- 05 Jun, 2017 40 commits
-
-
David Howells authored
Implement AuriStor's service upgrade facility. There are three problems that this is meant to deal with: (1) Various of the standard AFS RPC calls have IPv4 addresses in their requests and/or replies - but there's no room for including IPv6 addresses. (2) Definition of IPv6-specific RPC operations in the standard operation sets has not yet been achieved. (3) One could envision the creation a new service on the same port that as the original service. The new service could implement improved operations - and the client could try this first, falling back to the original service if it's not there. Unfortunately, certain servers ignore packets addressed to a service they don't implement and don't respond in any way - not even with an ABORT. This means that the client must then wait for the call timeout to occur. What service upgrade does is to see if the connection is marked as being 'upgradeable' and if so, change the service ID in the server and thus the request and reply formats. Note that the upgrade isn't mandatory - a server that supports only the original call set will ignore the upgrade request. In the protocol, the procedure is then as follows: (1) To request an upgrade, the first DATA packet in a new connection must have the userStatus set to 1 (this is normally 0). The userStatus value is normally ignored by the server. (2) If the server doesn't support upgrading, the reply packets will contain the same service ID as for the first request packet. (3) If the server does support upgrading, all future reply packets on that connection will contain the new service ID and the new service ID will be applied to *all* further calls on that connection as well. (4) The RPC op used to probe the upgrade must take the same request data as the shadow call in the upgrade set (but may return a different reply). GetCapability RPC ops were added to all standard sets for just this purpose. Ops where the request formats differ cannot be used for probing. (5) The client must wait for completion of the probe before sending any further RPC ops to the same destination. It should then use the service ID that recvmsg() reported back in all future calls. (6) The shadow service must have call definitions for all the operation IDs defined by the original service. To support service upgrading, a server should: (1) Call bind() twice on its AF_RXRPC socket before calling listen(). Each bind() should supply a different service ID, but the transport addresses must be the same. This allows the server to receive requests with either service ID. (2) Enable automatic upgrading by calling setsockopt(), specifying RXRPC_UPGRADEABLE_SERVICE and passing in a two-member array of unsigned shorts as the argument: unsigned short optval[2]; This specifies a pair of service IDs. They must be different and must match the service IDs bound to the socket. Member 0 is the service ID to upgrade from and member 1 is the service ID to upgrade to. Signed-off-by:David Howells <dhowells@redhat.com>
-
David Howells authored
Permit bind() to be called on an AF_RXRPC socket more than once (currently maximum twice) to bind multiple listening services to it. There are some restrictions: (1) All bind() calls involved must have a non-zero service ID. (2) The service IDs must all be different. (3) The rest of the address (notably the transport part) must be the same in all (a single UDP socket is shared). (4) This must be done before listen() or sendmsg() is called. This allows someone to connect to the service socket with different service IDs and lays the foundation for service upgrading. The service ID used by an incoming call can be extracted from the msg_name returned by recvmsg(). Signed-off-by: -
David Howells authored
Keep the rxrpc_connection struct's idea of the service ID that is exposed in the protocol separate from the service ID that's used as a lookup key. This allows the protocol service ID on a client connection to get upgraded without making the connection unfindable for other client calls that also would like to use the upgraded connection. The connection's actual service ID is then returned through recvmsg() by way of msg_name. Whilst we're at it, we get rid of the last_service_id field from each channel. The service ID is per-connection, not per-call and an entire connection is upgraded in one go. Signed-off-by: -
David S. Miller authored
Jiri Pirko says: ==================== mlxsw: Minor cleanup Fix small issues I noticed during the refactoring. First patch adds file name comments in the header file to make it clear what goes where. Second patch fixes a typo and third patch simply aligns RIF index allocation with similar allocations in the driver. ==================== Signed-off-by:David S. Miller <davem@davemloft.net>
-
Ido Schimmel authored
The way we usually allocate an index is by letting the allocation function return an error instead of an invalid index. Do the same for RIF index. Signed-off-by:
Ido Schimmel <idosch@mellanox.com> Reviewed-by:
Petr Machata <petrm@mellanox.com> Signed-off-by:
Jiri Pirko <jiri@mellanox.com> Signed-off-by:
-
Ido Schimmel authored
Signed-off-by:
-
Ido Schimmel authored
Make it clear where functions are defined and move misplaced declaration to their correct place. Signed-off-by:
-
Yotam Gigi authored
Change the firmware file name to be in "mellanox" directory. This commit is a followup to the linux-firmware commit a4c72696f5f4 ("Mellanox: Add firmware for mlxsw_spectrum") Signed-off-by:Yotam Gigi <yotamg@mellanox.com> Reviewed-by:
-
David S. Miller authored
Yuval Mintz says: qed*: Support VF XDP attachment ==================== Each driver queue [Rx, Tx, XDP-forwarding] requires an allocated HW/FW connection + configured queue-zone. VF handling by the PF has several limitations that prevented adding the capability to perform XDP at driver-level: - The VF assumes there's 1-to-1 correspondance between the VF queue and the used connection, meaning q<x> is always going to use cid<x>, whereas for its own queues the PF is acquiring a new cid per each new queue. - There's a 1-to-1 correspondate between the VF-queues and the HW queue zones. While this is necessary for Rx-queues [as the queue-zone contains the producer], transmission queues can share the underlaying queue-zone [only shared configuration is coalescing]. But all VF<->PF communication mechanisms assume there's a single identifier that identify a queue [as queue-zone == queue], while sharing queue-zones requires passing additional information. - VFs currently don't try mapping a doorbell bar - there's a small doorbell window in the regview allowing VFs to doorbell up to 16 connections; but this window isn's wide enough for the added XDP forwarding queues. This series is going to add the necessary infrastrucutre to finally let our VFs support XDP assuming both the PF and VF drivers are sufficiently new [Legacy support would be retained both for older VFs and older PFs, but both will be needed for this new support to work]. Basically, the various database driver maintains for its queue-cids would be revised, and queue-cids would be identified using the (queue-zone, unique index) pair. The TLV mechanism would then be extended to allow VFs to communicate that unique-index as well as the already provided queue-zone. Finally, the VFs would try to map their doorbell bar and inform their PF that they're using it. Almost all the changes are in qed, with exception of #3 [which does some cleanup in qede as well] and #11 that actually enables the feature. ==================== Signed-off-by: -
Mintz, Yuval authored
This introduces 2 changes needed for XDP to be supported for VFs: a. On VF-side, publish the NDO based on qed outputs b. On PF-side, request qed to allocate sufficient cids per-VF to allow the child vfs to support it Signed-off-by:Yuval Mintz <Yuval.Mintz@cavium.com> Signed-off-by:
-
Mintz, Yuval authored
The final addition on the qed front - - VFs would now require their PFs to provide multiple CIDs - Based on the availability of connections from PF, determine whether XDP is feasible and share it with qede via dev_info. Signed-off-by:
-
Mintz, Yuval authored
VFs are currently not mapping their doorbell bar, instead relying on the small doorbell window they have in their limited regview bar. In order to increase the number of possible Tx connections [queues] employeed by VF past 16, we need to start using the doorbell bar if one such is exposed - VF would communicate this fact to PF which would return the size-bar internally configured into chip, according to which the VF would decide whether to actually utilize the doorbell bar. Signed-off-by:
-
Mintz, Yuval authored
This adds the infrastructure for supporting VFs that want to open multiple transmission queues on the same queue-zone. At this point, there are no VFs that actually request this functionality, but later patches would remedy that. a. VF and PF would communicate the capability during ACQUIRE; Legacy VFs would continue on behaving as they do today b. PF would communicate number of supported CIDs to the VF and would enforce said limitation c. Whenever VF passes a request for a given queue configuration it would also pass an associated index within said queue-zone Signed-off-by: -
Mintz, Yuval authored
Allow the infrastructure a PF maintains for each one of its VFs to support multiple queue-cids on a single queue-zone. Signed-off-by:
-
Mintz, Yuval authored
Until now we used to have a single VF legacy compatibility mode, one that affected the place of the Rx producers of those VFs [mostly]. As PF would soon support allocating CIDs for VFs instead of having a static CID<->queue configuration for them, we'll need to have an additional legacy mode since existing VFs would need to continue on using the older mode of operation. Change the infrastrucutre so that the legacy would be able to indicate which of the legacy behaviors is needed for a given VF. Signed-off-by:
-
Mintz, Yuval authored
When a queue-cid is allocated, assign an index inside that's CID's queue-zone. For PFs and VFS, this number is going to be unique and derive from a per-queue-zone bitmap, while for PF's VFs queues the number is currently going to constant; Later, we'd add the capability of a VF to communicate such an index to its PF. Signed-off-by:
-
Mintz, Yuval authored
We're going to need additional information for queue-cids that a PF creates for its VFs, so start by refactoring existing logic used for initializing said struct into receiving a structure encapsulating the VF-specific information that needs to be provided. This also introduces QED_QUEUE_CID_SELF - each queue-cid would hold an indication to whether it belongs to the hw-function holding it [whether that's a PF or a VF], or else what's the VF id it belongs to. Signed-off-by:
-
Mintz, Yuval authored
Part of an effort of a cleaner seperation between qed and the protocol drivers, the L2 interface is to use the SB structure for initialization purposes opaquely. Signed-off-by:
-
Mintz, Yuval authored
First step in allowing a single PF/VF to open multiple queues on the same queue zone is to add per-hwfn database of queue-cids as a two-dimensional array where entry would be according to [queue zone][internal index]. Signed-off-by:
-
Mintz, Yuval authored
Each PF has a bitmap for its own ranges of CIDs, to allow easy grabbing of an available CID when such is needed. But VFs are not using the same mechanism, instead relying on hard-coded CIDs [ queue-index == cid ]. As an infrastructure step toward increasing number of CIDs of VFs, the PF is going to maintain bitmaps for the VF CIDs as well - the bitmaps would be per-VF and the ranges would be the same [in HW all VFs of a given PF have the same mapping of CIDs, and the HW is capable of distinguishing between those according to the VF index] Signed-off-by:
-
David S. Miller authored
Jason A. Donenfeld says: ==================== net: Avoiding stack overflow in skb_to_sgvec The recent bug with macsec and historical one with virtio have indicated that letting skb_to_sgvec trounce all over an sglist without checking the length is probably a bad idea. And it's not necessary either: an sglist already explicitly marks its last item, and the initialization functions are diligent in doing so. Thus there's a clear way of avoiding future overflows. So, this patchset, from a high level, makes skb_to_sgvec return a potential error code, and then adjusts all callers to check for the error code. There are two situations in which skb_to_sgvec might return such an error: 1) When the passed in sglist is too small; and 2) When the passed in skbuff is too deeply nested. So, the first patch in this series handles the issues with skb_to_sgvec directly, and the remaining ones then handle the call sites. ==================== Signed-off-by: -
Jason A. Donenfeld authored
Signed-off-by:
Jason A. Donenfeld <Jason@zx2c4.com> Reviewed-by:
Sergei Shtylyov <sergei.shtylyov@cogentembedded.com> Cc: "Michael S. Tsirkin" <mst@redhat.com> Cc: Jason Wang <jasowang@redhat.com> Signed-off-by:
-
Jason A. Donenfeld authored
Signed-off-by:
-
Jason A. Donenfeld authored
Signed-off-by:
-
Jason A. Donenfeld authored
Signed-off-by:
-
Jason A. Donenfeld authored
This is a defense-in-depth measure in response to bugs like 4d6fa57b ("macsec: avoid heap overflow in skb_to_sgvec"). There's not only a potential overflow of sglist items, but also a stack overflow potential, so we fix this by limiting the amount of recursion this function is allowed to do. Not actually providing a bounded base case is a future disaster that we can easily avoid here. As a small matter of house keeping, we take this opportunity to move the documentation comment over the actual function the documentation is for. While this could be implemented by using an explicit stack of skbuffs, when implementing this, the function complexity increased considerably, and I don't think such complexity and bloat is actually worth it. So, instead I built this and tested it on x86, x86_64, ARM, ARM64, and MIPS, and measured the stack usage there. I also reverted the recent MIPS changes that give it a separate IRQ stack, so that I could experience some worst-case situations. I found that limiting it to 24 layers deep yielded a good stack usage with room for safety, as well as being much deeper than any driver actually ever creates. Signed-off-by:
-
David S. Miller authored
Merge branch 'bpf-Add-BPF-support-to-all-perf_event' Alexei Starovoitov says: ==================== bpf: Add BPF support to all perf_event v3->v4: one more tweak to reject unsupported events at map update time as Peter suggested v2->v3: more refactoring to address Peter's feedback. Now all perf_events are attachable and readable v1->v2: address Peter's feedback. Refactor patch 1 to allow attaching bpf programs to all event types and reading counters from all of them as well patch 2 - more tests patch 3 - address Dave's feedback and document bpf_perf_event_read() and bpf_perf_event_output() properly ==================== Signed-off-by: -
Teng Qin authored
This commit updates documentation of the bpf_perf_event_output and bpf_perf_event_read helpers to match their implementation. Signed-off-by:
Teng Qin <qinteng@fb.com> Signed-off-by:
Alexei Starovoitov <ast@kernel.org> Signed-off-by:
-
Teng Qin authored
$ trace_event tests attaching BPF program to HW_CPU_CYCLES, SW_CPU_CLOCK, HW_CACHE_L1D and other events. It runs 'dd' in the background while bpf program collects user and kernel stack trace on counter overflow. User space expects to see sys_read and sys_write in the kernel stack. $ tracex6 tests reading of various perf counters from BPF program. Both tests were refactored to increase coverage and be more accurate. Signed-off-by:
-
Alexei Starovoitov authored
Allow BPF_PROG_TYPE_PERF_EVENT program types to attach to all perf_event types, including HW_CACHE, RAW, and dynamic pmu events. Only tracepoint/kprobe events are treated differently which require BPF_PROG_TYPE_TRACEPOINT/BPF_PROG_TYPE_KPROBE program types accordingly. Also add support for reading all event counters using bpf_perf_event_read() helper. Signed-off-by:
-
Sowmini Varadhan authored
The command # arp -s 62.2.0.1 a:b:c:d:e:f dev eth2 adds an entry like the following (listed by "arp -an") ? (62.2.0.1) at 0a:0b:0c:0d:0e:0f [ether] PERM on eth2 but the symmetric deletion command # arp -i eth2 -d 62.2.0.1 does not remove the PERM entry from the table, and instead leaves behind ? (62.2.0.1) at <incomplete> on eth2 The reason is that there is a refcnt of 1 for the arp_tbl itself (neigh_alloc starts off the entry with a refcnt of 1), thus the neigh_release() call from arp_invalidate() will (at best) just decrement the ref to 1, but will never actually free it from the table. To fix this, we need to do something like neigh_forced_gc: if the refcnt is 1 (i.e., on the table's ref), remove the entry from the table and free it. This patch refactors and shares common code between neigh_forced_gc and the newly added neigh_remove_one. A similar issue exists for IPv6 Neighbor Cache entries, and is fixed in a similar manner by this patch. Signed-off-by:
Sowmini Varadhan <sowmini.varadhan@oracle.com> Reviewed-by:
Julian Anastasov <ja@ssi.bg> Signed-off-by:
-
Andrew Lunn authored
Most of the PHYs supported by the SMSC driver have a counter of symbol errors. This is 16 bit wide and wraps around when it reaches its maximum value. Signed-off-by:
Andrew Lunn <andrew@lunn.ch> Reviewed-by:
Florian Fainelli <f.fainelli@gmail.com> Reviewed-By:
Woojung Huh <Woojung.Huh@microchip.com> Signed-off-by:
-
David S. Miller authored
Andrew Lunn says: ==================== dsa: Fixes for mv88e6161 Testing a board with an mv88e6161 turned up two issues. The PHYs were not found, because the wrong method to access them was used. The statistics did not work, because the wrong snapshot method was used ==================== Signed-off-by: -
Andrew Lunn authored
The mv88e6161 was using the wrong method to perform statistics snapshot. Signed-off-by:
Vivien Didelot <vivien.didelot@savoirfairelinux.com> Signed-off-by:
-
Andrew Lunn authored
Access to the internal PHYs of the 6161 and 6123 go through global 2 SMI registers. Fix the ops structure. Signed-off-by:
-
David S. Miller authored
Vivien Didelot says: ==================== net: dsa: mv88e6xxx: move registers macros This patchset brings no functional changes. It is the first step of a cleanup renaming the chip header file and moving the Register definitions _as is_ in their proper header files. A following patchset will prefix them with the appropriate model (MV88E6XXX_ or e.g. MV88E6390_) to respect an implicit namespace and easily identify model subtleties in registers layout, as correctly done in the newly added serdes.h header. ==================== Reviewed-by:
-
Vivien Didelot authored
Move the GLOBAL2_* macros where they belong, in the related global2.h header. Signed-off-by:
-
Vivien Didelot authored
Move the GLOBAL_* macros where they belong, in the related global1.h header. Include it in global2.c which uses GLOBAL_STATUS_IRQ_DEVICE. Signed-off-by:
-
Vivien Didelot authored
Move the PORT_* macros where they belong, in the related port.h header. Signed-off-by:
-
Vivien Didelot authored
Move the PHY_* macros where they belong, in the related phy.h header. Signed-off-by:
-
