disable SVM by default in KVM mode. Most of the bits that make enforce breaks were introduced in 2010 by commit 8560efed6a72a816c0115f41ddb9d79f7ce63f28. The intention behind that commit made sense, the only problem is that we can't guarantee guest ABI stability across hosts if we simply rely on trimming of CPU features based on host capabilities. So, this series remove CPUID bits from the CPU model definitions so they become defaults that: 1) won't unexpectly stop working when we start using the enforce flag; 2) won't silently break the guest ABI when TCG or KVM start supporting new features. There's only one non-trivial case left: the qemu32/qemu64 models. The problem with them is that we have conflicting expectations about it, from different users: TCG users expect the default CPU model to contain most TCG-supported features (and it makes sense). See, for example, commit f1e00a9cf326acc1f2386a72525af8859852e1df. KVM users expect the default CPU model to be a conservative choice which will work on most host CPUs (and will only contain features that are supported by KVM). We could solve the qemu32/qemu64 issue by having different defaults for TCG and KVM. But we have existing management code (libvirt) that already expects qemu32 or qemu64 to be the default, and changing the default would break that code. I will send an RFC to address that later. Cc: Aurelien Jarno aurel...@aurel32.net Cc: Paolo Bonzini pbonz...@redhat.com Cc: kvm@vger.kernel.org Eduardo Habkost (6): pc: Create pc_compat_2_1() functions target-i386: Rename KVM auto-feature-enable compat function target-i386: Disable CPUID_ACPI by default on KVM mode target-i386: Remove unsupported bits from all CPU models target-i386: Don't enable nested VMX by default target-i386: Disable SVM by default in KVM mode hw/i386/pc_piix.c | 22 ++ hw/i386/pc_q35.c | 18 -- target-i386/cpu.c | 42 -- target-i386/cpu.h | 3 ++- 4 files changed, 64 insertions(+), 21 deletions(-) Andreas, are you picking up this? Thanks, applied to qom-cpu: https://github.com/afaerber/qemu-cpu/commits/qom-cpu I do still have some comments, but let's get this in first. Regards, Andreas -- SUSE LINUX GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 21284 AG Nürnberg -- To unsubscribe from this list: send the line unsubscribe kvm in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
[PATCH v3 0/6] target-i386: Make most CPU models work with enforce out of the box
Changes v2 - v3: * None. This is just a rebase against latest qemu.git master (commit b00a0dd) Changes v1 - v2: * Commit message and comment changes. * Update compat code to change pc-*-2.1, not pc-*-2.0. * Added patch to disable SVM by default in KVM mode. Most of the bits that make enforce breaks were introduced in 2010 by commit 8560efed6a72a816c0115f41ddb9d79f7ce63f28. The intention behind that commit made sense, the only problem is that we can't guarantee guest ABI stability across hosts if we simply rely on trimming of CPU features based on host capabilities. So, this series remove CPUID bits from the CPU model definitions so they become defaults that: 1) won't unexpectly stop working when we start using the enforce flag; 2) won't silently break the guest ABI when TCG or KVM start supporting new features. There's only one non-trivial case left: the qemu32/qemu64 models. The problem with them is that we have conflicting expectations about it, from different users: TCG users expect the default CPU model to contain most TCG-supported features (and it makes sense). See, for example, commit f1e00a9cf326acc1f2386a72525af8859852e1df. KVM users expect the default CPU model to be a conservative choice which will work on most host CPUs (and will only contain features that are supported by KVM). We could solve the qemu32/qemu64 issue by having different defaults for TCG and KVM. But we have existing management code (libvirt) that already expects qemu32 or qemu64 to be the default, and changing the default would break that code. I will send an RFC to address that later. Cc: Aurelien Jarno aurel...@aurel32.net Cc: Paolo Bonzini pbonz...@redhat.com Cc: kvm@vger.kernel.org Eduardo Habkost (6): pc: Create pc_compat_2_1() functions target-i386: Rename KVM auto-feature-enable compat function target-i386: Disable CPUID_ACPI by default on KVM mode target-i386: Remove unsupported bits from all CPU models target-i386: Don't enable nested VMX by default target-i386: Disable SVM by default in KVM mode hw/i386/pc_piix.c | 22 ++ hw/i386/pc_q35.c | 18 -- target-i386/cpu.c | 42 -- target-i386/cpu.h | 3 ++- 4 files changed, 64 insertions(+), 21 deletions(-) -- 1.9.3 -- To unsubscribe from this list: send the line unsubscribe kvm in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCH v3 0/6] target-i386: Make most CPU models work with enforce out of the box
Il 03/10/2014 21:39, Eduardo Habkost ha scritto: Changes v2 - v3: * None. This is just a rebase against latest qemu.git master (commit b00a0dd) Changes v1 - v2: * Commit message and comment changes. * Update compat code to change pc-*-2.1, not pc-*-2.0. * Added patch to disable SVM by default in KVM mode. Most of the bits that make enforce breaks were introduced in 2010 by commit 8560efed6a72a816c0115f41ddb9d79f7ce63f28. The intention behind that commit made sense, the only problem is that we can't guarantee guest ABI stability across hosts if we simply rely on trimming of CPU features based on host capabilities. So, this series remove CPUID bits from the CPU model definitions so they become defaults that: 1) won't unexpectly stop working when we start using the enforce flag; 2) won't silently break the guest ABI when TCG or KVM start supporting new features. There's only one non-trivial case left: the qemu32/qemu64 models. The problem with them is that we have conflicting expectations about it, from different users: TCG users expect the default CPU model to contain most TCG-supported features (and it makes sense). See, for example, commit f1e00a9cf326acc1f2386a72525af8859852e1df. KVM users expect the default CPU model to be a conservative choice which will work on most host CPUs (and will only contain features that are supported by KVM). We could solve the qemu32/qemu64 issue by having different defaults for TCG and KVM. But we have existing management code (libvirt) that already expects qemu32 or qemu64 to be the default, and changing the default would break that code. I will send an RFC to address that later. Cc: Aurelien Jarno aurel...@aurel32.net Cc: Paolo Bonzini pbonz...@redhat.com Cc: kvm@vger.kernel.org Eduardo Habkost (6): pc: Create pc_compat_2_1() functions target-i386: Rename KVM auto-feature-enable compat function target-i386: Disable CPUID_ACPI by default on KVM mode target-i386: Remove unsupported bits from all CPU models target-i386: Don't enable nested VMX by default target-i386: Disable SVM by default in KVM mode hw/i386/pc_piix.c | 22 ++ hw/i386/pc_q35.c | 18 -- target-i386/cpu.c | 42 -- target-i386/cpu.h | 3 ++- 4 files changed, 64 insertions(+), 21 deletions(-) Andreas, are you picking up this? Paolo -- To unsubscribe from this list: send the line unsubscribe kvm in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
