Re: [PATCH v2 12/14] net: axienet: Upgrade descriptors to hold 64-bit addresses

From: Robert Hancock
Date: Mon Mar 09 2020 - 15:00:55 EST

Next message: Stephen Boyd: "Re: [PATCH v4 1/4] dt-bindings: Introduce SoC sleep stats bindings"
Previous message: Michael Walle: "Re: [PATCH 0/6] NXP DSPI bugfixes and support for LS1028A"
In reply to: Andre Przywara: "[PATCH v2 12/14] net: axienet: Upgrade descriptors to hold 64-bit addresses"
Next in thread: Andre Przywara: "Re: [PATCH v2 12/14] net: axienet: Upgrade descriptors to hold 64-bit addresses"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

On 2020-03-09 12:18 p.m., Andre Przywara wrote:
> Newer revisions of the AXI DMA IP (>= v7.1) support 64-bit addresses,
> both for the descriptors itself, as well as for the buffers they are
> pointing to.
> This is realised by adding "MSB" words for the next and phys pointer
> right behind the existing address word, now named "LSB". These MSB words
> live in formerly reserved areas of the descriptor.
>
> If the hardware supports it, write both words when setting an address.
> The buffer address is handled by two wrapper functions, the two
> occasions where we set the next pointers are open coded.
>
> For now this is guarded by a flag which we don't set yet.
>
> Signed-off-by: Andre Przywara <andre.przywara@xxxxxxx>
> ---
> drivers/net/ethernet/xilinx/xilinx_axienet.h | 9 +-
> .../net/ethernet/xilinx/xilinx_axienet_main.c | 113 ++++++++++++------
> 2 files changed, 83 insertions(+), 39 deletions(-)
>
> diff --git a/drivers/net/ethernet/xilinx/xilinx_axienet.h b/drivers/net/ethernet/xilinx/xilinx_axienet.h
> index fb7450ca5c53..84c4c3655516 100644
> --- a/drivers/net/ethernet/xilinx/xilinx_axienet.h
> +++ b/drivers/net/ethernet/xilinx/xilinx_axienet.h
> @@ -328,6 +328,7 @@
> #define XAE_FEATURE_PARTIAL_TX_CSUM (1 << 1)
> #define XAE_FEATURE_FULL_RX_CSUM (1 << 2)
> #define XAE_FEATURE_FULL_TX_CSUM (1 << 3)
> +#define XAE_FEATURE_DMA_64BIT (1 << 4)
>
> #define XAE_NO_CSUM_OFFLOAD 0
>
> @@ -340,9 +341,9 @@
> /**
> * struct axidma_bd - Axi Dma buffer descriptor layout
> * @next: MM2S/S2MM Next Descriptor Pointer
> - * @reserved1: Reserved and not used
> + * @next_msb: MM2S/S2MM Next Descriptor Pointer (high 32 bits)
> * @phys: MM2S/S2MM Buffer Address
> - * @reserved2: Reserved and not used
> + * @phys_msb: MM2S/S2MM Buffer Address (high 32 bits)
> * @reserved3: Reserved and not used
> * @reserved4: Reserved and not used
> * @cntrl: MM2S/S2MM Control value
> @@ -355,9 +356,9 @@
> */
> struct axidma_bd {
> u32 next; /* Physical address of next buffer descriptor */
> - u32 reserved1;
> + u32 next_msb; /* high 32 bits for IP >= v7.1, reserved on older IP */
> u32 phys;
> - u32 reserved2;
> + u32 phys_msb; /* for IP >= v7.1, reserved for older IP */
> u32 reserved3;
> u32 reserved4;
> u32 cntrl;
> diff --git a/drivers/net/ethernet/xilinx/xilinx_axienet_main.c b/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
> index ea44ef4cf288..edee0666d52c 100644
> --- a/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
> +++ b/drivers/net/ethernet/xilinx/xilinx_axienet_main.c
> @@ -153,6 +153,25 @@ static void axienet_dma_out_addr(struct axienet_local *lp, off_t reg,
> axienet_dma_out32(lp, reg, lower_32_bits(addr));
> }
>
> +static void desc_set_phys_addr(struct axienet_local *lp, dma_addr_t addr,
> + struct axidma_bd *desc)
> +{
> + desc->phys = lower_32_bits(addr);
> + if (lp->features & XAE_FEATURE_DMA_64BIT)
> + desc->phys_msb = upper_32_bits(addr);
> +}
> +
> +static dma_addr_t desc_get_phys_addr(struct axienet_local *lp,
> + struct axidma_bd *desc)
> +{
> + dma_addr_t ret = desc->phys;
> +
> + if (lp->features & XAE_FEATURE_DMA_64BIT)
> + ret |= (dma_addr_t)desc->phys_msb << 32;

Does this compile/work properly on a 32-bit kernel? If dma_addr_t was a
32-bit type, I would expect that left-shifting by 32 bits may not do
what you want. Not sure if there is an inverse to lower_32_bits and
upper_32_bits macros?

> +
> + return ret;
> +}
> +
> /**
> * axienet_dma_bd_release - Release buffer descriptor rings
> * @ndev: Pointer to the net_device structure
> @@ -176,6 +195,8 @@ static void axienet_dma_bd_release(struct net_device *ndev)
> return;
>
> for (i = 0; i < lp->rx_bd_num; i++) {
> + dma_addr_t phys;
> +
> /* A NULL skb means this descriptor has not been initialised
> * at all.
> */
> @@ -188,9 +209,11 @@ static void axienet_dma_bd_release(struct net_device *ndev)
> * descriptor size, after it had been successfully allocated.
> * So a non-zero value in there means we need to unmap it.
> */
> - if (lp->rx_bd_v[i].cntrl)
> - dma_unmap_single(ndev->dev.parent, lp->rx_bd_v[i].phys,
> + if (lp->rx_bd_v[i].cntrl) {
> + phys = desc_get_phys_addr(lp, &lp->rx_bd_v[i]);
> + dma_unmap_single(ndev->dev.parent, phys,
> lp->max_frm_size, DMA_FROM_DEVICE);
> + }
> }
>
> dma_free_coherent(ndev->dev.parent,
> @@ -235,29 +258,36 @@ static int axienet_dma_bd_init(struct net_device *ndev)
> goto out;
>
> for (i = 0; i < lp->tx_bd_num; i++) {
> - lp->tx_bd_v[i].next = lp->tx_bd_p +
> - sizeof(*lp->tx_bd_v) *
> - ((i + 1) % lp->tx_bd_num);
> + dma_addr_t addr = lp->tx_bd_p +
> + sizeof(*lp->tx_bd_v) *
> + ((i + 1) % lp->tx_bd_num);
> +
> + lp->tx_bd_v[i].next = lower_32_bits(addr);
> + if (lp->features & XAE_FEATURE_DMA_64BIT)
> + lp->tx_bd_v[i].next_msb = upper_32_bits(addr);
> }
>
> for (i = 0; i < lp->rx_bd_num; i++) {
> - lp->rx_bd_v[i].next = lp->rx_bd_p +
> - sizeof(*lp->rx_bd_v) *
> - ((i + 1) % lp->rx_bd_num);
> + dma_addr_t addr;
> +
> + addr = lp->rx_bd_p + sizeof(*lp->rx_bd_v) *
> + ((i + 1) % lp->rx_bd_num);
> + lp->rx_bd_v[i].next = lower_32_bits(addr);
> + if (lp->features & XAE_FEATURE_DMA_64BIT)
> + lp->rx_bd_v[i].next_msb = upper_32_bits(addr);
>
> skb = netdev_alloc_skb_ip_align(ndev, lp->max_frm_size);
> if (!skb)
> goto out;
>
> lp->rx_bd_v[i].skb = skb;
> - lp->rx_bd_v[i].phys = dma_map_single(ndev->dev.parent,
> - skb->data,
> - lp->max_frm_size,
> - DMA_FROM_DEVICE);
> - if (dma_mapping_error(ndev->dev.parent, lp->rx_bd_v[i].phys)) {
> + addr = dma_map_single(ndev->dev.parent, skb->data,
> + lp->max_frm_size, DMA_FROM_DEVICE);
> + if (dma_mapping_error(ndev->dev.parent, addr)) {
> netdev_err(ndev, "DMA mapping error\n");
> goto out;
> }
> + desc_set_phys_addr(lp, addr, &lp->rx_bd_v[i]);
>
> lp->rx_bd_v[i].cntrl = lp->max_frm_size;
> }
> @@ -573,6 +603,7 @@ static int axienet_free_tx_chain(struct net_device *ndev, u32 first_bd,
> struct axienet_local *lp = netdev_priv(ndev);
> int max_bds = (nr_bds != -1) ? nr_bds : lp->tx_bd_num;
> struct axidma_bd *cur_p;
> + dma_addr_t phys;
> unsigned int status;
> int i;
>
> @@ -586,9 +617,10 @@ static int axienet_free_tx_chain(struct net_device *ndev, u32 first_bd,
> if (nr_bds == -1 && !(status & XAXIDMA_BD_STS_COMPLETE_MASK))
> break;
>
> - dma_unmap_single(ndev->dev.parent, cur_p->phys,
> - (cur_p->cntrl & XAXIDMA_BD_CTRL_LENGTH_MASK),
> - DMA_TO_DEVICE);
> + phys = desc_get_phys_addr(lp, cur_p);
> + dma_unmap_single(ndev->dev.parent, phys,
> + (cur_p->cntrl & XAXIDMA_BD_CTRL_LENGTH_MASK),
> + DMA_TO_DEVICE);
>
> if (cur_p->skb && (status & XAXIDMA_BD_STS_COMPLETE_MASK))
> dev_consume_skb_irq(cur_p->skb);
> @@ -684,7 +716,7 @@ axienet_start_xmit(struct sk_buff *skb, struct net_device *ndev)
> u32 csum_start_off;
> u32 csum_index_off;
> skb_frag_t *frag;
> - dma_addr_t tail_p;
> + dma_addr_t tail_p, phys;
> struct axienet_local *lp = netdev_priv(ndev);
> struct axidma_bd *cur_p;
> u32 orig_tail_ptr = lp->tx_bd_tail;
> @@ -723,14 +755,15 @@ axienet_start_xmit(struct sk_buff *skb, struct net_device *ndev)
> cur_p->app0 |= 2; /* Tx Full Checksum Offload Enabled */
> }
>
> - cur_p->phys = dma_map_single(ndev->dev.parent, skb->data,
> - skb_headlen(skb), DMA_TO_DEVICE);
> - if (unlikely(dma_mapping_error(ndev->dev.parent, cur_p->phys))) {
> + phys = dma_map_single(ndev->dev.parent, skb->data,
> + skb_headlen(skb), DMA_TO_DEVICE);
> + if (unlikely(dma_mapping_error(ndev->dev.parent, phys))) {
> if (net_ratelimit())
> netdev_err(ndev, "TX DMA mapping error\n");
> ndev->stats.tx_dropped++;
> return NETDEV_TX_OK;
> }
> + desc_set_phys_addr(lp, phys, cur_p);
> cur_p->cntrl = skb_headlen(skb) | XAXIDMA_BD_CTRL_TXSOF_MASK;
>
> for (ii = 0; ii < num_frag; ii++) {
> @@ -738,11 +771,11 @@ axienet_start_xmit(struct sk_buff *skb, struct net_device *ndev)
> lp->tx_bd_tail = 0;
> cur_p = &lp->tx_bd_v[lp->tx_bd_tail];
> frag = &skb_shinfo(skb)->frags[ii];
> - cur_p->phys = dma_map_single(ndev->dev.parent,
> - skb_frag_address(frag),
> - skb_frag_size(frag),
> - DMA_TO_DEVICE);
> - if (unlikely(dma_mapping_error(ndev->dev.parent, cur_p->phys))) {
> + phys = dma_map_single(ndev->dev.parent,
> + skb_frag_address(frag),
> + skb_frag_size(frag),
> + DMA_TO_DEVICE);
> + if (unlikely(dma_mapping_error(ndev->dev.parent, phys))) {
> if (net_ratelimit())
> netdev_err(ndev, "TX DMA mapping error\n");
> ndev->stats.tx_dropped++;
> @@ -752,6 +785,7 @@ axienet_start_xmit(struct sk_buff *skb, struct net_device *ndev)
>
> return NETDEV_TX_OK;
> }
> + desc_set_phys_addr(lp, phys, cur_p);
> cur_p->cntrl = skb_frag_size(frag);
> }
>
> @@ -790,10 +824,12 @@ static void axienet_recv(struct net_device *ndev)
> cur_p = &lp->rx_bd_v[lp->rx_bd_ci];
>
> while ((cur_p->status & XAXIDMA_BD_STS_COMPLETE_MASK)) {
> + dma_addr_t phys;
> +
> tail_p = lp->rx_bd_p + sizeof(*lp->rx_bd_v) * lp->rx_bd_ci;
>
> - dma_unmap_single(ndev->dev.parent, cur_p->phys,
> - lp->max_frm_size,
> + phys = desc_get_phys_addr(lp, cur_p);
> + dma_unmap_single(ndev->dev.parent, phys, lp->max_frm_size,
> DMA_FROM_DEVICE);
>
> skb = cur_p->skb;
> @@ -829,15 +865,16 @@ static void axienet_recv(struct net_device *ndev)
> if (!new_skb)
> return;
>
> - cur_p->phys = dma_map_single(ndev->dev.parent, new_skb->data,
> - lp->max_frm_size,
> - DMA_FROM_DEVICE);
> - if (unlikely(dma_mapping_error(ndev->dev.parent, cur_p->phys))) {
> + phys = dma_map_single(ndev->dev.parent, new_skb->data,
> + lp->max_frm_size,
> + DMA_FROM_DEVICE);
> + if (unlikely(dma_mapping_error(ndev->dev.parent, phys))) {
> if (net_ratelimit())
> netdev_err(ndev, "RX DMA mapping error\n");
> dev_kfree_skb(new_skb);
> return;
> }
> + desc_set_phys_addr(lp, phys, cur_p);
>
> cur_p->cntrl = lp->max_frm_size;
> cur_p->status = 0;
> @@ -882,7 +919,8 @@ static irqreturn_t axienet_tx_irq(int irq, void *_ndev)
> return IRQ_NONE;
> if (status & XAXIDMA_IRQ_ERROR_MASK) {
> dev_err(&ndev->dev, "DMA Tx error 0x%x\n", status);
> - dev_err(&ndev->dev, "Current BD is at: 0x%x\n",
> + dev_err(&ndev->dev, "Current BD is at: 0x%x%08x\n",
> + (lp->tx_bd_v[lp->tx_bd_ci]).phys_msb,
> (lp->tx_bd_v[lp->tx_bd_ci]).phys);
>
> cr = axienet_dma_in32(lp, XAXIDMA_TX_CR_OFFSET);
> @@ -931,7 +969,8 @@ static irqreturn_t axienet_rx_irq(int irq, void *_ndev)
> return IRQ_NONE;
> if (status & XAXIDMA_IRQ_ERROR_MASK) {
> dev_err(&ndev->dev, "DMA Rx error 0x%x\n", status);
> - dev_err(&ndev->dev, "Current BD is at: 0x%x\n",
> + dev_err(&ndev->dev, "Current BD is at: 0x%x%08x\n",
> + (lp->rx_bd_v[lp->rx_bd_ci]).phys_msb,
> (lp->rx_bd_v[lp->rx_bd_ci]).phys);
>
> cr = axienet_dma_in32(lp, XAXIDMA_TX_CR_OFFSET);
> @@ -1633,14 +1672,18 @@ static void axienet_dma_err_handler(struct work_struct *work)
>
> for (i = 0; i < lp->tx_bd_num; i++) {
> cur_p = &lp->tx_bd_v[i];
> - if (cur_p->cntrl)
> - dma_unmap_single(ndev->dev.parent, cur_p->phys,
> + if (cur_p->cntrl) {
> + dma_addr_t addr = desc_get_phys_addr(lp, cur_p);
> +
> + dma_unmap_single(ndev->dev.parent, addr,
> (cur_p->cntrl &
> XAXIDMA_BD_CTRL_LENGTH_MASK),
> DMA_TO_DEVICE);
> + }
> if (cur_p->skb)
> dev_kfree_skb_irq(cur_p->skb);
> cur_p->phys = 0;
> + cur_p->phys_msb = 0;
> cur_p->cntrl = 0;
> cur_p->status = 0;
> cur_p->app0 = 0;
>

--
Robert Hancock
Senior Hardware Designer
SED Systems, a division of Calian Ltd.
Email: hancock@xxxxxxxxxxxxx

Next message: Stephen Boyd: "Re: [PATCH v4 1/4] dt-bindings: Introduce SoC sleep stats bindings"
Previous message: Michael Walle: "Re: [PATCH 0/6] NXP DSPI bugfixes and support for LS1028A"
In reply to: Andre Przywara: "[PATCH v2 12/14] net: axienet: Upgrade descriptors to hold 64-bit addresses"
Next in thread: Andre Przywara: "Re: [PATCH v2 12/14] net: axienet: Upgrade descriptors to hold 64-bit addresses"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]