Messages in this thread | | | Subject | Re: [PATCH v3] nvme-tcp: Do not reset transport on data digest errors | From | Sagi Grimberg <> | Date | Mon, 30 Aug 2021 14:25:04 +0300 |
| |
On 8/26/21 1:21 AM, Daniel Wagner wrote: > The spec says > > 7.4.6.1 Digest Error handling > > When a host detects a data digest error in a C2HData PDU, that host > shall continue processing C2HData PDUs associated with the command and > when the command processing has completed, if a successful status was > returned by the controller, the host shall fail the command with a > non-fatal transport error. > > Currently the transport is reseted when a data digest error is > detected. Instead, when a digest error is detected, mark the final > status as NVME_SC_DATA_XFER_ERROR and let the upper layer handle > the error. > > In order to keep track of the final result maintain a status field in > nvme_tcp_request object and use it to overwrite the completion queue > status (which might be successful even though a digest error has been > detected) when completing the request. > > Reviewed-by: Hannes Reinecke <hare@suse.de> > Signed-off-by: Daniel Wagner <dwagner@suse.de> > --- > The status member placed so that it fills up a hole in struct > nvme_tcp_request: > > struct nvme_tcp_request { > struct nvme_request req; /* 0 32 */ > void * pdu; /* 32 8 */ > struct nvme_tcp_queue * queue; /* 40 8 */ > u32 data_len; /* 48 4 */ > u32 pdu_len; /* 52 4 */ > u32 pdu_sent; /* 56 4 */ > u16 ttag; /* 60 2 */ > u16 status; /* 62 2 */ > /* --- cacheline 1 boundary (64 bytes) --- */ > struct list_head entry; /* 64 16 */ > struct llist_node lentry; /* 80 8 */ > __le32 ddgst; /* 88 4 */ > > /* XXX 4 bytes hole, try to pack */ > > struct bio * curr_bio; /* 96 8 */ > struct iov_iter iter; /* 104 40 */ > /* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */ > size_t offset; /* 144 8 */ > size_t data_sent; /* 152 8 */ > enum nvme_tcp_send_state state; /* 160 4 */ > > /* size: 168, cachelines: 3, members: 16 */ > /* sum members: 160, holes: 1, sum holes: 4 */ > /* padding: 4 */ > /* last cacheline: 40 bytes */ > }; > > v3: > - initialize req->status in nvme_tcp_setup_cmd_pdu() > - add rb tag from Hannes > > v2: > - https://lore.kernel.org/linux-nvme/20210825124259.28707-1-dwagner@suse.de/ > - moved 'status' from nvme_tcp_queue to nvme_tcp_request. > > v1: > - https://lore.kernel.org/linux-nvme/20210805121541.77613-1-dwagner@suse.de/ > > drivers/nvme/host/tcp.c | 23 +++++++++++++++++++---- > 1 file changed, 19 insertions(+), 4 deletions(-) > > diff --git a/drivers/nvme/host/tcp.c b/drivers/nvme/host/tcp.c > index 645025620154..29ef0f74f620 100644 > --- a/drivers/nvme/host/tcp.c > +++ b/drivers/nvme/host/tcp.c > @@ -45,6 +45,7 @@ struct nvme_tcp_request { > u32 pdu_len; > u32 pdu_sent; > u16 ttag; > + u16 status; > struct list_head entry; > struct llist_node lentry; > __le32 ddgst; > @@ -485,7 +486,9 @@ static void nvme_tcp_error_recovery(struct nvme_ctrl *ctrl) > static int nvme_tcp_process_nvme_cqe(struct nvme_tcp_queue *queue, > struct nvme_completion *cqe) > { > + struct nvme_tcp_request *req; > struct request *rq; > + u16 status; > > rq = nvme_find_rq(nvme_tcp_tagset(queue), cqe->command_id); > if (!rq) { > @@ -496,7 +499,12 @@ static int nvme_tcp_process_nvme_cqe(struct nvme_tcp_queue *queue, > return -EINVAL; > } > > - if (!nvme_try_complete_req(rq, cqe->status, cqe->result)) > + req = blk_mq_rq_to_pdu(rq); > + status = req->status; > + if (status == NVME_SC_SUCCESS) > + status = cqe->status;
Maybe more intuitive to skip the local status variable? /* */ if (req->status == NVME_SC_SUCCESS) req->status = cqe->status;
This way it is always consistent completing with req->status.
| |