From: Simon Horman <[email protected]> Date: Sat, 1 Jun 2024 09:53:08 +0100
> On Tue, May 28, 2024 at 03:48:37PM +0200, Alexander Lobakin wrote: >> Currently, sizeof(struct idpf_queue) is 32 Kb. >> This is due to the 12-bit hashtable declaration at the end of the queue. >> This HT is needed only for Tx queues when the flow scheduling mode is >> enabled. But &idpf_queue is unified for all of the queue types, >> provoking excessive memory usage. >> The unified structure in general makes the code less effective via >> suboptimal fields placement. You can't avoid that unless you make unions >> each 2 fields. Even then, different field alignment etc., doesn't allow >> you to optimize things to the limit. >> Split &idpf_queue into 4 structures corresponding to the queue types: >> RQ (Rx queue), SQ (Tx queue), FQ (buffer queue), and CQ (completion >> queue). Place only needed fields there and shortcuts handy for hotpath. >> Allocate the abovementioned hashtable dynamically and only when needed, >> keeping &idpf_tx_queue relatively short (192 bytes, same as Rx). This HT >> is used only for OOO completions, which aren't really hotpath anyway. >> Note that this change must be done atomically, otherwise it's really >> easy to get lost and miss something. >> >> Signed-off-by: Alexander Lobakin <[email protected]> > > ... > >> diff --git a/drivers/net/ethernet/intel/idpf/idpf_txrx.c >> b/drivers/net/ethernet/intel/idpf/idpf_txrx.c > > ... > >> @@ -1158,20 +1325,22 @@ static void idpf_rxq_set_descids(struct idpf_vport >> *vport, struct idpf_queue *q) >> */ >> static int idpf_txq_group_alloc(struct idpf_vport *vport, u16 num_txq) >> { >> - bool flow_sch_en; >> - int err, i; >> + bool split, flow_sch_en; >> + int i; >> >> vport->txq_grps = kcalloc(vport->num_txq_grp, >> sizeof(*vport->txq_grps), GFP_KERNEL); >> if (!vport->txq_grps) >> return -ENOMEM; >> >> + split = idpf_is_queue_model_split(vport->txq_model); >> flow_sch_en = !idpf_is_cap_ena(vport->adapter, IDPF_OTHER_CAPS, >> VIRTCHNL2_CAP_SPLITQ_QSCHED); >> >> for (i = 0; i < vport->num_txq_grp; i++) { >> struct idpf_txq_group *tx_qgrp = &vport->txq_grps[i]; >> struct idpf_adapter *adapter = vport->adapter; >> + struct idpf_txq_stash *stashes; >> int j; >> >> tx_qgrp->vport = vport; >> @@ -1180,45 +1349,62 @@ static int idpf_txq_group_alloc(struct idpf_vport >> *vport, u16 num_txq) >> for (j = 0; j < tx_qgrp->num_txq; j++) { >> tx_qgrp->txqs[j] = kzalloc(sizeof(*tx_qgrp->txqs[j]), >> GFP_KERNEL); >> - if (!tx_qgrp->txqs[j]) { >> - err = -ENOMEM; >> + if (!tx_qgrp->txqs[j]) >> goto err_alloc; >> - } >> + } >> + >> + if (split && flow_sch_en) { >> + stashes = kcalloc(num_txq, sizeof(*stashes), >> + GFP_KERNEL); > > Hi Alexander, > > Here stashes is assigned a memory allocation and > then then assigned to tx_qgrp->stashes a few lines below... > >> + if (!stashes) >> + goto err_alloc; >> + >> + tx_qgrp->stashes = stashes; >> } >> >> for (j = 0; j < tx_qgrp->num_txq; j++) { >> - struct idpf_queue *q = tx_qgrp->txqs[j]; >> + struct idpf_tx_queue *q = tx_qgrp->txqs[j]; >> >> q->dev = &adapter->pdev->dev; >> q->desc_count = vport->txq_desc_count; >> q->tx_max_bufs = idpf_get_max_tx_bufs(adapter); >> q->tx_min_pkt_len = idpf_get_min_tx_pkt_len(adapter); >> - q->vport = vport; >> + q->netdev = vport->netdev; >> q->txq_grp = tx_qgrp; >> - hash_init(q->sched_buf_hash); >> >> - if (flow_sch_en) >> - set_bit(__IDPF_Q_FLOW_SCH_EN, q->flags); >> + if (!split) { >> + q->clean_budget = vport->compln_clean_budget; >> + idpf_queue_assign(CRC_EN, q, >> + vport->crc_enable); >> + } >> + >> + if (!flow_sch_en) >> + continue; >> + >> + if (split) { > > ... but here elements of stashes seem to be assigned to q->stash > without stashes having being initialised. > > Flagged by Smatch Hi! Yes, I saw the report, but isn't it a false positive? Allocation happens when `split && flow_sch_en`, and here we have if (!flow_sch_en) continue; if (split) // assign IOW the assignment can't happen without the allocation? > >> + q->stash = &stashes[j]; >> + hash_init(q->stash->sched_buf_hash); >> + } >> + >> + idpf_queue_set(FLOW_SCH_EN, q); Thanks, Olek
