On 2024-05-10 08:23, Andy Gospodarek wrote: > On Fri, May 10, 2024 at 11:35:35AM +0100, Vadim Fedorenko wrote: >> On 10.05.2024 04:55, Ajit Khaparde wrote: >>> On Thu, May 9, 2024 at 2:50 PM Vadim Fedorenko >>> <vadim.fedorenko@xxxxxxxxx> wrote: >>>> >>>> On 09/05/2024 17:27, Wei Huang wrote: >>>>> From: Manoj Panicker <manoj.panicker2@xxxxxxx> >>>>> >>>>> As a usage example, this patch implements TPH support in Broadcom BNXT >>>>> device driver by invoking pcie_tph_set_st() function when interrupt >>>>> affinity is changed. >>>>> >>>>> Reviewed-by: Ajit Khaparde <ajit.khaparde@xxxxxxxxxxxx> >>>>> Reviewed-by: Andy Gospodarek <andrew.gospodarek@xxxxxxxxxxxx> >>>>> Reviewed-by: Wei Huang <wei.huang2@xxxxxxx> >>>>> Signed-off-by: Manoj Panicker <manoj.panicker2@xxxxxxx> >>>>> --- >>>>> drivers/net/ethernet/broadcom/bnxt/bnxt.c | 51 +++++++++++++++++++++++ >>>>> drivers/net/ethernet/broadcom/bnxt/bnxt.h | 4 ++ >>>>> 2 files changed, 55 insertions(+) >>>>> >>>>> diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.c b/drivers/net/ethernet/broadcom/bnxt/bnxt.c >>>>> index 2c2ee79c4d77..be9c17566fb4 100644 >>>>> --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.c >>>>> +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.c >>>>> @@ -55,6 +55,7 @@ >>>>> #include <net/page_pool/helpers.h> >>>>> #include <linux/align.h> >>>>> #include <net/netdev_queues.h> >>>>> +#include <linux/pci-tph.h> >>>>> >>>>> #include "bnxt_hsi.h" >>>>> #include "bnxt.h" >>>>> @@ -10491,6 +10492,7 @@ static void bnxt_free_irq(struct bnxt *bp) >>>>> free_cpumask_var(irq->cpu_mask); >>>>> irq->have_cpumask = 0; >>>>> } >>>>> + irq_set_affinity_notifier(irq->vector, NULL); >>>>> free_irq(irq->vector, bp->bnapi[i]); >>>>> } >>>>> >>>>> @@ -10498,6 +10500,45 @@ static void bnxt_free_irq(struct bnxt *bp) >>>>> } >>>>> } >>>>> >>>>> +static void bnxt_rtnl_lock_sp(struct bnxt *bp); >>>>> +static void bnxt_rtnl_unlock_sp(struct bnxt *bp); >>>>> +static void bnxt_irq_affinity_notify(struct irq_affinity_notify *notify, >>>>> + const cpumask_t *mask) >>>>> +{ >>>>> + struct bnxt_irq *irq; >>>>> + >>>>> + irq = container_of(notify, struct bnxt_irq, affinity_notify); >>>>> + cpumask_copy(irq->cpu_mask, mask); >>>>> + >>>>> + if (!pcie_tph_set_st(irq->bp->pdev, irq->msix_nr, >>>>> + cpumask_first(irq->cpu_mask), >>>>> + TPH_MEM_TYPE_VM, PCI_TPH_REQ_TPH_ONLY)) >>>>> + pr_err("error in configuring steering tag\n"); >>>>> + >>>>> + if (netif_running(irq->bp->dev)) { >>>>> + rtnl_lock(); >>>>> + bnxt_close_nic(irq->bp, false, false); >>>>> + bnxt_open_nic(irq->bp, false, false); >>>>> + rtnl_unlock(); >>>>> + } >>>> >>>> Is it really needed? It will cause link flap and pause in the traffic >>>> service for the device. Why the device needs full restart in this case? >>> >>> In that sequence only the rings are recreated for the hardware to sync >>> up the tags. >>> >>> Actually its not a full restart. There is no link reinit or other >>> heavy lifting in this sequence. >>> The pause in traffic may be momentary. Do IRQ/CPU affinities change frequently? >>> Probably not? >> >> From what I can see in bnxt_en, proper validation of link_re_init parameter is >> not (yet?) implemented, __bnxt_open_nic will unconditionally call >> netif_carrier_off() which will be treated as loss of carrier with counters >> increment and proper events posted. Changes to CPU affinities were >> non-disruptive before the patch, but now it may break user-space >> assumptions. > > From my testing the link should not flap. I just fired up a recent net-next > and confirmed the same by calling $ ethtool -G ens7f0np0 rx 1024 which does a > similar bnxt_close_nic(bp, false, false)/bnxt_open_nic(bp, false, false) as > this patch. Link remained up -- even with a non-Broadocm link-partner. > >> Does FW need full rings re-init to update target value, which is one u32 write? >> It looks like overkill TBH. > > Full rings do not, but the initialization of that particular ring associated > with this irq does need to be done. On my list of things we need to do in > bnxt_en is implement the new ndo_queue_stop/start and ndo_queue_mem_alloc/free > operations and once those are done we could make a switch as that may be less > disruptive. Hi Andy, I have an implementation of the new ndo_queue_stop/start() API [1] and would appreciate comments. I've been trying to test it but without avail due to FW issues. [1]: https://lore.kernel.org/netdev/20240502045410.3524155-1-dw@xxxxxxxxxxx/ > >> And yes, affinities can be change on fly according to the changes of the >> workload on the host. >> >>>> >>>> >>>>> +} >>>>> + >>>>> +static void bnxt_irq_affinity_release(struct kref __always_unused *ref) >>>>> +{ >>>>> +} >>>>> + >>>>> +static inline void __bnxt_register_notify_irqchanges(struct bnxt_irq *irq) >>>> >>>> No inlines in .c files, please. Let compiler decide what to inline. >>>> >>>>> +{ >>>>> + struct irq_affinity_notify *notify; >>>>> + >>>>> + notify = &irq->affinity_notify; >>>>> + notify->irq = irq->vector; >>>>> + notify->notify = bnxt_irq_affinity_notify; >>>>> + notify->release = bnxt_irq_affinity_release; >>>>> + >>>>> + irq_set_affinity_notifier(irq->vector, notify); >>>>> +} >>>>> + >>>>> static int bnxt_request_irq(struct bnxt *bp) >>>>> { >>>>> int i, j, rc = 0; >>>>> @@ -10543,6 +10584,7 @@ static int bnxt_request_irq(struct bnxt *bp) >>>>> int numa_node = dev_to_node(&bp->pdev->dev); >>>>> >>>>> irq->have_cpumask = 1; >>>>> + irq->msix_nr = map_idx; >>>>> cpumask_set_cpu(cpumask_local_spread(i, numa_node), >>>>> irq->cpu_mask); >>>>> rc = irq_set_affinity_hint(irq->vector, irq->cpu_mask); >>>>> @@ -10552,6 +10594,15 @@ static int bnxt_request_irq(struct bnxt *bp) >>>>> irq->vector); >>>>> break; >>>>> } >>>>> + >>>>> + if (!pcie_tph_set_st(bp->pdev, i, >>>>> + cpumask_first(irq->cpu_mask), >>>>> + TPH_MEM_TYPE_VM, PCI_TPH_REQ_TPH_ONLY)) { >>>>> + netdev_err(bp->dev, "error in setting steering tag\n"); >>>>> + } else { >>>>> + irq->bp = bp; >>>>> + __bnxt_register_notify_irqchanges(irq); >>>>> + } >>>>> } >>>>> } >>>>> return rc; >>>>> diff --git a/drivers/net/ethernet/broadcom/bnxt/bnxt.h b/drivers/net/ethernet/broadcom/bnxt/bnxt.h >>>>> index dd849e715c9b..0d3442590bb4 100644 >>>>> --- a/drivers/net/ethernet/broadcom/bnxt/bnxt.h >>>>> +++ b/drivers/net/ethernet/broadcom/bnxt/bnxt.h >>>>> @@ -1195,6 +1195,10 @@ struct bnxt_irq { >>>>> u8 have_cpumask:1; >>>>> char name[IFNAMSIZ + 2]; >>>>> cpumask_var_t cpu_mask; >>>>> + >>>>> + int msix_nr; >>>>> + struct bnxt *bp; >>>>> + struct irq_affinity_notify affinity_notify; >>>>> }; >>>>> >>>>> #define HWRM_RING_ALLOC_TX 0x1 >>>> >>