| Message ID | 1501704648-20159-13-git-send-email-longli@exchange.microsoft.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
You seem to be doing memory allocations and frees for every packet on the write. At least for other RDMA protocols that would have been a major performance issue. Do you have any performance numbers and/or profiles of the code? -- To unsubscribe from this list: send the line "unsubscribe linux-cifs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
> -----Original Message----- > From: Christoph Hellwig [mailto:hch@infradead.org] > Sent: Sunday, August 13, 2017 3:20 AM > To: Long Li <longli@microsoft.com> > Cc: Steve French <sfrench@samba.org>; linux-cifs@vger.kernel.org; samba- > technical@lists.samba.org; linux-kernel@vger.kernel.org; Long Li > <longli@microsoft.com> > Subject: Re: [[PATCH v1] 12/37] [CIFS] SMBD: Handle send completion from > CQ > > You seem to be doing memory allocations and frees for every packet on the > write. At least for other RDMA protocols that would have been a major > performance issue. The size of SGE array passed to IB is unknown, so I don't know how much to pre-allocate in advance. But it seems this size is not big when passed down from CIFS. I will look at pre-allocating buffer if this is an issue. > > Do you have any performance numbers and/or profiles of the code? I will look into profiling. -- To unsubscribe from this list: send the line "unsubscribe linux-cifs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/fs/cifs/cifsrdma.c b/fs/cifs/cifsrdma.c index 20237b7..ecbc832 100644 --- a/fs/cifs/cifsrdma.c +++ b/fs/cifs/cifsrdma.c @@ -197,6 +197,31 @@ cifs_rdma_qp_async_error_upcall(struct ib_event *event, void *context) } } +/* Called in softirq, when a RDMA send is donea */ +static void send_done(struct ib_cq *cq, struct ib_wc *wc) +{ + int i; + struct cifs_rdma_request *request = + container_of(wc->wr_cqe, struct cifs_rdma_request, cqe); + + log_rdma_send("cifs_rdma_request %p completed wc->status=%d\n", + request, wc->status); + + if (wc->status != IB_WC_SUCCESS || wc->opcode != IB_WC_SEND) { + log_rdma_send("wc->status=%d wc->opcode=%d\n", + wc->status, wc->opcode); + } + + for (i=0; i<request->num_sge; i++) + ib_dma_unmap_single(request->info->id->device, + request->sge[i].addr, + request->sge[i].length, + DMA_TO_DEVICE); + + kfree(request->sge); + mempool_free(request, request->info->request_mempool); +} + /* Called from softirq, when recv is done */ static void recv_done(struct ib_cq *cq, struct ib_wc *wc) {