linux-kernel - Re: [PATCH v2 2/3] kvm/arm: Introduce a new vmid allocator for KVM

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20210721160614.GC11003@willie-the-truck>
Date:   Wed, 21 Jul 2021 17:06:14 +0100
From:   Will Deacon <will@...nel.org>
To:     Shameer Kolothum <shameerali.kolothum.thodi@...wei.com>
Cc:     linux-arm-kernel@...ts.infradead.org, kvmarm@...ts.cs.columbia.edu,
        linux-kernel@...r.kernel.org, maz@...nel.org,
        catalin.marinas@....com, james.morse@....com,
        julien.thierry.kdev@...il.com, suzuki.poulose@....com,
        jean-philippe@...aro.org, Alexandru.Elisei@....com,
        linuxarm@...wei.com
Subject: Re: [PATCH v2 2/3] kvm/arm: Introduce a new vmid allocator for KVM

On Wed, Jun 16, 2021 at 04:56:05PM +0100, Shameer Kolothum wrote:
> A new VMID allocator for arm64 KVM use. This is based on
> arm64 asid allocator algorithm.
> 
> Signed-off-by: Shameer Kolothum <shameerali.kolothum.thodi@...wei.com>
> ---
>  arch/arm64/include/asm/kvm_host.h |   4 +
>  arch/arm64/kvm/vmid.c             | 206 ++++++++++++++++++++++++++++++
>  2 files changed, 210 insertions(+)
>  create mode 100644 arch/arm64/kvm/vmid.c

Generally, I prefer this to the alternative of creating a library. However,
I'd probably remove all the duplicated comments in favour of a reference
to the ASID allocator. That way, we can just comment any VMID-specific
behaviour in here.

Some comments below...

> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> index 7cd7d5c8c4bc..75a7e8071012 100644
> --- a/arch/arm64/include/asm/kvm_host.h
> +++ b/arch/arm64/include/asm/kvm_host.h
> @@ -680,6 +680,10 @@ int kvm_arm_pvtime_get_attr(struct kvm_vcpu *vcpu,
>  int kvm_arm_pvtime_has_attr(struct kvm_vcpu *vcpu,
>  			    struct kvm_device_attr *attr);
>  
> +int kvm_arm_vmid_alloc_init(void);
> +void kvm_arm_vmid_alloc_free(void);
> +void kvm_arm_update_vmid(atomic64_t *id);
> +
>  static inline void kvm_arm_pvtime_vcpu_init(struct kvm_vcpu_arch *vcpu_arch)
>  {
>  	vcpu_arch->steal.base = GPA_INVALID;
> diff --git a/arch/arm64/kvm/vmid.c b/arch/arm64/kvm/vmid.c
> new file mode 100644
> index 000000000000..687e18d33130
> --- /dev/null
> +++ b/arch/arm64/kvm/vmid.c
> @@ -0,0 +1,206 @@
> +// SPDX-License-Identifier: GPL-2.0
> +/*
> + * VMID allocator.
> + *
> + * Based on arch/arm64/mm/context.c
> + *
> + * Copyright (C) 2002-2003 Deep Blue Solutions Ltd, all rights reserved.
> + * Copyright (C) 2012 ARM Ltd.
> + */
> +
> +#include <linux/bitfield.h>
> +#include <linux/bitops.h>
> +
> +#include <asm/kvm_asm.h>
> +#include <asm/kvm_mmu.h>
> +
> +static u32 vmid_bits;
> +static DEFINE_RAW_SPINLOCK(cpu_vmid_lock);
> +
> +static atomic64_t vmid_generation;
> +static unsigned long *vmid_map;
> +
> +static DEFINE_PER_CPU(atomic64_t, active_vmids);
> +static DEFINE_PER_CPU(u64, reserved_vmids);
> +static cpumask_t tlb_flush_pending;
> +
> +#define VMID_MASK		(~GENMASK(vmid_bits - 1, 0))
> +#define VMID_FIRST_VERSION	(1UL << vmid_bits)
> +
> +#define NUM_USER_VMIDS		VMID_FIRST_VERSION
> +#define vmid2idx(vmid)		((vmid) & ~VMID_MASK)
> +#define idx2vmid(idx)		vmid2idx(idx)
> +
> +#define vmid_gen_match(vmid) \
> +	(!(((vmid) ^ atomic64_read(&vmid_generation)) >> vmid_bits))
> +
> +static void flush_context(void)
> +{
> +	int cpu;
> +	u64 vmid;
> +
> +	bitmap_clear(vmid_map, 0, NUM_USER_VMIDS);
> +
> +	for_each_possible_cpu(cpu) {
> +		vmid = atomic64_xchg_relaxed(&per_cpu(active_vmids, cpu), 0);
> +		/*
> +		 * If this CPU has already been through a
> +		 * rollover, but hasn't run another task in
> +		 * the meantime, we must preserve its reserved
> +		 * VMID, as this is the only trace we have of
> +		 * the process it is still running.
> +		 */
> +		if (vmid == 0)
> +			vmid = per_cpu(reserved_vmids, cpu);
> +		__set_bit(vmid2idx(vmid), vmid_map);
> +		per_cpu(reserved_vmids, cpu) = vmid;
> +	}

Hmm, so here we're copying the active_vmids into the reserved_vmids on a
rollover, but I wonder if that's overly pessismistic? For the ASID
allocator, every CPU tends to have a current task so it makes sense, but
I'm not sure it's necessarily the case that every CPU tends to have a
vCPU as the current task. For example, imagine you have a nasty 128-CPU
system with 8-bit VMIDs and each CPU has at some point run a vCPU. Then,
on rollover, we'll immediately reserve half of the VMID space, even if
those vCPUs don't even exist any more.

Not sure if it's worth worrying about, but I wanted to mention it.

> +void kvm_arm_update_vmid(atomic64_t *id)
> +{

Take the kvm_vmid here? That would make:

> +	/* Check that our VMID belongs to the current generation. */
> +	vmid = atomic64_read(id);
> +	if (!vmid_gen_match(vmid)) {
> +		vmid = new_vmid(id);
> +		atomic64_set(id, vmid);
> +	}

A bit more readable, as you could pass the pointer directly to new_vmid
for initialisation.

Will