)]}'
{
  "commit": "ef56c5ef358f3385174c059f49a39196ea68444b",
  "tree": "db84de1d669837fb015eae7bc0e0ddfe39d3e91c",
  "parents": [
    "905dd8f431e84572b7f1a1f9ad95581c3e48bf66"
  ],
  "author": {
    "name": "Fuad Tabba",
    "email": "fuad.tabba@linux.dev",
    "time": "Tue Jul 14 16:44:26 2026 +0100"
  },
  "committer": {
    "name": "Fuad Tabba",
    "email": "fuad.tabba@linux.dev",
    "time": "Fri Aug 21 07:54:52 2026 +0100"
  },
  "message": "arm64: Query steal-time support on the VM fd\n\nCommit 1688c26f837b (\"arm64: Query per-VM capabilities when selecting\nvCPU features\") moved kvm_cpu__select_features()\u0027s feature probes\n(SVE, PMU, EL1_32BIT, PTRAUTH) from the system fd to the VM fd,\nbecause vCPU feature availability is a per-VM property that the\nsystem fd cannot reflect. kvm_cpu__setup_pvtime() still probes\nKVM_CAP_STEAL_TIME with kvm__supports_extension() on the system fd;\nsteal-time is the same kind of per-VM capability.\n\nThe gap is user visible under pKVM: the system fd tells kvmtool\nsteal-time is available even for a protected VM, so kvmtool proceeds\nto KVM_HAS_DEVICE_ATTR, hits the kernel\u0027s -EPERM for that VM class,\nand dies in kvm_cpu__arch_init():\n\n  Fatal: Unable to initialise vcpu\n\nSwitch to kvm__supports_vm_extension(), completing the VM-fd\nconversion for the one per-VM capability probe it missed. kvmtool\nthen correctly takes its existing no_pvtime skip path when the VM fd\nreports no support.\n\nSigned-off-by: Fuad Tabba \u003ctabba@google.com\u003e\n",
  "tree_diff": [
    {
      "type": "modify",
      "old_id": "2933ac7ca959df43a670d20168f2543c5e7ce6ca",
      "old_mode": 33188,
      "old_path": "arm64/pvtime.c",
      "new_id": "839aa8a7532c4ac6b8452e84e3212978bc5b56cb",
      "new_mode": 33188,
      "new_path": "arm64/pvtime.c"
    }
  ]
}
