On Thu, Jun 13, 2024 at 01:03:00PM +0200, Alexander Lobakin wrote:
> From: Simon Horman <ho...@kernel.org>
> 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 <aleksander.loba...@intel.com>
> >
> > ...
> >
> >> 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?
Thanks, and sorry for missing the points you highlight above.
I agree that this is a false positive.
>
> >
> >> + q->stash = &stashes[j];
> >> + hash_init(q->stash->sched_buf_hash);
> >> + }
> >> +
> >> + idpf_queue_set(FLOW_SCH_EN, q);
>
> Thanks,
> Olek
>
