| Message ID | 20231220085424.97407-2-chengyou@linux.alibaba.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | RDMA/erdma: Introduce hardware statistics support | expand |
On Wed, Dec 20, 2023 at 04:54:23PM +0800, Cheng Xu wrote: > Hardware response, such as the result of query statistics, may be too > long to be directly accommodated within the CQE structure. To address > this, we introduce a DMA pool to hold the hardware's responses of CMDQ > requests. > > Signed-off-by: Cheng Xu <chengyou@linux.alibaba.com> > --- > drivers/infiniband/hw/erdma/erdma.h | 2 ++ > drivers/infiniband/hw/erdma/erdma_main.c | 38 ++++++++++++++++++++++-- > 2 files changed, 38 insertions(+), 2 deletions(-) > > diff --git a/drivers/infiniband/hw/erdma/erdma.h b/drivers/infiniband/hw/erdma/erdma.h > index f190111840e9..5df401a30cb9 100644 > --- a/drivers/infiniband/hw/erdma/erdma.h > +++ b/drivers/infiniband/hw/erdma/erdma.h > @@ -212,6 +212,8 @@ struct erdma_dev { > > atomic_t num_ctx; > struct list_head cep_list; > + > + struct dma_pool *resp_pool; > }; > > static inline void *get_queue_entry(void *qbuf, u32 idx, u32 depth, u32 shift) > diff --git a/drivers/infiniband/hw/erdma/erdma_main.c b/drivers/infiniband/hw/erdma/erdma_main.c > index 0880c79a978c..541e77aea494 100644 > --- a/drivers/infiniband/hw/erdma/erdma_main.c > +++ b/drivers/infiniband/hw/erdma/erdma_main.c > @@ -168,18 +168,48 @@ static void erdma_comm_irq_uninit(struct erdma_dev *dev) > free_irq(dev->comm_irq.msix_vector, dev); > } > > +static int erdma_dma_pools_init(struct erdma_dev *dev) > +{ > + dev->resp_pool = dma_pool_create("erdma_resp_pool", &dev->pdev->dev, > + ERDMA_HW_RESP_SIZE, ERDMA_HW_RESP_SIZE, > + 0); > + if (!dev->resp_pool) > + return -ENOMEM; > + > + return 0; > +} > + > +static void erdma_dma_pools_destroy(struct erdma_dev *dev) > +{ > + dma_pool_destroy(dev->resp_pool); > +} Please don't add extra layer of functions which will be called in same file anyway. Call directly to dma_pool_destroy(), same goes for dma_pool_create(). Thanks
On 12/20/23 6:12 PM, Leon Romanovsky wrote: > On Wed, Dec 20, 2023 at 04:54:23PM +0800, Cheng Xu wrote: >> Hardware response, such as the result of query statistics, may be too >> long to be directly accommodated within the CQE structure. To address >> this, we introduce a DMA pool to hold the hardware's responses of CMDQ >> requests. >> >> Signed-off-by: Cheng Xu <chengyou@linux.alibaba.com> >> --- >> drivers/infiniband/hw/erdma/erdma.h | 2 ++ >> drivers/infiniband/hw/erdma/erdma_main.c | 38 ++++++++++++++++++++++-- >> 2 files changed, 38 insertions(+), 2 deletions(-) >> <...> >> +static int erdma_dma_pools_init(struct erdma_dev *dev) >> +{ >> + dev->resp_pool = dma_pool_create("erdma_resp_pool", &dev->pdev->dev, >> + ERDMA_HW_RESP_SIZE, ERDMA_HW_RESP_SIZE, >> + 0); >> + if (!dev->resp_pool) >> + return -ENOMEM; >> + >> + return 0; >> +} >> + >> +static void erdma_dma_pools_destroy(struct erdma_dev *dev) >> +{ >> + dma_pool_destroy(dev->resp_pool); >> +} > > Please don't add extra layer of functions which will be called in same > file anyway. Call directly to dma_pool_destroy(), same goes for dma_pool_create(). Get it. Will fix in v2. Thanks, Cheng Xu > Thanks
diff --git a/drivers/infiniband/hw/erdma/erdma.h b/drivers/infiniband/hw/erdma/erdma.h index f190111840e9..5df401a30cb9 100644 --- a/drivers/infiniband/hw/erdma/erdma.h +++ b/drivers/infiniband/hw/erdma/erdma.h @@ -212,6 +212,8 @@ struct erdma_dev { atomic_t num_ctx; struct list_head cep_list; + + struct dma_pool *resp_pool; }; static inline void *get_queue_entry(void *qbuf, u32 idx, u32 depth, u32 shift) diff --git a/drivers/infiniband/hw/erdma/erdma_main.c b/drivers/infiniband/hw/erdma/erdma_main.c index 0880c79a978c..541e77aea494 100644 --- a/drivers/infiniband/hw/erdma/erdma_main.c +++ b/drivers/infiniband/hw/erdma/erdma_main.c @@ -168,18 +168,48 @@ static void erdma_comm_irq_uninit(struct erdma_dev *dev) free_irq(dev->comm_irq.msix_vector, dev); } +static int erdma_dma_pools_init(struct erdma_dev *dev) +{ + dev->resp_pool = dma_pool_create("erdma_resp_pool", &dev->pdev->dev, + ERDMA_HW_RESP_SIZE, ERDMA_HW_RESP_SIZE, + 0); + if (!dev->resp_pool) + return -ENOMEM; + + return 0; +} + +static void erdma_dma_pools_destroy(struct erdma_dev *dev) +{ + dma_pool_destroy(dev->resp_pool); +} + static int erdma_device_init(struct erdma_dev *dev, struct pci_dev *pdev) { int ret; + ret = erdma_dma_pools_init(dev); + if (ret) + return ret; + ret = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(ERDMA_PCI_WIDTH)); if (ret) - return ret; + goto destroy_pool; dma_set_max_seg_size(&pdev->dev, UINT_MAX); return 0; + +destroy_pool: + erdma_dma_pools_destroy(dev); + + return ret; +} + +static void erdma_device_uninit(struct erdma_dev *dev) +{ + erdma_dma_pools_destroy(dev); } static void erdma_hw_reset(struct erdma_dev *dev) @@ -273,7 +303,7 @@ static int erdma_probe_dev(struct pci_dev *pdev) err = erdma_request_vectors(dev); if (err) - goto err_iounmap_func_bar; + goto err_uninit_device; err = erdma_comm_irq_init(dev); if (err) @@ -314,6 +344,9 @@ static int erdma_probe_dev(struct pci_dev *pdev) err_free_vectors: pci_free_irq_vectors(dev->pdev); +err_uninit_device: + erdma_device_uninit(dev); + err_iounmap_func_bar: devm_iounmap(&pdev->dev, dev->func_bar); @@ -339,6 +372,7 @@ static void erdma_remove_dev(struct pci_dev *pdev) erdma_aeq_destroy(dev); erdma_comm_irq_uninit(dev); pci_free_irq_vectors(dev->pdev); + erdma_device_uninit(dev); devm_iounmap(&pdev->dev, dev->func_bar); pci_release_selected_regions(pdev, ERDMA_BAR_MASK);
Hardware response, such as the result of query statistics, may be too long to be directly accommodated within the CQE structure. To address this, we introduce a DMA pool to hold the hardware's responses of CMDQ requests. Signed-off-by: Cheng Xu <chengyou@linux.alibaba.com> --- drivers/infiniband/hw/erdma/erdma.h | 2 ++ drivers/infiniband/hw/erdma/erdma_main.c | 38 ++++++++++++++++++++++-- 2 files changed, 38 insertions(+), 2 deletions(-)