From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: X-Spam-Checker-Version: SpamAssassin 3.4.0 (2014-02-07) on aws-us-west-2-korg-lkml-1.web.codeaurora.org X-Spam-Level: X-Spam-Status: No, score=-7.3 required=3.0 tests=BAYES_00,DKIM_SIGNED, DKIM_VALID,DKIM_VALID_AU,HEADER_FROM_DIFFERENT_DOMAINS,MAILING_LIST_MULTI, SPF_HELO_NONE,SPF_PASS,USER_AGENT_SANE_2 autolearn=no autolearn_force=no version=3.4.0 Received: from mail.kernel.org (mail.kernel.org [198.145.29.99]) by smtp.lore.kernel.org (Postfix) with ESMTP id F065FC11F64 for ; Thu, 1 Jul 2021 17:39:08 +0000 (UTC) Received: from vger.kernel.org (vger.kernel.org [23.128.96.18]) by mail.kernel.org (Postfix) with ESMTP id C5B256044F for ; Thu, 1 Jul 2021 17:39:08 +0000 (UTC) Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S232593AbhGARle (ORCPT ); Thu, 1 Jul 2021 13:41:34 -0400 Received: from lindbergh.monkeyblade.net ([23.128.96.19]:47722 "EHLO lindbergh.monkeyblade.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S229978AbhGARle (ORCPT ); Thu, 1 Jul 2021 13:41:34 -0400 Received: from bombadil.infradead.org (bombadil.infradead.org [IPv6:2607:7c80:54:e::133]) by lindbergh.monkeyblade.net (Postfix) with ESMTPS id D0174C061762 for ; Thu, 1 Jul 2021 10:39:03 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; q=dns/txt; c=relaxed/relaxed; d=infradead.org; s=bombadil.20210309; h=Mime-Version:Content-Type:References: In-Reply-To:Date:Cc:To:From:Subject:Message-ID:Sender:Reply-To: Content-Transfer-Encoding:Content-ID:Content-Description; bh=wutHQnSFuSsbpWDjhp5LbXwI1xk6Pmj5T8hInVm+r8E=; b=eU4Tr9VYJv79RXyPaw+Vby2lKS GlcGCxDZGJ4rqImZbxCQrbC/7urrb4uy5D7jyNqq6iuNrlh/4kalNf7YxT61Qy2oCjoCE95ny0ErX D0m3CY1+jQrUDvUgOBIll68Z/9QqzYqrN23gyHkNyyuxNgbtdAc2cemDhkfcs79hCk6b5jDLqjoR7 TtunRpzUMP544Bn7Byt+B4vTB6U2mc4WxnVSOBcyo/0t70rpNmeoRFb9ri+z4R0wnXybWRN8qAWhk wknMI3pOcd2oG5D9llLYjSI8SnFRZ1HD6bDDbxHaBHbTM/vSSn9cDAsjwrZwYPw5GliNpHuWM+c9Z enrOluGA==; Received: from 54-240-197-232.amazon.com ([54.240.197.232] helo=freeip.amazon.com) by bombadil.infradead.org with esmtpsa (Exim 4.94.2 #2 (Red Hat Linux)) id 1lz0eV-000dwP-2U; Thu, 01 Jul 2021 17:39:03 +0000 Message-ID: <1d5b8251e8d9e67613295d5b7f51c49c1ee8c0a8.camel@infradead.org> Subject: Re: [PATCH v3 3/5] vhost_net: remove virtio_net_hdr validation, let tun/tap do it themselves From: David Woodhouse To: Jason Wang , netdev@vger.kernel.org Cc: Eugenio =?ISO-8859-1?Q?P=E9rez?= , Willem de Bruijn , "Michael S.Tsirkin" Date: Thu, 01 Jul 2021 18:39:00 +0100 In-Reply-To: References: <03ee62602dd7b7101f78e0802249a6e2e4c10b7f.camel@infradead.org> <20210624123005.1301761-1-dwmw2@infradead.org> <20210624123005.1301761-3-dwmw2@infradead.org> <1c6110d9-2a45-f766-9d9a-e2996c14b748@redhat.com> <72dfecd426d183615c0dd4c2e68690b0e95dd739.camel@infradead.org> <80f61c54a2b39cb129e8606f843f7ace605d67e0.camel@infradead.org> <99496947-8171-d252-66d3-0af12c62fd2c@redhat.com> <500370cc-d030-6c2d-8e96-533a3533a8e2@redhat.com> Content-Type: multipart/signed; micalg="sha-256"; protocol="application/x-pkcs7-signature"; boundary="=-CAlq5QINaohkmriR4npb" X-Mailer: Evolution 3.28.5-0ubuntu0.18.04.2 Mime-Version: 1.0 X-SRS-Rewrite: SMTP reverse-path rewritten from by bombadil.infradead.org. See http://www.infradead.org/rpr.html Precedence: bulk List-ID: X-Mailing-List: netdev@vger.kernel.org --=-CAlq5QINaohkmriR4npb Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable On Thu, 2021-07-01 at 12:13 +0800, Jason Wang wrote: > =E5=9C=A8 2021/6/30 =E4=B8=8B=E5=8D=886:02, David Woodhouse =E5=86=99=E9= =81=93: > > On Wed, 2021-06-30 at 12:39 +0800, Jason Wang wrote: > > > =E5=9C=A8 2021/6/29 =E4=B8=8B=E5=8D=886:49, David Woodhouse =E5=86=99= =E9=81=93: > > > > So as I expected, the throughput is better with vhost-net once I ge= t to > > > > the point of 100% CPU usage in my main thread, because it offloads = the > > > > kernel=E2=86=90=E2=86=92user copies. But latency is somewhat worse. > > > >=20 > > > > I'm still using select() instead of epoll() which would give me a > > > > little back =E2=80=94 but only a little, as I only poll on 3-4 fds,= and more to > > > > the point it'll give me just as much win in the non-vhost case too,= so > > > > it won't make much difference to the vhost vs. non-vhost comparison= . > > > >=20 > > > > Perhaps I really should look into that trick of "if the vhost TX ri= ng > > > > is already stopped and would need a kick, and I only have a few pac= kets > > > > in the batch, just write them directly to /dev/net/tun". > > >=20 > > > That should work on low throughput. > >=20 > > Indeed it works remarkably well, as I noted in my follow-up. I also > > fixed a minor stupidity where I was reading from the 'call' eventfd > > *before* doing the real work of moving packets around. And that gives > > me a few tens of microseconds back too. > >=20 > > > > I'm wondering how that optimisation would translate to actual guest= s, > > > > which presumably have the same problem. Perhaps it would be an > > > > operation on the vhost fd, which ends up processing the ring right > > > > there in the context of *that* process instead of doing a wakeup? > > >=20 > > > It might improve the latency in an ideal case but several possible is= sues: > > >=20 > > > 1) this will blocks vCPU running until the sent is done > > > 2) copy_from_user() may sleep which will block the vcpu thread furthe= r > >=20 > > Yes, it would block the vCPU for a short period of time, but we could > > limit that. The real win is to improve latency of single, short packets > > like a first SYN, or SYNACK. It should work fine even if it's limited > > to *one* *short* packet which *is* resident in memory. >=20 >=20 > This looks tricky since we need to poke both virtqueue metadata as well= =20 > as the payload. That's OK as we'd *only* do it if the thread is quiescent anyway. > And we need to let the packet iterate the network stack which might have= =20 > extra latency (qdiscs, eBPF, switch/OVS). >=20 > So it looks to me it's better to use vhost_net busy polling instead=20 > (VHOST_SET_VRING_BUSYLOOP_TIMEOUT). Or something very similar, with the *trylock* and bailing out. > Userspace can detect this feature by validating the existence of the ioct= l. Yep. Or if we want to get fancy, we could even offer it to the guest. As a *different* doorbell register to poke if they want to relinquish the physical CPU to process the packet quicker. We wouldn't even *need* to go through userspace at all, if we put that into a separate page... but that probably *is* overengineering it :) > > Although actually I'm not *overly* worried about the 'resident' part. > > For a transmit packet, especially a short one not a sendpage(), it's > > fairly likely the the guest has touched the buffer right before sending > > it. And taken the hit of faulting it in then, if necessary. If the host > > is paging out memory which is *active* use by a guest, that guest is > > screwed anyway :) >=20 >=20 > Right, but there could be workload that is unrelated to the networking.= =20 > Block vCPU thread in this case seems sub-optimal. >=20 Right, but the VMM (or the guest, if we're letting the guest choose) wouldn't have to use it for those cases. > > Alternatively, there's still the memory map thing I need to fix before > > I can commit this in my application: > >=20 > > #ifdef __x86_64__ > > vmem->regions[0].guest_phys_addr =3D 4096; > > vmem->regions[0].memory_size =3D 0x7fffffffe000; > > vmem->regions[0].userspace_addr =3D 4096; > > #else > > #error FIXME > > #endif > > if (ioctl(vpninfo->vhost_fd, VHOST_SET_MEM_TABLE, vmem) < 0) { > >=20 > > Perhaps if we end up with a user-visible feature to deal with that, > > then I could use the presence of *that* feature to infer that the tun > > bugs are fixed. >=20 >=20 > As we discussed before it could be a new backend feature. VHOST_NET_SVA= =20 > (shared virtual address)? Yeah, I'll take a look. > > Another random thought as I stare at this... can't we handle checksums > > in tun_get_user() / tun_put_user()? We could always set NETIF_F_HW_CSUM > > on the tun device, and just do it *while* we're copying the packet to > > userspace, if userspace doesn't support it. That would be better than > > having the kernel complete the checksum in a separate pass *before* > > handing the skb to tun_net_xmit(). >=20 >=20 > I'm not sure I get this, but for performance reason we don't do any csum= =20 > in this case? I think we have to; the packets can't leave the box without a valid checksum. If the skb isn't CHECKSUM_COMPLETE at the time it's handed off to the ->hard_start_xmit of a netdev which doesn't advertise hardware checksum support, the network stack will do it manually in an extra pass. Which is kind of silly if the tun device is going to do a pass over all the data *anyway* as it copies it up to userspace. Even in the normal case without vhost-net. > > We could similarly do a partial checksum in tun_get_user() and hand it > > off to the network stack with ->ip_summed =3D=3D CHECKSUM_PARTIAL. >=20 >=20 > I think that's is how it is expected to work (via vnet header), see=20 > virtio_net_hdr_to_skb(). But only if the "guest" supports it; it doesn't get handled by the tun device. It *could*, since we already have the helpers to checksum *as* we copy to/from userspace. It doesn't help for me to advertise that I support TX checksums in userspace because I'd have to do an extra pass for that. I only do one pass over the data as I encrypt it, and in many block cipher modes the encryption of the early blocks affects the IV for the subsequent blocks... do I can't just go back and "fix" the checksum at the start of the packet, once I'm finished. So doing the checksum as the packet is copied up to userspace would be very useful. --=-CAlq5QINaohkmriR4npb Content-Type: application/x-pkcs7-signature; name="smime.p7s" Content-Disposition: attachment; filename="smime.p7s" Content-Transfer-Encoding: base64 MIAGCSqGSIb3DQEHAqCAMIACAQExDzANBglghkgBZQMEAgEFADCABgkqhkiG9w0BBwEAAKCCECow ggUcMIIEBKADAgECAhEA4rtJSHkq7AnpxKUY8ZlYZjANBgkqhkiG9w0BAQsFADCBlzELMAkGA1UE BhMCR0IxGzAZBgNVBAgTEkdyZWF0ZXIgTWFuY2hlc3RlcjEQMA4GA1UEBxMHU2FsZm9yZDEaMBgG A1UEChMRQ09NT0RPIENBIExpbWl0ZWQxPTA7BgNVBAMTNENPTU9ETyBSU0EgQ2xpZW50IEF1dGhl bnRpY2F0aW9uIGFuZCBTZWN1cmUgRW1haWwgQ0EwHhcNMTkwMTAyMDAwMDAwWhcNMjIwMTAxMjM1 OTU5WjAkMSIwIAYJKoZIhvcNAQkBFhNkd213MkBpbmZyYWRlYWQub3JnMIIBIjANBgkqhkiG9w0B AQEFAAOCAQ8AMIIBCgKCAQEAsv3wObLTCbUA7GJqKj9vHGf+Fa+tpkO+ZRVve9EpNsMsfXhvFpb8 RgL8vD+L133wK6csYoDU7zKiAo92FMUWaY1Hy6HqvVr9oevfTV3xhB5rQO1RHJoAfkvhy+wpjo7Q cXuzkOpibq2YurVStHAiGqAOMGMXhcVGqPuGhcVcVzVUjsvEzAV9Po9K2rpZ52FE4rDkpDK1pBK+ uOAyOkgIg/cD8Kugav5tyapydeWMZRJQH1vMQ6OVT24CyAn2yXm2NgTQMS1mpzStP2ioPtTnszIQ Ih7ASVzhV6csHb8Yrkx8mgllOyrt9Y2kWRRJFm/FPRNEurOeNV6lnYAXOymVJwIDAQABo4IB0zCC Ac8wHwYDVR0jBBgwFoAUgq9sjPjF/pZhfOgfPStxSF7Ei8AwHQYDVR0OBBYEFLfuNf820LvaT4AK xrGK3EKx1DE7MA4GA1UdDwEB/wQEAwIFoDAMBgNVHRMBAf8EAjAAMB0GA1UdJQQWMBQGCCsGAQUF BwMEBggrBgEFBQcDAjBGBgNVHSAEPzA9MDsGDCsGAQQBsjEBAgEDBTArMCkGCCsGAQUFBwIBFh1o dHRwczovL3NlY3VyZS5jb21vZG8ubmV0L0NQUzBaBgNVHR8EUzBRME+gTaBLhklodHRwOi8vY3Js LmNvbW9kb2NhLmNvbS9DT01PRE9SU0FDbGllbnRBdXRoZW50aWNhdGlvbmFuZFNlY3VyZUVtYWls Q0EuY3JsMIGLBggrBgEFBQcBAQR/MH0wVQYIKwYBBQUHMAKGSWh0dHA6Ly9jcnQuY29tb2RvY2Eu Y29tL0NPTU9ET1JTQUNsaWVudEF1dGhlbnRpY2F0aW9uYW5kU2VjdXJlRW1haWxDQS5jcnQwJAYI KwYBBQUHMAGGGGh0dHA6Ly9vY3NwLmNvbW9kb2NhLmNvbTAeBgNVHREEFzAVgRNkd213MkBpbmZy YWRlYWQub3JnMA0GCSqGSIb3DQEBCwUAA4IBAQALbSykFusvvVkSIWttcEeifOGGKs7Wx2f5f45b nv2ghcxK5URjUvCnJhg+soxOMoQLG6+nbhzzb2rLTdRVGbvjZH0fOOzq0LShq0EXsqnJbbuwJhK+ PnBtqX5O23PMHutP1l88AtVN+Rb72oSvnD+dK6708JqqUx2MAFLMevrhJRXLjKb2Mm+/8XBpEw+B 7DisN4TMlLB/d55WnT9UPNHmQ+3KFL7QrTO8hYExkU849g58Dn3Nw3oCbMUgny81ocrLlB2Z5fFG Qu1AdNiBA+kg/UxzyJZpFbKfCITd5yX49bOriL692aMVDyqUvh8fP+T99PqorH4cIJP6OxSTdxKM MIIFHDCCBASgAwIBAgIRAOK7SUh5KuwJ6cSlGPGZWGYwDQYJKoZIhvcNAQELBQAwgZcxCzAJBgNV BAYTAkdCMRswGQYDVQQIExJHcmVhdGVyIE1hbmNoZXN0ZXIxEDAOBgNVBAcTB1NhbGZvcmQxGjAY BgNVBAoTEUNPTU9ETyBDQSBMaW1pdGVkMT0wOwYDVQQDEzRDT01PRE8gUlNBIENsaWVudCBBdXRo ZW50aWNhdGlvbiBhbmQgU2VjdXJlIEVtYWlsIENBMB4XDTE5MDEwMjAwMDAwMFoXDTIyMDEwMTIz NTk1OVowJDEiMCAGCSqGSIb3DQEJARYTZHdtdzJAaW5mcmFkZWFkLm9yZzCCASIwDQYJKoZIhvcN AQEBBQADggEPADCCAQoCggEBALL98Dmy0wm1AOxiaio/bxxn/hWvraZDvmUVb3vRKTbDLH14bxaW /EYC/Lw/i9d98CunLGKA1O8yogKPdhTFFmmNR8uh6r1a/aHr301d8YQea0DtURyaAH5L4cvsKY6O 0HF7s5DqYm6tmLq1UrRwIhqgDjBjF4XFRqj7hoXFXFc1VI7LxMwFfT6PStq6WedhROKw5KQytaQS vrjgMjpICIP3A/CroGr+bcmqcnXljGUSUB9bzEOjlU9uAsgJ9sl5tjYE0DEtZqc0rT9oqD7U57My ECIewElc4VenLB2/GK5MfJoJZTsq7fWNpFkUSRZvxT0TRLqznjVepZ2AFzsplScCAwEAAaOCAdMw ggHPMB8GA1UdIwQYMBaAFIKvbIz4xf6WYXzoHz0rcUhexIvAMB0GA1UdDgQWBBS37jX/NtC72k+A CsaxitxCsdQxOzAOBgNVHQ8BAf8EBAMCBaAwDAYDVR0TAQH/BAIwADAdBgNVHSUEFjAUBggrBgEF BQcDBAYIKwYBBQUHAwIwRgYDVR0gBD8wPTA7BgwrBgEEAbIxAQIBAwUwKzApBggrBgEFBQcCARYd aHR0cHM6Ly9zZWN1cmUuY29tb2RvLm5ldC9DUFMwWgYDVR0fBFMwUTBPoE2gS4ZJaHR0cDovL2Ny bC5jb21vZG9jYS5jb20vQ09NT0RPUlNBQ2xpZW50QXV0aGVudGljYXRpb25hbmRTZWN1cmVFbWFp bENBLmNybDCBiwYIKwYBBQUHAQEEfzB9MFUGCCsGAQUFBzAChklodHRwOi8vY3J0LmNvbW9kb2Nh LmNvbS9DT01PRE9SU0FDbGllbnRBdXRoZW50aWNhdGlvbmFuZFNlY3VyZUVtYWlsQ0EuY3J0MCQG CCsGAQUFBzABhhhodHRwOi8vb2NzcC5jb21vZG9jYS5jb20wHgYDVR0RBBcwFYETZHdtdzJAaW5m cmFkZWFkLm9yZzANBgkqhkiG9w0BAQsFAAOCAQEAC20spBbrL71ZEiFrbXBHonzhhirO1sdn+X+O W579oIXMSuVEY1LwpyYYPrKMTjKECxuvp24c829qy03UVRm742R9Hzjs6tC0oatBF7KpyW27sCYS vj5wbal+TttzzB7rT9ZfPALVTfkW+9qEr5w/nSuu9PCaqlMdjABSzHr64SUVy4ym9jJvv/FwaRMP gew4rDeEzJSwf3eeVp0/VDzR5kPtyhS+0K0zvIWBMZFPOPYOfA59zcN6AmzFIJ8vNaHKy5QdmeXx RkLtQHTYgQPpIP1Mc8iWaRWynwiE3ecl+PWzq4i+vdmjFQ8qlL4fHz/k/fT6qKx+HCCT+jsUk3cS jDCCBeYwggPOoAMCAQICEGqb4Tg7/ytrnwHV2binUlYwDQYJKoZIhvcNAQEMBQAwgYUxCzAJBgNV BAYTAkdCMRswGQYDVQQIExJHcmVhdGVyIE1hbmNoZXN0ZXIxEDAOBgNVBAcTB1NhbGZvcmQxGjAY BgNVBAoTEUNPTU9ETyBDQSBMaW1pdGVkMSswKQYDVQQDEyJDT01PRE8gUlNBIENlcnRpZmljYXRp b24gQXV0aG9yaXR5MB4XDTEzMDExMDAwMDAwMFoXDTI4MDEwOTIzNTk1OVowgZcxCzAJBgNVBAYT AkdCMRswGQYDVQQIExJHcmVhdGVyIE1hbmNoZXN0ZXIxEDAOBgNVBAcTB1NhbGZvcmQxGjAYBgNV BAoTEUNPTU9ETyBDQSBMaW1pdGVkMT0wOwYDVQQDEzRDT01PRE8gUlNBIENsaWVudCBBdXRoZW50 aWNhdGlvbiBhbmQgU2VjdXJlIEVtYWlsIENBMIIBIjANBgkqhkiG9w0BAQEFAAOCAQ8AMIIBCgKC AQEAvrOeV6wodnVAFsc4A5jTxhh2IVDzJXkLTLWg0X06WD6cpzEup/Y0dtmEatrQPTRI5Or1u6zf +bGBSyD9aH95dDSmeny1nxdlYCeXIoymMv6pQHJGNcIDpFDIMypVpVSRsivlJTRENf+RKwrB6vcf WlP8dSsE3Rfywq09N0ZfxcBa39V0wsGtkGWC+eQKiz4pBZYKjrc5NOpG9qrxpZxyb4o4yNNwTqza aPpGRqXB7IMjtf7tTmU2jqPMLxFNe1VXj9XB1rHvbRikw8lBoNoSWY66nJN/VCJv5ym6Q0mdCbDK CMPybTjoNCQuelc0IAaO4nLUXk0BOSxSxt8kCvsUtQIDAQABo4IBPDCCATgwHwYDVR0jBBgwFoAU u69+Aj36pvE8hI6t7jiY7NkyMtQwHQYDVR0OBBYEFIKvbIz4xf6WYXzoHz0rcUhexIvAMA4GA1Ud DwEB/wQEAwIBhjASBgNVHRMBAf8ECDAGAQH/AgEAMBEGA1UdIAQKMAgwBgYEVR0gADBMBgNVHR8E RTBDMEGgP6A9hjtodHRwOi8vY3JsLmNvbW9kb2NhLmNvbS9DT01PRE9SU0FDZXJ0aWZpY2F0aW9u QXV0aG9yaXR5LmNybDBxBggrBgEFBQcBAQRlMGMwOwYIKwYBBQUHMAKGL2h0dHA6Ly9jcnQuY29t b2RvY2EuY29tL0NPTU9ET1JTQUFkZFRydXN0Q0EuY3J0MCQGCCsGAQUFBzABhhhodHRwOi8vb2Nz cC5jb21vZG9jYS5jb20wDQYJKoZIhvcNAQEMBQADggIBAHhcsoEoNE887l9Wzp+XVuyPomsX9vP2 SQgG1NgvNc3fQP7TcePo7EIMERoh42awGGsma65u/ITse2hKZHzT0CBxhuhb6txM1n/y78e/4ZOs 0j8CGpfb+SJA3GaBQ+394k+z3ZByWPQedXLL1OdK8aRINTsjk/H5Ns77zwbjOKkDamxlpZ4TKSDM KVmU/PUWNMKSTvtlenlxBhh7ETrN543j/Q6qqgCWgWuMAXijnRglp9fyadqGOncjZjaaSOGTTFB+ E2pvOUtY+hPebuPtTbq7vODqzCM6ryEhNhzf+enm0zlpXK7q332nXttNtjv7VFNYG+I31gnMrwfH M5tdhYF/8v5UY5g2xANPECTQdu9vWPoqNSGDt87b3gXb1AiGGaI06vzgkejL580ul+9hz9D0S0U4 jkhJiA7EuTecP/CFtR72uYRBcunwwH3fciPjviDDAI9SnC/2aPY8ydehzuZutLbZdRJ5PDEJM/1t yZR2niOYihZ+FCbtf3D9mB12D4ln9icgc7CwaxpNSCPt8i/GqK2HsOgkL3VYnwtx7cJUmpvVdZ4o gnzgXtgtdk3ShrtOS1iAN2ZBXFiRmjVzmehoMof06r1xub+85hFQzVxZx5/bRaTKTlL8YXLI8nAb R9HWdFqzcOoB/hxfEyIQpx9/s81rgzdEZOofSlZHynoSMYIDyjCCA8YCAQEwga0wgZcxCzAJBgNV BAYTAkdCMRswGQYDVQQIExJHcmVhdGVyIE1hbmNoZXN0ZXIxEDAOBgNVBAcTB1NhbGZvcmQxGjAY BgNVBAoTEUNPTU9ETyBDQSBMaW1pdGVkMT0wOwYDVQQDEzRDT01PRE8gUlNBIENsaWVudCBBdXRo ZW50aWNhdGlvbiBhbmQgU2VjdXJlIEVtYWlsIENBAhEA4rtJSHkq7AnpxKUY8ZlYZjANBglghkgB ZQMEAgEFAKCCAe0wGAYJKoZIhvcNAQkDMQsGCSqGSIb3DQEHATAcBgkqhkiG9w0BCQUxDxcNMjEw NzAxMTczOTAwWjAvBgkqhkiG9w0BCQQxIgQgVhewBAFQcYyIFEK/LXK5fkSb/rJPosiwWyfI+Gx2 22Awgb4GCSsGAQQBgjcQBDGBsDCBrTCBlzELMAkGA1UEBhMCR0IxGzAZBgNVBAgTEkdyZWF0ZXIg TWFuY2hlc3RlcjEQMA4GA1UEBxMHU2FsZm9yZDEaMBgGA1UEChMRQ09NT0RPIENBIExpbWl0ZWQx PTA7BgNVBAMTNENPTU9ETyBSU0EgQ2xpZW50IEF1dGhlbnRpY2F0aW9uIGFuZCBTZWN1cmUgRW1h aWwgQ0ECEQDiu0lIeSrsCenEpRjxmVhmMIHABgsqhkiG9w0BCRACCzGBsKCBrTCBlzELMAkGA1UE BhMCR0IxGzAZBgNVBAgTEkdyZWF0ZXIgTWFuY2hlc3RlcjEQMA4GA1UEBxMHU2FsZm9yZDEaMBgG A1UEChMRQ09NT0RPIENBIExpbWl0ZWQxPTA7BgNVBAMTNENPTU9ETyBSU0EgQ2xpZW50IEF1dGhl bnRpY2F0aW9uIGFuZCBTZWN1cmUgRW1haWwgQ0ECEQDiu0lIeSrsCenEpRjxmVhmMA0GCSqGSIb3 DQEBAQUABIIBAHSDYc7iqjtvXKKTDoPkcDH9xgRnOreGSA9a7D0r7wBr90jDVbi16YWI/2jA5kAb nzyii+qMD0RHsXkSvMUPUxwDAc1kgoLS3wj6DWxqMzsykirc6lloZw9dqSCw5sczXp04OJjywYZT qSuhsFApGpOTTf5S9cf04ZXVKiYMNli8AMkwNdeDsN2l/GZRGdd3aLxTOO7qfJLHrgPaz5W1iO7R +3PGBVAhkU8LGNKcvd920z0jP0IxDTB6z1QpoQOC85NlI0U6peAGEW93j8xo6GV01xFLhVLc12oR SUyRlrK1ydwcUrQxkqh1+4tjDmNVVjRsdLQPr46y3tfglrXCmbIAAAAAAAA= --=-CAlq5QINaohkmriR4npb--