"value":"In the Linux kernel, the following vulnerability has been resolved:\n\narm64: entry: always set GIC_PRIO_PSR_I_SET during entry\n\nZenghui reports that booting a kernel with \"irqchip.gicv3_pseudo_nmi=1\"\non the command line hits a warning during kernel entry, due to the way\nwe manipulate the PMR.\n\nEarly in the entry sequence, we call lockdep_hardirqs_off() to inform\nlockdep that interrupts have been masked (as the HW sets DAIF wqhen\nentering an exception). Architecturally PMR_EL1 is not affected by\nexception entry, and we don't set GIC_PRIO_PSR_I_SET in the PMR early in\nthe exception entry sequence, so early in exception entry the PMR can\nindicate that interrupts are unmasked even though they are masked by\nDAIF.\n\nIf DEBUG_LOCKDEP is selected, lockdep_hardirqs_off() will check that\ninterrupts are masked, before we set GIC_PRIO_PSR_I_SET in any of the\nexception entry paths, and hence lockdep_hardirqs_off() will WARN() that\nsomething is amiss.\n\nWe can avoid this by consistently setting GIC_PRIO_PSR_I_SET during\nexception entry so that kernel code sees a consistent environment. We\nmust also update local_daif_inherit() to undo this, as currently only\ntouches DAIF. For other paths, local_daif_restore() will update both\nDAIF and the PMR. With this done, we can remove the existing special\ncases which set this later in the entry code.\n\nWe always use (GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET) for consistency with\nlocal_daif_save(), as this will warn if it ever encounters\n(GIC_PRIO_IRQOFF | GIC_PRIO_PSR_I_SET), and never sets this itself. This\nmatches the gic_prio_kentry_setup that we have to retain for\nret_to_user.\n\nThe original splat from Zenghui's report was:\n\n| DEBUG_LOCKS_WARN_ON(!irqs_disabled())\n| WARNING: CPU: 3 PID: 125 at kernel/locking/lockdep.c:4258 lockdep_hardirqs_off+0xd4/0xe8\n| Modules linked in:\n| CPU: 3 PID: 125 Comm: modprobe Tainted: G W 5.12.0-rc8+ #463\n| Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015\n| pstate: 604003c5 (nZCv DAIF +PAN -UAO -TCO BTYPE=--)\n| pc : lockdep_hardirqs_off+0xd4/0xe8\n| lr : lockdep_hardirqs_off+0xd4/0xe8\n| sp : ffff80002a39bad0\n| pmr_save: 000000e0\n| x29: ffff80002a39bad0 x28: ffff0000de214bc0\n| x27: ffff0000de1c0400 x26: 000000000049b328\n| x25: 0000000000406f30 x24: ffff0000de1c00a0\n| x23: 0000000020400005 x22: ffff8000105f747c\n| x21: 0000000096000044 x20: 0000000000498ef9\n| x19: ffff80002a39bc88 x18: ffffffffffffffff\n| x17: 0000000000000000 x16: ffff800011c61eb0\n| x15: ffff800011700a88 x14: 0720072007200720\n| x13: 0720072007200720 x12: 0720072007200720\n| x11: 0720072007200720 x10: 0720072007200720\n| x9 : ffff80002a39bad0 x8 : ffff80002a39bad0\n| x7 : ffff8000119f0800 x6 : c0000000ffff7fff\n| x5 : ffff8000119f07a8 x4 : 0000000000000001\n| x3 : 9bcdab23f2432800 x2 : ffff800011730538\n| x1 : 9bcdab23f2432800 x0 : 0000000000000000\n| Call trace:\n| lockdep_hardirqs_off+0xd4/0xe8\n| enter_from_kernel_mode.isra.5+0x7c/0xa8\n| el1_abort+0x24/0x100\n| el1_sync_handler+0x80/0xd0\n| el1_sync+0x6c/0x100\n| __arch_clear_user+0xc/0x90\n| load_elf_binary+0x9fc/0x1450\n| bprm_execve+0x404/0x880\n| kernel_execve+0x180/0x188\n| call_usermodehelper_exec_async+0xdc/0x158\n| ret_from_fork+0x10/0x18"