Michael S. Tsirkin
2022-Sep-23 06:02 UTC
[PATCH v3] vdpa/ifcvf: fix the calculation of queuepair
On Fri, Sep 23, 2022 at 01:43:36PM +0800, Zhu, Lingshan wrote:> > > On 9/23/2022 1:32 PM, Michael S. Tsirkin wrote: > > On Fri, Sep 23, 2022 at 11:21:41AM +0800, Zhu, Lingshan wrote: > > > > > > On 9/23/2022 9:03 AM, Angus Chen wrote: > > > > The queuepair should be divided by 2 and > > > > it should not be hw->nr_vring when multi-queue feature was enabled > > > This is not about MQ, even no MQ, there are still space in the cap storing > > > (though invalid)queue states that can be addressed. > > > So I think just describe q_pair_id = qid / 2 here. > > hmm without MQ qid is always 0 or 1 so the old calculation > > and the new one give the same result right? > Yes, if no MQ, the queue pair id is always 0. The original code > is buggy anyway, we need the queue pair id = qid /2 even without MQ. > > If there are 8 queues, even no MQ feature bit negotiated, the HW > still stores the queue states though invalid but can be addressed. > > And we are working on a new layout of this LM cap/bar. > > I guess this is what Angus Chen meant. > > Would you mind writing a good commit log for us then? > I think it can be as easy as: > The q_pair_id to address a queue pair in the lm bar should be > calculated by queue_id / 2 than queue_id / nr_vringEnglish requires "rather than" and not "than" here. Otherwise, ok. Angus Chen could you post v4 with this corrected and Lingshan Zhu's ack included please?> Thanks, > Zhu Lingshan > > > > > for the code change: > > > Acked-by: Zhu Lingshan<lingshan.zhu at intel.com> > > > > > > Thanks > > > > Fixes: 2ddae773c93b ("vDPA/ifcvf: detect and use the onboard number of queues directly") > > > > > > > > changes in v2: > > > > - add commitid which to be fixed > > > > - no code changes > > > > > > > > changes in v3: > > > > - updated commit log with more info, addressing comments by MST > > > > - no code changes > > > > > > > > Signed-off-by: Angus Chen <angus.chen at jaguarmicro.com> > > > > --- > > > > drivers/vdpa/ifcvf/ifcvf_base.c | 4 ++-- > > > > 1 file changed, 2 insertions(+), 2 deletions(-) > > > > > > > > diff --git a/drivers/vdpa/ifcvf/ifcvf_base.c b/drivers/vdpa/ifcvf/ifcvf_base.c > > > > index 75a703b803a2..3e4486bfa0b7 100644 > > > > --- a/drivers/vdpa/ifcvf/ifcvf_base.c > > > > +++ b/drivers/vdpa/ifcvf/ifcvf_base.c > > > > @@ -323,7 +323,7 @@ u16 ifcvf_get_vq_state(struct ifcvf_hw *hw, u16 qid) > > > > u32 q_pair_id; > > > > ifcvf_lm = (struct ifcvf_lm_cfg __iomem *)hw->lm_cfg; > > > > - q_pair_id = qid / hw->nr_vring; > > > > + q_pair_id = qid / 2; > > > > avail_idx_addr = &ifcvf_lm->vring_lm_cfg[q_pair_id].idx_addr[qid % 2]; > > > > last_avail_idx = vp_ioread16(avail_idx_addr); > > > > @@ -337,7 +337,7 @@ int ifcvf_set_vq_state(struct ifcvf_hw *hw, u16 qid, u16 num) > > > > u32 q_pair_id; > > > > ifcvf_lm = (struct ifcvf_lm_cfg __iomem *)hw->lm_cfg; > > > > - q_pair_id = qid / hw->nr_vring; > > > > + q_pair_id = qid / 2; > > > > avail_idx_addr = &ifcvf_lm->vring_lm_cfg[q_pair_id].idx_addr[qid % 2]; > > > > hw->vring[qid].last_avail_idx = num; > > > > vp_iowrite16(num, avail_idx_addr);