I re-ran the numbers on 10GE against the actual alacrityvm v0.1 release available in git on kernel.org. I tried to include the newly announced "vhost" driver (Michael Tsirkin) for virtio acceleration, but ran into issues getting the patches to apply. For now, this includes native, virtio-u (virtio-userspace), and venet all running on 31-rc4. If I can resolve the issue with Michaels patches, I will add "virtio-k" (virtio-kernel) to the mix as well. For now, here are the results for ...

Nice quote. Even 15 years after it was published, those first two graphs are prime "bad examples". The second two could stand some improvement also - like put the legend and numbers *on* each solid bar. @J.G. - I think you are fighting an uphill battle here against human nature. People tend to go with their idea of "pretty". Mike --

Anthony is correct. The new numbers after fixing the HRT clock issue are: virtio-u: 2670Mb/s, 266us rtt (3764 tps udp-rr) I will update the charts later tonight. Sorry for the confusion. -Greg

RFC 4653 specifies Non-Congestion Robustness (NCR) for TCP. In the absence of explicit congestion notification from the network, TCP uses loss as an indication of congestion. One of the ways TCP detects loss is using the arrival of three duplicate acknowledgments. However, this heuristic is not always correct, notably in the case when network paths reorder segments (for whatever reason), resulting in degraded performance. TCP-NCR is designed to mitigate this degraded performance by increasing ...

On Wed, 12 Aug 2009 20:50:59 +0200 Patch has funny indentation and awkward naming for socket elements. Your style needs to match existing code. What is the usage model for this? I expect that some user who wants to enable this would be stuck somewhere with a lossy network and would want to enable it. Or is it something only researchers will want to play with? It would be easier to use a sysctl value for this because otherwise each application has to be changed to select the socket ...

Signed-off-by: Yitchak Gertner <gertner@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_dump.h | 890 +++++++++++++++++++++++----------------------- 1 files changed, 449 insertions(+), 441 deletions(-) diff --git a/drivers/net/bnx2x_dump.h b/drivers/net/bnx2x_dump.h index 78c6b03..3bb9a91 100644 --- a/drivers/net/bnx2x_dump.h +++ b/drivers/net/bnx2x_dump.h @@ -13,31 +13,35 @@ * The signature is time stamp, diag version and grc_dump version ...

Moving the code to a more logical place and beautifying it. No real change in behavior. Signed-off-by: Vladislav Zolotarov <vladz@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x.h | 28 +++- drivers/net/bnx2x_init.h | 320 ++++++-------------------------- drivers/net/bnx2x_init_ops.h | 419 ++++++++++++++++++++++++------------------ drivers/net/bnx2x_main.c | 79 ++++++-- drivers/net/bnx2x_reg.h | 113 +++++++++++ 5 ...

Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_link.c | 108 ++++++++++++--------------------------------- drivers/net/bnx2x_link.h | 12 +++-- drivers/net/bnx2x_main.c | 8 +--- 3 files changed, 39 insertions(+), 89 deletions(-) diff --git a/drivers/net/bnx2x_link.c b/drivers/net/bnx2x_link.c index 74f4d10..c2b0010 100644 --- a/drivers/net/bnx2x_link.c +++ b/drivers/net/bnx2x_link.c @@ -1547,10 +1547,7 @@ static u8 bnx2x_ext_phy_resove_fc(struct ...

Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_main.c | 4 ++-- 1 files changed, 2 insertions(+), 2 deletions(-) diff --git a/drivers/net/bnx2x_main.c b/drivers/net/bnx2x_main.c index 54e3ef9..037d862 100644 --- a/drivers/net/bnx2x_main.c +++ b/drivers/net/bnx2x_main.c @@ -56,8 +56,8 @@ #include "bnx2x_init_ops.h" #include "bnx2x_dump.h" -#define DRV_MODULE_VERSION "1.48.114-1" -#define DRV_MODULE_RELDATE "2009/07/29" +#define ...

Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_main.c | 9 +++------ 1 files changed, 3 insertions(+), 6 deletions(-) diff --git a/drivers/net/bnx2x_main.c b/drivers/net/bnx2x_main.c index a409767..23b9c7d 100644 --- a/drivers/net/bnx2x_main.c +++ b/drivers/net/bnx2x_main.c @@ -138,12 +138,9 @@ static struct { static const struct pci_device_id bnx2x_pci_tbl[] = { - { PCI_VENDOR_ID_BROADCOM, PCI_DEVICE_ID_NX2_57710, - PCI_ANY_ID, PCI_ANY_ID, 0, 0, ...

Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_main.c | 6 +++--- 1 files changed, 3 insertions(+), 3 deletions(-) diff --git a/drivers/net/bnx2x_main.c b/drivers/net/bnx2x_main.c index bdf7228..a409767 100644 --- a/drivers/net/bnx2x_main.c +++ b/drivers/net/bnx2x_main.c @@ -2926,7 +2926,7 @@ static inline void bnx2x_attn_int_deasserted0(struct bnx2x *bp, u32 attn) REG_WR(bp, reg_offset, val); BNX2X_ERR("FATAL HW block attention set0 0x%x\n", - ...

Disable bmac access while working with emac and keep the single lane SerDes in reset while working with 4 lanes XGXS Signed-off-by: Yaniv Rosner <yanivr@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_link.c | 7 ++++--- 1 files changed, 4 insertions(+), 3 deletions(-) diff --git a/drivers/net/bnx2x_link.c b/drivers/net/bnx2x_link.c index dc3b69e..b81a057 100644 --- a/drivers/net/bnx2x_link.c +++ b/drivers/net/bnx2x_link.c @@ -397,7 +397,8 @@ ...

Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_main.c | 69 ++++++++++++++++++++++------------------------ 1 files changed, 33 insertions(+), 36 deletions(-) diff --git a/drivers/net/bnx2x_main.c b/drivers/net/bnx2x_main.c index 347036f..dd9a77f 100644 --- a/drivers/net/bnx2x_main.c +++ b/drivers/net/bnx2x_main.c @@ -8946,50 +8946,15 @@ static int bnx2x_set_settings(struct net_device *dev, struct ethtool_cmd *cmd) return 0; } -#define ...

Signed-off-by: Benjamin Li <benli@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x.h | 1 + drivers/net/bnx2x_main.c | 8 ++++++++ 2 files changed, 9 insertions(+), 0 deletions(-) diff --git a/drivers/net/bnx2x.h b/drivers/net/bnx2x.h index 004f4a8..633acca 100644 --- a/drivers/net/bnx2x.h +++ b/drivers/net/bnx2x.h @@ -89,6 +89,7 @@ } while (0) #else #define bnx2x_panic() do { \ + bp->panic = 1; \ BNX2X_ERR("driver assert\n"); ...

Move printing of version from probe to the init function Rather then checking if this is the first module probe call to print the version of the driver only once, the statement is moved to the init function of the module where init is only called once Signed-off-by: Benjamin Li <benli@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_main.c | 6 ++---- 1 files changed, 2 insertions(+), 4 deletions(-) diff --git a/drivers/net/bnx2x_main.c ...

When a given ring is running out of space, the FW can send pause towards the network. When working with multi-queues, when one queue is getting out of space it can block all other queues. The preferred scheme is to send pause frames only when running out of the shared internal chip buffers and if a given queue cannot place a packet on the host, it will drop it. Since some users might want to work in drop-less mode, allowing changing the behavior as a module parameter. Signed-off-by: Eilon ...

Signed-off-by: Vladislav Zolotarov <vladz@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_main.c | 7 +++++++ 1 files changed, 7 insertions(+), 0 deletions(-) diff --git a/drivers/net/bnx2x_main.c b/drivers/net/bnx2x_main.c index e6fde5e..594168a 100644 --- a/drivers/net/bnx2x_main.c +++ b/drivers/net/bnx2x_main.c @@ -1497,6 +1497,13 @@ static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget) bd_prod = RX_BD(bd_prod); bd_cons = ...

There are 3 operations that the driver needs to support to allow applications to access the PHY FW (on top of the MDC/MDIO access). Since those are essentially nvram access commands, adding them to the ethtool -E interface. Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_link.c | 21 ++++++------- drivers/net/bnx2x_link.h | 4 ++ drivers/net/bnx2x_main.c | 72 ++++++++++++++++++++++++++++++++++----------- 3 files changed, 67 insertions(+), 30 ...

Adding CL73 support to the built in PHY in the 5771x device. Also supporting fallbacks to CL73 if the link partner does not respond. Signed-off-by: Yaniv Rosner <yanivr@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_link.c | 197 +++++++++++++++++++++++++++++++++++++++------- drivers/net/bnx2x_reg.h | 13 +++ 2 files changed, 180 insertions(+), 30 deletions(-) diff --git a/drivers/net/bnx2x_link.c b/drivers/net/bnx2x_link.c index ...

The BCM8481 does not generate LASI interrupt for 10M, 100M and 1G link, so we are using LED4 output as the interrupt input to the 57711. This requires some adaptation in the link interrupt routines Signed-off-by: Yaniv Rosner <yanivr@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_link.c | 498 ++++++++++++++++++++++++++++++++++++++++------ drivers/net/bnx2x_reg.h | 32 +++ 2 files changed, 468 insertions(+), 62 deletions(-) diff --git ...

Signed-off-by: Yaniv Rosner <yanivr@broadcom.com> Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_link.c | 118 ++++++++++++++++++++++++++++++++++++++++++++-- 1 files changed, 113 insertions(+), 5 deletions(-) diff --git a/drivers/net/bnx2x_link.c b/drivers/net/bnx2x_link.c index 9e1f19a..c925249 100644 --- a/drivers/net/bnx2x_link.c +++ b/drivers/net/bnx2x_link.c @@ -2046,6 +2046,111 @@ static void bnx2x_save_bcm_spirom_ver(struct bnx2x *bp, u8 port, ...

Signed-off-by: Eilon Greenstein <eilong@broadcom.com> --- drivers/net/bnx2x_main.c | 4 +--- 1 files changed, 1 insertions(+), 3 deletions(-) diff --git a/drivers/net/bnx2x_main.c b/drivers/net/bnx2x_main.c index 442ba61..47b687b 100644 --- a/drivers/net/bnx2x_main.c +++ b/drivers/net/bnx2x_main.c @@ -2148,9 +2148,7 @@ static u8 bnx2x_initial_phy_init(struct bnx2x *bp, int load_mode) /* Initialize link parameters structure variables */ /* It is recommended to turn off RX FC for ...

Hi Dave, Here is a patch series for the bnx2x. This patch series also replace the FW, so it contains two big blobs - the new fw and the removal of the old one. Those patches do not contain anything but the ihex - the actually change to the driver is in patch number 2 which is small enough to fit the mailing list. For those who wish to see all the patches, including the ihex, I also updated http://linux.broadcom.com/eilong/ to contain this patch series. Please consider applying to ...

If an interface has multiple addresses, the current message for DAD failure isn't really helpful, so this patch adds the address itself to the printk. Signed-off-by: Jens Rosenboom <jens@mcbone.net> --- diff --git a/net/ipv6/addrconf.c b/net/ipv6/addrconf.c index 43b3c9f..01a4b25 100644 --- a/net/ipv6/addrconf.c +++ b/net/ipv6/addrconf.c @@ -1403,8 +1403,8 @@ void addrconf_dad_failure(struct inet6_ifaddr *ifp) struct inet6_dev *idev = ifp->idev; if ...

This patch adds basic checkpoint/restart support for AF_UNIX sockets. It has been tested with a single and multiple processes, and with data inflight at the time of checkpoint. It supports socketpair()s, path-based, and abstract sockets. Changes in v9: - Fix double-free of skb's in the list and target holding queue in the error path of sock_copy_buffers() - Adjust use of ckpt_read_string() to match new signature Changes in v8: - Fix stale dev_alloc_skb() from before the ...

950.000.00 GBP has been Awarded to you in our LG Electronics,send our office your Names:............ Address:.......... Country:.......... --

Subject: virtio: net refill on out-of-memory If we run out of memory, use keventd to fill the buffer. There's a report of this happening: "Page allocation failures in guest", Message-ID: <20090713115158.0a4892b0@mjolnir.ossman.eu> Signed-off-by: Rusty Russell <rusty@rustcorp.com.au> diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -71,6 +71,9 @@ struct virtnet_info struct sk_buff_head recv; struct ...

Hmm, might as well, although this is v. unlikely to happen. Thanks, Rusty. diff --git a/drivers/net/virtio_net.c b/drivers/net/virtio_net.c --- a/drivers/net/virtio_net.c +++ b/drivers/net/virtio_net.c @@ -72,7 +72,7 @@ struct virtnet_info struct sk_buff_head send; /* Work struct for refilling if we run low on memory. */ - struct work_struct refill; + struct delayed_work refill; /* Chain pages by the private ptr. */ struct page *pages; @@ -402,19 +402,16 @@ static void ...

It inserts memory barriers on architectures that require them (currently only the Alpha), and, more importantly, documents I don't think so. sync-barrier has nothing to do with it as it comes --

Is it? I don't see where RCU actually enters the equation if you are not bracketing the dereference with rcu_read_lock/unlock. I'm sure you have this correct, its just that it goes against my understanding of how to use RCU properly so I am trying to understand what you did. I'm I still am not seeing it. Even rcupdate.h says: /** * rcu_dereference - fetch an RCU-protected pointer in an * RCU read-side critical section. This pointer may later * be safely dereferenced. note ...

Yes, that is understood. Perhaps you should just use a normal barrier, however. (Or at least a comment that says "I am just using this for its More correctly: it "smells like" RCU, but its not. ;) It's rcu-like, but you are not really using the rcu facilities. I think anyone that knows RCU and reads your code will likely be scratching their heads as well. Its probably not a big deal, as I understand your code now. Just a suggestion to help clarify it. Regards, -Greg

If you are using call_rcu(), synchronize_rcu(), or one of the similar primitives, then you absolutely need rcu_read_lock() and rcu_read_unlock(), or one of the similar pairs of primitives. If you -don't- use rcu_read_lock(), then you are pretty much restricted to adding data, but never removing it. Make sense? ;-) Thanx, Paul --

Well, you -do- need -something- that takes on the role of rcu_read_lock(), and in your case you in fact actually do. Your equivalent of rcu_read_lock() is the beginning of execution of a workqueue item, and the equivalent of rcu_read_unlock() is the end of execution of that same workqueue item. Implicit, but no less real. If a couple more uses like this show up, I might need to add this to Documentation/RCU. ;-) Thanx, Paul --

And I idly wonder if this approach could replace SRCU. Probably not for protecting the CPU-hotplug notifier chains, but worth some thought. Thanx, Paul --

Are you saying SRIOV is a requirement, and I can either program the SRIOV adapter with a mac or use promis? Or are you saying I can use Out of curiosity, wouldnt you only need to disable LRO on the adapter, since the other two (IIUC) are transmit path and are therefore You mean via "ip link" I assume? Regards, -Greg

Makes sense. Got it. I was going to add guest-to-guest to the test matrix, but I assume that is not supported with vhost unless you have something like a VEPA enabled bridge? Ah, ok. Even better. Thanks! -Greg

Presumably you mean on the same host? There were also some patches to enable local guest to guest for macvlan, that would be a nice software-only solution. For back to back, I just tried over veth, seems --

Oh, hopefully macvlan will soon allow that. -- MST --

Yeah, this would be the config I would be interested in. Regards, -Greg

I can try to do a few variations, but what I am interested is in performance in a real-world L2 configuration. This would generally mean all hosts (virtual or physical) in the same L2 domain. If I get a chance, though, I will try to also wire them up in isolation as another data point. Regards, -Greg

From: "John W. Linville" <linville@tuxdriver.com> Ok, I'll take care of that. --

Well I've been applying the equivalent of this patch privately since we started using the libertas driver. We build systems with one or two wired Ethernets and then an optional wireless module. A fixed name wlan0 is a lot easier than explaining to a user that the interface might be eth1 or eth2 depending on which model they have, or that eth1 might be wired or wireless. It also simplifies scripts and configuration file handling. I'm sure there are many other solutions for big systems but ...

I don't care either way, it's completely historical. Most of the fullmac drivers used 'eth' back when. Might want to get buy-in from the OLPC crew since they probably have the most deployed units using libertas (cc-ed Daniel Drake). Daniel, is it a problem for you guys if the libertas wifi interface name went from 'eth' -> 'wlan' ? Mesh name would be unchanged. Dan --

Hi all, In the -next tree, it looks like this: int genl_register_mc_group(struct genl_family *family, struct genl_multicast_group *grp) { int id; unsigned long *new_groups; int err; BUG_ON(grp->name[0] == '\0'); genl_lock(); /* special-case our own group */ if (grp == &notify_grp) id = GENL_ID_CTRL; else id = find_first_zero_bit(mc_groups, mc_groups_longs * BITS_PER_LONG); if (id >= mc_groups_longs * BITS_PER_LONG) { size_t nlen = ...

Nope, kfree_skb() frees the frags. It needs to, otherwise we leak on every received packet! Cheers, Rusty. --

We discussed this before, and I still think this could be directly derived from struct virtqueue, in the same way that vring_virtqueue is derived from struct virtqueue. That would make it possible for simple device drivers to use the same driver in both host and guest, similar to how Ira Snyder used virtqueues to make virtio_net run between two hosts running the same code [1]. Ideally, I guess you should be able to even make virtio_net work in the host if you do that, but that could bring ...

I prefer keeping it simple. Much of abstraction in virtio is due to the fact that it needs to work on top of different hardware emulations: lguest,kvm, possibly others in the future. vhost is always working on I don't think so. For example, there's a callback field that gets invoked in guest when buffers are consumed. It could be overloaded to mean "buffers are available" in host but you never handle both As I pointed out earlier, most code in virtio net is asymmetrical: guest provides ...

Ok, that makes sense. Let me see if I understand the concept of the driver. Here's a picture of what makes sense to me: guest system --------------------------------- | userspace applications | --------------------------------- | kernel network stack | --------------------------------- | virtio-net | --------------------------------- | transport (virtio-ring, etc.) | --------------------------------- | ...

Actually, vhost may not always be limited to real hardware. We may on day use vhost as the basis of a driver domain. There's quite a lot of interest in this for networking. At any rate, I'd like to see performance results before we consider trying to reuse virtio code. Regards, Anthony Liguori --

Much better -- a couple of documentation nits below. How about something like "Therefore the beginning of workqueue execution acts as rcu_read_lock() and the end of workqueue execution acts as rcu_read_lock()"? It would also be good to add comments to the workqueue functions themselves saying that they act as read-side critical sections for your kind of RCU. --

--

IIUC it is possible for multiple processes to attach to the same tun device. Will it get confusing/incorrect to have each attach --

Ah yes, you're right - I saw the check for (ifr->ifr_flags & IFF_TUN_EXCL) in Ok, thanks. To my untrained eye the class addition looks right too, so with the trivial change: Acked-by: Serge Hallyn <serue@us.ibm.com> thanks, -serge --

OL> Before pulling this one, I took a quick look at this patch, and I OL> saw that it still uses skb_morph despite the changelog and my OL> memory... That's correct. We've been through several ways of allocating the skb's, so it's definitely confusing. We're back to skb_morph() because I'm pre-allocating them for lock safety when traversing the queues. The only thing that is allocated is the actual skb structure itself; the buffers are still shared like with skb_clone(). OL> 1) Move ...

Currently we do all the work from special worker threads, so it's possible for this callback to sleep. The error isn't yet propagated to We were using master netdevices for several purposes: 1) ip link add link mwpanX type wpan, so that we have out-of-box support for radio additions. That's really nice thing to have. 2) for SOCK_RAW implementation that can be used to send raw packets over-the-air/receive raw packets. I think we can use af_packet for this, but I'm still not sure ...

Hmmm. Really weird. Then, if we want to pass data from socket layer to MAC layer, we should place data in skb->data and not in skb->cb (like Nice idea. Thanks a lot! -- With best wishes Dmitry --

I did have NUMA enabled, and memory was configured as independent rather than interleaved. Based on all the discussions, it seemed a good possibility that the BIOS was broken. Today a colleague checked the SuperMicro site, and discovered and installed a newer version of the BIOS. Things seem better now, but not totally correct. There are now NUMA nodes 0 and 1 instead of 0 and 2, and the CPUs for node 0 are 0 through 3 while the CPUs for node 1 are 4 through 7 (previously the even CPUs ...

That might be ok, depending on how the APICs are configured. Of course you should have the same number of CPUs on the different Most likely you need the appended patch from linux-next. It should be probably in .31, but I can't see it in linus' tree only in -next. Jesse? No. -Andi commit eaf2f454cc9a76dbe1890af6269e60fe9978a3a5 Author: Jesse Barnes <jbarnes@virtuousgeek.org> Date: Fri Jul 10 14:04:30 2009 -0700 x86/PCI: initialize PCI bus node numbers early ...

From: Bill Fink <billfink@mindspring.com> This, unfortunately, won't be comprehensive. You'd also need to kludge the NUMA node used for allocation of the skb->data buffer via the netdev_alloc_skb() calls in myri10ge_rx_done() and friends. This could possibly account for why, with your kludge, you still were only getting 56.4703 Gbps --

The macro DRV_NAME contains "korina", the field dev->name points to the actual interface name. So messages were formerly prefixed with 'korinaeth2:' (on my system). Signed-off-by: Phil Sutter <n0-1@freewrt.org> --- drivers/net/korina.c | 32 +++++++++++++++++--------------- 1 files changed, 17 insertions(+), 15 deletions(-) diff --git a/drivers/net/korina.c b/drivers/net/korina.c index b4cf602..6df9d25 100644 --- a/drivers/net/korina.c +++ b/drivers/net/korina.c @@ -338,7 +338,7 @@ ...

On Thu, Aug 06, 2009 at 05:57:02PM -0400, Dave Jones wrote: > I'm chasing yet another dma-debug warning where we're unmapping a different > size to what we mapped. > > > WARNING: at lib/dma-debug.c:803 check_unmap+0x1f5/0x509() (Not tainted) > > Hardware name: > > 8139cp 0000:00:03.0: DMA-API: device driver frees DMA memory with different > > size [device address=0x000000001e9f8852] [map size=1536 bytes] [unmap size=1538 > > bytes] > > Modules linked in: ipv6 dm_multipath ...

BTW, I've seen the same issue in 2.6.28 and 2.6.29 while doing a bunch of NFS-over-UDP testing. I've seen the issue reported in 2.6.27 as well, but it went by ignored. It's not easy to reproduce as it seems like it requires quite a bit traffic over over multiple interfaces. I've been looking at this for a while and haven't caught the bugger. Here is the stack trace from 2.6.28: May 13 16:17:38 dl380g6-2 kernel: [ 4473.086015] ------------[ cut here ]------- ----- May 13 16:17:38 ...

I simply did not think of the sole pn_sock_seq_fops as "so many things"... But I can change it if you think it would be better. -- Rémi Denis-Courmont Nokia Devices R&D, Maemo Software, Helsinki --

Right, that question is still open, and dont't see it as very important Not yet, but I guess it comes as a natural extension when I fix multicast/broadcast delivery from the reflective relay for VEPA. The logic that I would use there is: broadcast from a dowstream port: if (bridge_mode(source_port)) { forward_to_upstream(frame); for_each_downstream(port) { /* deliver to all bridge ports except self, do not deliver to any VEPA ...

Right, this would be a logical extension in that scenario. I would imagine that in many scenarios running a VEB also means that you want to use the advanced ebtables/iptables filtering of the bridge subsystem, but if all guests trust each other, using macvlan to bridge between them You can of course do that by adding one port of the S-component to a port of a bridge, and using another port of the S-component to create macvlan devices, or you could have multiple ports of the S-component each ...