From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from ozlabs.org (bilbo.ozlabs.org [203.11.71.1]) (using TLSv1.2 with cipher ADH-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by lists.ozlabs.org (Postfix) with ESMTPS id 40VTnt40gVzF1sD for ; Tue, 24 Apr 2018 13:48:34 +1000 (AEST) Date: Tue, 24 Apr 2018 13:48:25 +1000 From: David Gibson To: Sam Bobroff Cc: =?iso-8859-1?Q?C=E9dric?= Le Goater , linuxppc-dev@lists.ozlabs.org, kvm@vger.kernel.org, kvm-ppc@vger.kernel.org, paulus@samba.org Subject: Re: [PATCH RFC 1/1] KVM: PPC: Book3S HV: pack VCORE IDs to access full VCPU ID space Message-ID: <20180424034825.GN19804@umbus.fritz.box> References: <70974cfb62a7f09a53ec914d2909639884228244.1523516498.git.sam.bobroff@au1.ibm.com> <20180416040942.GB20551@umbus.fritz.box> <1e01ea66-6103-94c8-ccb1-ed35b3a3104b@kaod.org> <20180424031914.GA25846@tungsten.ozlabs.ibm.com> MIME-Version: 1.0 Content-Type: multipart/signed; micalg=pgp-sha256; protocol="application/pgp-signature"; boundary="qYrsQHciA3Wqs7Iv" In-Reply-To: <20180424031914.GA25846@tungsten.ozlabs.ibm.com> List-Id: Linux on PowerPC Developers Mail List List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , --qYrsQHciA3Wqs7Iv Content-Type: text/plain; charset=iso-8859-1 Content-Disposition: inline Content-Transfer-Encoding: quoted-printable On Tue, Apr 24, 2018 at 01:19:15PM +1000, Sam Bobroff wrote: > On Mon, Apr 23, 2018 at 11:06:35AM +0200, C=E9dric Le Goater wrote: > > On 04/16/2018 06:09 AM, David Gibson wrote: > > > On Thu, Apr 12, 2018 at 05:02:06PM +1000, Sam Bobroff wrote: > > >> It is not currently possible to create the full number of possible > > >> VCPUs (KVM_MAX_VCPUS) on Power9 with KVM-HV when the guest uses less > > >> threads per core than it's core stride (or "VSMT mode"). This is > > >> because the VCORE ID and XIVE offsets to grow beyond KVM_MAX_VCPUS > > >> even though the VCPU ID is less than KVM_MAX_VCPU_ID. > > >> > > >> To address this, "pack" the VCORE ID and XIVE offsets by using > > >> knowledge of the way the VCPU IDs will be used when there are less > > >> guest threads per core than the core stride. The primary thread of > > >> each core will always be used first. Then, if the guest uses more th= an > > >> one thread per core, these secondary threads will sequentially follow > > >> the primary in each core. > > >> > > >> So, the only way an ID above KVM_MAX_VCPUS can be seen, is if the > > >> VCPUs are being spaced apart, so at least half of each core is empty > > >> and IDs between KVM_MAX_VCPUS and (KVM_MAX_VCPUS * 2) can be mapped > > >> into the second half of each core (4..7, in an 8-thread core). > > >> > > >> Similarly, if IDs above KVM_MAX_VCPUS * 2 are seen, at least 3/4 of > > >> each core is being left empty, and we can map down into the second a= nd > > >> third quarters of each core (2, 3 and 5, 6 in an 8-thread core). > > >> > > >> Lastly, if IDs above KVM_MAX_VCPUS * 4 are seen, only the primary > > >> threads are being used and 7/8 of the core is empty, allowing use of > > >> the 1, 3, 5 and 7 thread slots. > > >> > > >> (Strides less than 8 are handled similarly.) > > >> > > >> This allows the VCORE ID or offset to be calculated quickly from the > > >> VCPU ID or XIVE server numbers, without access to the VCPU structure. > > >> > > >> Signed-off-by: Sam Bobroff > > >> --- > > >> Hello everyone, > > >> > > >> I've tested this on P8 and P9, in lots of combinations of host and g= uest > > >> threading modes and it has been fine but it does feel like a "tricky" > > >> approach, so I still feel somewhat wary about it. > >=20 > > Have you done any migration ?=20 >=20 > No, but I will :-) >=20 > > >> I've posted it as an RFC because I have not tested it with guest nat= ive-XIVE, > > >> and I suspect that it will take some work to support it. > >=20 > > The KVM XIVE device will be different for XIVE exploitation mode, same = structures=20 > > though. I will send a patchset shortly.=20 >=20 > Great. This is probably where conflicts between the host and guest > numbers will show up. (See dwg's question below.) >=20 > > >> arch/powerpc/include/asm/kvm_book3s.h | 19 +++++++++++++++++++ > > >> arch/powerpc/kvm/book3s_hv.c | 14 ++++++++++---- > > >> arch/powerpc/kvm/book3s_xive.c | 9 +++++++-- > > >> 3 files changed, 36 insertions(+), 6 deletions(-) > > >> > > >> diff --git a/arch/powerpc/include/asm/kvm_book3s.h b/arch/powerpc/in= clude/asm/kvm_book3s.h > > >> index 376ae803b69c..1295056d564a 100644 > > >> --- a/arch/powerpc/include/asm/kvm_book3s.h > > >> +++ b/arch/powerpc/include/asm/kvm_book3s.h > > >> @@ -368,4 +368,23 @@ extern int kvmppc_h_logical_ci_store(struct kvm= _vcpu *vcpu); > > >> #define SPLIT_HACK_MASK 0xff000000 > > >> #define SPLIT_HACK_OFFS 0xfb000000 > > >> =20 > > >> +/* Pack a VCPU ID from the [0..KVM_MAX_VCPU_ID) space down to the > > >> + * [0..KVM_MAX_VCPUS) space, while using knowledge of the guest's c= ore stride > > >> + * (but not it's actual threading mode, which is not available) to = avoid > > >> + * collisions. > > >> + */ > > >> +static inline u32 kvmppc_pack_vcpu_id(struct kvm *kvm, u32 id) > > >> +{ > > >> + const int block_offsets[MAX_SMT_THREADS] =3D {0, 4, 2, 6, 1, 5, 3,= 7}; > > >=20 > > > I'd suggest 1,3,5,7 at the end rather than 1,5,3,7 - accomplishes > > > roughly the same thing, but I think makes the pattern more obvious. >=20 > OK. >=20 > > >> + int stride =3D kvm->arch.emul_smt_mode > 1 ? > > >> + kvm->arch.emul_smt_mode : kvm->arch.smt_mode; > > >=20 > > > AFAICT from BUG_ON()s etc. at the callsites, kvm->arch.smt_mode must > > > always be 1 when this is called, so the conditional here doesn't seem > > > useful. >=20 > Ah yes, right. (That was an older version when I was thinking of using > it for P8 as well but that didn't seem to be a good idea.) >=20 > > >> + int block =3D (id / KVM_MAX_VCPUS) * (MAX_SMT_THREADS / stride); > > >> + u32 packed_id; > > >> + > > >> + BUG_ON(block >=3D MAX_SMT_THREADS); > > >> + packed_id =3D (id % KVM_MAX_VCPUS) + block_offsets[block]; > > >> + BUG_ON(packed_id >=3D KVM_MAX_VCPUS); > > >> + return packed_id; > > >> +} > > >=20 > > > It took me a while to wrap my head around the packing function, but I > > > think I got there in the end. It's pretty clever. >=20 > Thanks, I'll try to add a better description as well :-) >=20 > > > One thing bothers me, though. This certainly packs things under > > > KVM_MAX_VCPUS, but not necessarily under the actual number of vcpus. > > > e.g. KVM_MAC_VCPUS=3D=3D16, 8 vcpus total, stride 8, 2 vthreads/vcore= (as > > > qemu sees it), gives both unpacked IDs (0, 1, 8, 9, 16, 17, 24, 25) > > > and packed ids of (0, 1, 8, 9, 4, 5, 12, 13) - leaving 2, 3, 6, 7 > > > etc. unused. >=20 > That's right. The property it provides is that all the numbers are under > KVM_MAX_VCPUS (which, see below, is the size of the fixed areas) not > that they are sequential. >=20 > > > So again, the question is what exactly are these remapped IDs useful > > > for. If we're indexing into a bare array of structures of size > > > KVM_MAX_VCPUS then we're *already* wasting a bunch of space by having > > > more entries than vcpus. If we're indexing into something sparser, > > > then why is the remapping worthwhile? >=20 > Well, here's my thinking: >=20 > At the moment, kvm->vcores[] and xive->vp_base are both sized by NR_CPUS > (via KVM_MAX_VCPUS and KVM_MAX_VCORES which are both NR_CPUS). This is > enough space for the maximum number of VCPUs, and some space is wasted > when the guest uses less than this (but KVM doesn't know how many will > be created, so we can't do better easily). The problem is that the > indicies overflow before all of those VCPUs can be created, not that > more space is needed. >=20 > We could fix the overflow by expanding these areas to KVM_MAX_VCPU_ID > but that will use 8x the space we use now, and we know that no more than > KVM_MAX_VCPUS will be used so all this new space is basically wasted. >=20 > So remapping seems better if it will work. (Ben H. was strongly against > wasting more XIVE space if possible.) Hm, ok. Are the relevant arrays here per-VM, or global? Or some of both? > In short, remapping provides a way to allow the guest to create it's full= set > of VCPUs without wasting any more space than we do currently, without > having to do something more complicated like tracking used IDs or adding > additional KVM CAPs. >=20 > > >> + > > >> #endif /* __ASM_KVM_BOOK3S_H__ */ > > >> diff --git a/arch/powerpc/kvm/book3s_hv.c b/arch/powerpc/kvm/book3s_= hv.c > > >> index 9cb9448163c4..49165cc90051 100644 > > >> --- a/arch/powerpc/kvm/book3s_hv.c > > >> +++ b/arch/powerpc/kvm/book3s_hv.c > > >> @@ -1762,7 +1762,7 @@ static int threads_per_vcore(struct kvm *kvm) > > >> return threads_per_subcore; > > >> } > > >> =20 > > >> -static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, in= t core) > > >> +static struct kvmppc_vcore *kvmppc_vcore_create(struct kvm *kvm, in= t id) > > >> { > > >> struct kvmppc_vcore *vcore; > > >> =20 > > >> @@ -1776,7 +1776,7 @@ static struct kvmppc_vcore *kvmppc_vcore_creat= e(struct kvm *kvm, int core) > > >> init_swait_queue_head(&vcore->wq); > > >> vcore->preempt_tb =3D TB_NIL; > > >> vcore->lpcr =3D kvm->arch.lpcr; > > >> - vcore->first_vcpuid =3D core * kvm->arch.smt_mode; > > >> + vcore->first_vcpuid =3D id; > > >> vcore->kvm =3D kvm; > > >> INIT_LIST_HEAD(&vcore->preempt_list); > > >> =20 > > >> @@ -1992,12 +1992,18 @@ static struct kvm_vcpu *kvmppc_core_vcpu_cre= ate_hv(struct kvm *kvm, > > >> mutex_lock(&kvm->lock); > > >> vcore =3D NULL; > > >> err =3D -EINVAL; > > >> - core =3D id / kvm->arch.smt_mode; > > >> + if (cpu_has_feature(CPU_FTR_ARCH_300)) { > > >> + BUG_ON(kvm->arch.smt_mode !=3D 1); > > >> + core =3D kvmppc_pack_vcpu_id(kvm, id); > > >> + } else { > > >> + core =3D id / kvm->arch.smt_mode; > > >> + } > > >> if (core < KVM_MAX_VCORES) { > > >> vcore =3D kvm->arch.vcores[core]; > > >> + BUG_ON(cpu_has_feature(CPU_FTR_ARCH_300) && vcore); > > >> if (!vcore) { > > >> err =3D -ENOMEM; > > >> - vcore =3D kvmppc_vcore_create(kvm, core); > > >> + vcore =3D kvmppc_vcore_create(kvm, id & ~(kvm->arch.smt_mode - 1= )); > > >> kvm->arch.vcores[core] =3D vcore; > > >> kvm->arch.online_vcores++; > > >> } > > >> diff --git a/arch/powerpc/kvm/book3s_xive.c b/arch/powerpc/kvm/book3= s_xive.c > > >> index f9818d7d3381..681dfe12a5f3 100644 > > >> --- a/arch/powerpc/kvm/book3s_xive.c > > >> +++ b/arch/powerpc/kvm/book3s_xive.c > > >> @@ -317,6 +317,11 @@ static int xive_select_target(struct kvm *kvm, = u32 *server, u8 prio) > > >> return -EBUSY; > > >> } > > >> =20 > > >> +static u32 xive_vp(struct kvmppc_xive *xive, u32 server) > > >> +{ > > >> + return xive->vp_base + kvmppc_pack_vcpu_id(xive->kvm, server); > > >> +} > > >> + > > >=20 > > > I'm finding the XIVE indexing really baffling. There are a bunch of > > > other places where the code uses (xive->vp_base + NUMBER) directly. >=20 > Ugh, yes. It looks like I botched part of my final cleanup and all the > cases you saw in kvm/book3s_xive.c should have been replaced with a call = to > xive_vp(). I'll fix it and sorry for the confusion. Ok. > > This links the QEMU vCPU server NUMBER to a XIVE virtual processor numb= er=20 > > in OPAL. So we need to check that all used NUMBERs are, first, consiste= nt=20 > > and then, in the correct range. >=20 > Right. My approach was to allow XIVE to keep using server numbers that > are equal to VCPU IDs, and just pack down the ID before indexing into > the vp_base area. >=20 > > > If those are host side references, I guess they don't need updates for > > > this. >=20 > These are all guest side references. >=20 > > > But if that's the case, then how does indexing into the same array > > > with both host and guest server numbers make sense? >=20 > Right, it doesn't make sense to mix host and guest server numbers when > we're remapping only the guest ones, but in this case (without native > guest XIVE support) it's just guest ones. Right. Will this remapping be broken by guest-visible XIVE? That is for the guest visible XIVE are we going to need to expose un-remapped XIVE server IDs to the guest? > > yes. VPs are allocated with KVM_MAX_VCPUS : > >=20 > > xive->vp_base =3D xive_native_alloc_vp_block(KVM_MAX_VCPUS); > >=20 > > but > >=20 > > #define KVM_MAX_VCPU_ID (threads_per_subcore * KVM_MAX_VCORES) > >=20 > > WE would need to change the allocation of the VPs I guess. >=20 > Yes, this is one of the structures that overflow if we don't pack the IDs. >=20 > > >> static u8 xive_lock_and_mask(struct kvmppc_xive *xive, > > >> struct kvmppc_xive_src_block *sb, > > >> struct kvmppc_xive_irq_state *state) > > >> @@ -1084,7 +1089,7 @@ int kvmppc_xive_connect_vcpu(struct kvm_device= *dev, > > >> pr_devel("Duplicate !\n"); > > >> return -EEXIST; > > >> } > > >> - if (cpu >=3D KVM_MAX_VCPUS) { > > >> + if (cpu >=3D KVM_MAX_VCPU_ID) {>> > > >> pr_devel("Out of bounds !\n"); > > >> return -EINVAL; > > >> } > > >> @@ -1098,7 +1103,7 @@ int kvmppc_xive_connect_vcpu(struct kvm_device= *dev, > > >> xc->xive =3D xive; > > >> xc->vcpu =3D vcpu; > > >> xc->server_num =3D cpu; > > >> - xc->vp_id =3D xive->vp_base + cpu; > > >> + xc->vp_id =3D xive_vp(xive, cpu); > > >> xc->mfrr =3D 0xff; > > >> xc->valid =3D true; > > >> =20 > > >=20 > >=20 --=20 David Gibson | I'll have my music baroque, and my code david AT gibson.dropbear.id.au | minimalist, thank you. NOT _the_ _other_ | _way_ _around_! http://www.ozlabs.org/~dgibson --qYrsQHciA3Wqs7Iv Content-Type: application/pgp-signature; name="signature.asc" -----BEGIN PGP SIGNATURE----- iQIzBAEBCAAdFiEEdfRlhq5hpmzETofcbDjKyiDZs5IFAlreqQcACgkQbDjKyiDZ s5KwBg/+LjFqaZXkcb/TLl74AcvCOmpb1O4CU9fVN5NXoKgLVCRCkHY5gl3SCScZ uZuvEEVz86n6ILFWzFBReZUlmiRjpfurRJUoABDaLBKF8Wd1nGwcbMq+w+MJGDH6 b6HQoO18xwEGuO2v1i38qVoar0ev+jcllnPTeIyQmiQgSvUB7gA/pLi9o0oLiDoO czEp9he/iWdcj3WTPu8lX2OG4ZCSpvwFBg+47szULCV8pvVaWksg8CnavxJ139w8 PCYD5n5VGKPm92epL1A6E3pKwcAqAgIgIke7JXn7PWGIWHoF+Y32rvjwB3d1bdu9 a1iPjAiuDxuWLxqsApzMGpIiaVpl0/LTB79RFJWuy0nJ2pWF7+ZOT6XV0vO6RRtJ iuG7SeGYIjoQrn/7iW1P0Imasiocqef12GN7uSxj6SW8perWemdlsr2MrXSZknR3 1gcjYFD128nudBRbb6xKGVFrm4rz+7oF9WAbJOUPxiwMm+h+rU1LS/Pa6y0RDjxy EJLlnBNqNDlTjOepsROm2maJmoij0BYYmpIyZs3zQJDmSxQVkK8IahgyxTO5M3u6 oGiY6Bg4TN8i6HFIvQdrQYTcmYuwLb4X9hXWx9BDCyJ3iuwfJc+6Rhj5JGxnfScH NIo/W/Z6daE9z7m9TvcelbDGshuLUOWa/84Esf7h+E+VDxTK1Zo= =x8Y9 -----END PGP SIGNATURE----- --qYrsQHciA3Wqs7Iv--