RE: [PATCHv8 07/11] ib_core: Add API to support IBoE from userspace
If we have a dedicated ABI call for this mapping, then it seems reasonable to have it device independent. However, this mapping is really only used when creating address handles. So, we can base the mapping on the (device specific) create_ah() flow, but provide generic mapping functions for all devices to use (this is kind of what happens now). Also, using create_ah() doesn't introduce an ABI call that is specific to ib--eth mappings. This is similar to how device-specific ib_reg_user_mr() functions call the generic ib_umem_get()... -Original Message- From: linux-rdma-ow...@vger.kernel.org [mailto:linux-rdma-ow...@vger.kernel.org] On Behalf Of Roland Dreier Sent: Thursday, May 13, 2010 10:18 PM To: Sean Hefty Cc: 'Eli Cohen'; Eli Cohen; Linux RDMA list; ewg Subject: Re: [PATCHv8 07/11] ib_core: Add API to support IBoE from userspace Basically, what I want to understand is why does this change make sense? @@ -1139,6 +1139,10 @@ struct ib_device { struct ib_grh *in_grh, struct ib_mad *in_mad, struct ib_mad *out_mad); +int(*get_eth_l2_addr)(struct ib_device *device, u8 port, + union ib_gid *dgid, int sgid_idx, + u8 *mac, u16 *vlan_id, u8 *tagged); + Yes, that was pretty much my original question. Why do we have a verb for userspace to call a device-specific method to do the mapping? The layering seems wrong somewhere if we have a generic verb to do this mapping, but then put the mapping in device-specific code. - R. -- Roland Dreier rola...@cisco.com || For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/index.html -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCHv8 07/11] ib_core: Add API to support IBoE from userspace
On Wed, May 12, 2010 at 01:28:31PM -0700, Roland Dreier wrote: Add ib_uverbs_get_eth_l2_addr() to allow ibv_create_ah() to resolve sgid, dgid to vlan, dmac for any gid type. Although user-space might bypass this call for link-local gids, it is better not to replicate the kernel resolution policy. Port link layer is also returned by ibv_query_port(). A high-level comment/question, followed by some notes about the specific patch. At the highest level, is having this very low-level command exposed as part of the kernel uverbs - userspace API the right place to split things? Making the Ethernet address resolution part of the low-level driver implies that it's not really a generic part of the verbs interface. Correct - Ethernet address resolution really isn't part of the verbs interface accoring to the RoCE specification; it should happen below the verbs. Maybe it is generic, and we should have a generic function instead of calling into the low-level driver. I see the argument that we should keep the policy in the kernel, although I'm not sure how strong that argument is -- once we start shipping a kernel with a certain policy (and I guess OFED has in effect already done that), how could we ever change that policy? We'll have interoperability issues anyway, so it seems having userspace and kernel use different policies doesn't cause much further problems anyway. Address mapping policies depend on the address type. This patch only defines a policy for mapping link-local addresses, and we should indeed take care not to change it (if possible). Later on, we can add more policies for other address types (e.g., normal IPv6 addresses, mapped IPv4 addresses, etc.) Or maybe it is device-specific, and we could wrap it up into the create AH uverbs call we already have? Interesting idea. Not sure what is better here: a seperate ABI call or some additional 'u8 ctx[32]' field in struct ib_uverbs_create_ah_resp that will be interpreted by the hw-specific user-space driver. Low-level comments: +ssize_t ib_uverbs_get_eth_l2_addr(struct ib_uverbs_file *file, const char __user *buf, +int in_len, int out_len) +{ + struct ib_uverbs_get_eth_l2_addr cmd; + struct ib_uverbs_get_eth_l2_addr_resp resp; + int ret; + struct ib_pd*pd; + + if (out_len sizeof resp) + return -ENOSPC; + + if (copy_from_user(cmd, buf, sizeof cmd)) + return -EFAULT; + + pd = idr_read_pd(cmd.pd_handle, file-ucontext); + if (!pd) + return -EINVAL; + + ret = ib_get_eth_l2_addr(pd-device, cmd.port, (union ib_gid *)cmd.gid, + cmd.sgid_idx, resp.mac, resp.vlan_id, resp.tagged); + put_pd_read(pd); + if (!ret) { + if (copy_to_user((void __user *) (unsigned long) cmd.response, + resp, sizeof resp)) + return -EFAULT; This leaks kernel memory contents to userspace since the stack variable resp is never cleared. Also will cause problems if we ever need to use the reserved fields for anything. I see, I missed that subtle detail. Also I'm not sure I understand why you pass the PD into this method? It seems you just use it to get a pointer to the device, but you already have that from the uverbs_file structure that's passed into all commands. True, missed that too. +int ib_get_eth_l2_addr(struct ib_device *device, u8 port, union ib_gid *gid, + int sgid_idx, u8 *mac, __u16 *vlan_id, u8 *tagged) +{ + if (!device-get_eth_l2_addr) + return -ENOSYS; + + return device-get_eth_l2_addr(device, port, gid, sgid_idx, mac, vlan_id, tagged); +} +EXPORT_SYMBOL(ib_get_eth_l2_addr); I don't think we need this wrapper, since uverbs can just call the get_eth_l2_addr method directly (we already do that for quite a few other methods, eg alloc_ucontext is a uverbs-only method that has no in-kernel wrapper). Also the -ENOSYS test isn't necessary, since a device-driver shouldn't allow this method unless it actually implements it. I agree. By the way, do you prefer that I re-create the patch with the fixes or would you make the necessary changes. -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
RE: [PATCHv8 07/11] ib_core: Add API to support IBoE from userspace
Address mapping policies depend on the address type. This patch only defines a policy for mapping link-local addresses, and we should indeed take care not to change it (if possible). Later on, we can add more policies for other address types (e.g., normal IPv6 addresses, mapped IPv4 addresses, etc.) Or maybe it is device-specific, and we could wrap it up into the create AH uverbs call we already have? Interesting idea. Not sure what is better here: a seperate ABI call or some additional 'u8 ctx[32]' field in struct ib_uverbs_create_ah_resp that will be interpreted by the hw-specific user-space driver. I don't understand why mapping remote addresses should be driver specific. -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCHv8 07/11] ib_core: Add API to support IBoE from userspace
On Thu, May 13, 2010 at 08:55:13AM -0700, Sean Hefty wrote: Interesting idea. Not sure what is better here: a seperate ABI call or some additional 'u8 ctx[32]' field in struct ib_uverbs_create_ah_resp that will be interpreted by the hw-specific user-space driver. I don't understand why mapping remote addresses should be driver specific. In this case it will not be just mapping remote addresses but creating the required AH information which is unique to each device. -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
RE: [PATCHv8 07/11] ib_core: Add API to support IBoE from userspace
In this case it will not be just mapping remote addresses but creating the required AH information which is unique to each device. I understand that the AH is per device. What I don't get is why we would want each device to perform the mapping. We don't expect the device to map GIDs to LIDs when creating an AH. The user must have done the mapping beforehand. The mapping does not depend on which device I'm using, it should be the same. Basically, what I want to understand is why does this change make sense? @@ -1139,6 +1139,10 @@ struct ib_device { struct ib_grh *in_grh, struct ib_mad *in_mad, struct ib_mad *out_mad); + int(*get_eth_l2_addr)(struct ib_device *device, u8 port, + union ib_gid *dgid, int sgid_idx, + u8 *mac, u16 *vlan_id, u8 *tagged); + - Sean -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
Re: [PATCHv8 07/11] ib_core: Add API to support IBoE from userspace
Basically, what I want to understand is why does this change make sense? @@ -1139,6 +1139,10 @@ struct ib_device { struct ib_grh *in_grh, struct ib_mad *in_mad, struct ib_mad *out_mad); +int(*get_eth_l2_addr)(struct ib_device *device, u8 port, + union ib_gid *dgid, int sgid_idx, + u8 *mac, u16 *vlan_id, u8 *tagged); + Yes, that was pretty much my original question. Why do we have a verb for userspace to call a device-specific method to do the mapping? The layering seems wrong somewhere if we have a generic verb to do this mapping, but then put the mapping in device-specific code. - R. -- Roland Dreier rola...@cisco.com || For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/index.html -- To unsubscribe from this list: send the line unsubscribe linux-rdma in the body of a message to majord...@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html