[PATCH v7 1/3] dmaengine: Add support for APM X-Gene SoC DMA engine driver
Rameshwar Sahu
rsahu at apm.com
Mon Mar 16 03:30:22 PDT 2015
Hi Vinod,
On Mon, Mar 16, 2015 at 2:57 PM, Vinod Koul <vinod.koul at intel.com> wrote:
> On Thu, Mar 12, 2015 at 04:45:19PM +0530, Rameshwar Prasad Sahu wrote:
>> +/* DMA ring csr registers and bit definations */
>> +#define DMA_RING_CONFIG 0x04
>> +#define DMA_RING_ENABLE BIT(31)
>> +#define DMA_RING_ID 0x08
>> +#define DMA_RING_ID_SETUP(v) ((v) | BIT(31))
>> +#define DMA_RING_ID_BUF 0x0C
>> +#define DMA_RING_ID_BUF_SETUP(v) (((v) << 9) | BIT(21))
>> +#define DMA_RING_THRESLD0_SET1 0x30
>> +#define DMA_RING_THRESLD0_SET1_VAL 0X64
>> +#define DMA_RING_THRESLD1_SET1 0x34
>> +#define DMA_RING_THRESLD1_SET1_VAL 0xC8
>> +#define DMA_RING_HYSTERESIS 0x68
>> +#define DMA_RING_HYSTERESIS_VAL 0xFFFFFFFF
>> +#define DMA_RING_STATE 0x6C
>> +#define DMA_RING_STATE_WR_BASE 0x70
>> +#define DMA_RING_NE_INT_MODE 0x017C
>> +#define DMA_RING_NE_INT_MODE_SET(m, v) \
>> + ((m) = ((m) & ~BIT(31 - (v))) | BIT(31 - (v)))
>> +#define DMA_RING_NE_INT_MODE_RESET(m, v) \
>> + ((m) &= (~BIT(31 - (v))))
>> +#define DMA_RING_CLKEN 0xC208
>> +#define DMA_RING_SRST 0xC200
>> +#define DMA_RING_MEM_RAM_SHUTDOWN 0xD070
>> +#define DMA_RING_BLK_MEM_RDY 0xD074
>> +#define DMA_RING_BLK_MEM_RDY_VAL 0xFFFFFFFF
>> +#define DMA_RING_DESC_CNT(v) (((v) & 0x0001FFFE) >> 1)
>> +#define DMA_RING_ID_GET(owner, num) (((owner) << 6) | (num))
>> +#define DMA_RING_DST_ID(v) ((1 << 10) | (v))
>> +#define DMA_RING_CMD_OFFSET 0x2C
>> +#define DMA_RING_CMD_BASE_OFFSET(v) ((v) << 6)
>> +#define DMA_RING_COHERENT_SET(m) (((u32 *)(m))[2] |= BIT(4))
>> +#define DMA_RING_ADDRL_SET(m, v) (((u32 *)(m))[2] |= (((v) >> 8) << 5))
>> +#define DMA_RING_ADDRH_SET(m, v) (((u32 *)(m))[3] |= ((v) >> 35))
>> +#define DMA_RING_ACCEPTLERR_SET(m) (((u32 *)(m))[3] |= BIT(19))
>> +#define DMA_RING_SIZE_SET(m, v) (((u32 *)(m))[3] |= ((v) << 23))
>> +#define DMA_RING_RECOMBBUF_SET(m) (((u32 *)(m))[3] |= BIT(27))
>> +#define DMA_RING_RECOMTIMEOUTL_SET(m) (((u32 *)(m))[3] |= (0x7 << 28))
>> +#define DMA_RING_RECOMTIMEOUTH_SET(m) (((u32 *)(m))[4] |= 0x3)
>> +#define DMA_RING_SELTHRSH_SET(m) (((u32 *)(m))[4] |= BIT(3))
>> +#define DMA_RING_TYPE_SET(m, v) (((u32 *)(m))[4] |= ((v) << 19))
> These are very generic name as can conflicts, can you please namespace
> these...
Okay
>
>> +/* DMA device csr registers and bit definitions */
>> +#define DMA_IPBRR 0x0
>> +#define DMA_DEV_ID_RD(v) ((v) & 0x00000FFF)
>> +#define DMA_BUS_ID_RD(v) (((v) >> 12) & 3)
>> +#define DMA_REV_NO_RD(v) (((v) >> 14) & 3)
>> +#define DMA_GCR 0x10
>> +#define DMA_CH_SETUP(v) ((v) = ((v) & ~0x000FFFFF) | 0x000AAFFF)
>> +#define DMA_ENABLE(v) ((v) |= BIT(31))
>> +#define DMA_DISABLE(v) ((v) &= ~BIT(31))
>> +#define DMA_RAID6_CONT 0x14
>> +#define DMA_RAID6_MULTI_CTRL(v) ((v) << 24)
>> +#define DMA_INT 0x70
>> +#define DMA_INT_MASK 0x74
>> +#define DMA_INT_ALL_MASK 0xFFFFFFFF
>> +#define DMA_INT_ALL_UNMASK 0x0
>> +#define DMA_INT_MASK_SHIFT 0x14
>> +#define DMA_RING_INT0_MASK 0x90A0
>> +#define DMA_RING_INT1_MASK 0x90A8
>> +#define DMA_RING_INT2_MASK 0x90B0
>> +#define DMA_RING_INT3_MASK 0x90B8
>> +#define DMA_RING_INT4_MASK 0x90C0
>> +#define DMA_CFG_RING_WQ_ASSOC 0x90E0
>> +#define DMA_ASSOC_RING_MNGR1 0xFFFFFFFF
>> +#define DMA_MEM_RAM_SHUTDOWN 0xD070
>> +#define DMA_BLK_MEM_RDY 0xD074
>> +#define DMA_BLK_MEM_RDY_VAL 0xFFFFFFFF
> same here and throughout the driver..
>
Okay...
>> +static void xgene_dma_free_descriptor(struct xgene_dma_chan *chan,
>> + struct xgene_dma_desc_sw *desc)
>> +{
>> + list_del(&desc->node);
>> + chan_dbg(chan, "LD %p free\n", desc);
>> + dma_pool_free(chan->desc_pool, desc, desc->tx.phys);
> where is the descriptor freed? Perhpas we can say clean rather than free
> here to not confuse...
xgene_dma_clean_completed_descriptor() is freeing the descriptor once
descriptor acked by client.
This is in the completion path.
>
>> +}
>> +
>> +static struct xgene_dma_desc_sw *xgene_dma_alloc_descriptor(
>> + struct xgene_dma_chan *chan)
>> +{
>> + struct xgene_dma_desc_sw *desc;
>> + dma_addr_t phys;
>> +
>> + desc = dma_pool_alloc(chan->desc_pool, GFP_NOWAIT, &phys);
>> + if (!desc) {
>> + chan_dbg(chan, "Failed to allocate LDs\n");
> not error?
Yes it's error only by lacking of dma memory, do I need to use dev_err
to show the error msg ??
>
>> +static void xgene_dma_free_desc_list_reverse(struct xgene_dma_chan *chan,
>> + struct list_head *list)
> do we really care about free order?
Yes it start dellocation of descriptor by tail.
>
>> +{
>> + struct xgene_dma_desc_sw *desc, *_desc;
>> +
>> + list_for_each_entry_safe_reverse(desc, _desc, list, node)
>> + xgene_dma_free_descriptor(chan, desc);
>> +}
>> +
>> +static void xgene_dma_free_chan_resources(struct dma_chan *dchan)
>> +{
>> + struct xgene_dma_chan *chan = to_dma_chan(dchan);
>> +
>> + chan_dbg(chan, "Free all resources\n");
>> +
>> + if (!chan->desc_pool)
>> + return;
>> +
>> + spin_lock_bh(&chan->lock);
>> +
>> + /* Process all running descriptor */
>> + xgene_dma_cleanup_descriptors(chan);
>> +
>> + /* Clean all link descriptor queues */
>> + xgene_dma_free_desc_list(chan, &chan->ld_pending);
>> + xgene_dma_free_desc_list(chan, &chan->ld_running);
>> + xgene_dma_free_desc_list(chan, &chan->ld_completed);
>> +
>> + spin_unlock_bh(&chan->lock);
>> +
>> + /* Delete this channel DMA pool */
>> + dma_pool_destroy(chan->desc_pool);
>> + chan->desc_pool = NULL;
>> +}
>> +
>> +static struct dma_async_tx_descriptor *xgene_dma_prep_memcpy(
>> + struct dma_chan *dchan, dma_addr_t dst, dma_addr_t src,
>> + size_t len, unsigned long flags)
>> +{
>> + struct xgene_dma_desc_sw *first = NULL, *new;
>> + struct xgene_dma_chan *chan;
>> + size_t copy;
>> + u32 desc_id;
>> +
>> + if (unlikely(!dchan || !len))
>> + return NULL;
>> +
>> + chan = to_dma_chan(dchan);
>> +
>> + /* Get the id for this group of descriptors */
>> + desc_id = atomic_inc_return(&chan->desc_id);
>> +
>> + do {
>> + /* Allocate the link descriptor from DMA pool */
>> + new = xgene_dma_alloc_descriptor(chan);
>> + if (!new)
>> + goto fail;
>> +
>> + /* Create the largest transaction possible */
>> + copy = min_t(size_t, len, DMA_MAX_64B_DESC_BYTE_CNT);
>> +
>> + /* Prepare DMA descriptor */
>> + xgene_dma_prep_cpy_desc(chan, new, dst, src, copy, desc_id);
>> +
>> + if (!first)
>> + first = new;
>> +
>> + new->first = first;
>> + first->desc_cnt++;
>> + new->desc_id = desc_id;
>> +
>> + new->tx.cookie = 0;
>> + async_tx_ack(&new->tx);
>> +
>> + /* Update metadata */
>> + len -= copy;
>> + dst += copy;
>> + src += copy;
>> +
>> + /* Insert the link descriptor to the LD ring */
>> + list_add_tail(&new->node, &first->tx_list);
>> + } while (len);
>> +
>> + first->tx.flags = flags; /* client is in control of this ack */
>> + first->tx.cookie = -EBUSY;
>> + first->flags |= DMA_FLAG_FIRST_DESC;
>> +
>> + return &first->tx;
> where are you mapping dma buffers?
I didn't get you here. Can you please explain me here what you mean.
As per my understanding client should map the dma buffer and give the
physical address and size to this callback prep routines.
>
>> +static enum dma_status xgene_dma_find_tx_desc_status(
>> + struct xgene_dma_chan *chan, dma_cookie_t cookie,
>> + struct dma_tx_state *txstate)
>> +{
>> + struct xgene_dma_desc_sw *desc;
>> +
>> + spin_lock_bh(&chan->lock);
>> +
>> + /* Check if this tx descriptor is still in pending queue */
>> + list_for_each_entry(desc, &chan->ld_pending, node) {
>> + if (desc->tx.cookie == cookie) {
>> + xgene_chan_xfer_ld_pending(chan);
> why are you calling this here, status check shouldnt do this...
Okay, I will remove it.
>> + spin_unlock_bh(&chan->lock);
>> + return DMA_IN_PROGRESS;
> residue here is size of transacation.
We can't calculate here residue size. We don't have any controller
register which will tell about remaining transaction size.
>> + }
>> + }
>> +
>> + /* Check if this descriptor is in running queue */
>> + list_for_each_entry(desc, &chan->ld_running, node) {
>> + if (desc->tx.cookie == cookie) {
>> + /* Cleanup any running and executed descriptors */
>> + xgene_dma_cleanup_descriptors(chan);
> ditto?
Okay
>> + spin_unlock_bh(&chan->lock);
>> + return dma_cookie_status(&chan->dma_chan,
>> + cookie, txstate);
> and you havent touched txstate so what is the intent here...?
txstate can filled by caller, so it may be NULL or not NULL, we are
passing same.
>> + }
>> + }
>> +
>> + spin_unlock_bh(&chan->lock);
>> +
>> + return DMA_COMPLETE;
>> +}
>> +
>> +static enum dma_status xgene_dma_tx_status(struct dma_chan *dchan,
>> + dma_cookie_t cookie,
>> + struct dma_tx_state *txstate)
>> +{
>> + struct xgene_dma_chan *chan = to_dma_chan(dchan);
>> +
>> + enum dma_status status;
>> +
>> + status = dma_cookie_status(dchan, cookie, txstate);
>> + if (status == DMA_COMPLETE)
> you should do this if txstate in NULL, no point doing calculations..
>
>> + return status;
>> +
>> + return xgene_dma_find_tx_desc_status(chan, cookie, txstate);
>> +}
>> +
>> +static void xgene_dma_tasklet_cb(unsigned long data)
>> +{
>> + struct xgene_dma_chan *chan = (struct xgene_dma_chan *)data;
>> +
>> + spin_lock_bh(&chan->lock);
>> +
>> + /* Run all cleanup for descriptors which have been completed */
>> + xgene_dma_cleanup_descriptors(chan);
>> +
>> + /* Re-enable DMA channel IRQ */
>> + enable_irq(chan->rx_irq);
>> +
>> + spin_unlock_bh(&chan->lock);
>> +}
>> +
>> +static irqreturn_t xgene_dma_chan_ring_isr(int irq, void *id)
>> +{
>> + struct xgene_dma_chan *chan = (struct xgene_dma_chan *)id;
>> +
>> + BUG_ON(!chan);
>> +
>> + /*
>> + * Disable DMA channel IRQ until we process completed
>> + * descriptors
>> + */
>> + disable_irq_nosync(chan->rx_irq);
> and why is that?
This will reduce the interrupt overheads and will give us better performance.
we are enabling it again when we are done with tasklet function.
>
>> +
>> + /*
>> + * Schedule the tasklet to handle all cleanup of the current
>> + * transaction. It will start a new transaction if there is
>> + * one pending.
>> + */
>> + tasklet_schedule(&chan->tasklet);
> for better results why not schedule the next transaction here..?
I agree, I will do this.
>> +
>> + return IRQ_HANDLED;
>> +}
>> +
>
>> +static void xgene_dma_async_unregister(struct xgene_dma *pdma)
>> +{
>> + int i;
>> +
>> + for (i = 0; i < DMA_MAX_CHANNEL; i++)
>> + dma_async_device_unregister(&pdma->dma_dev[i]);
> how do you ensure irq is disabled and tasklets killed?
After un-registering async dma device we are freeing irqs and killing tasklet.
Does this make sense. and do we need to add some way to make sure
about irq disabled.
>
>> +MODULE_DESCRIPTION("APM X-Gene SoC DMA driver");
>> +MODULE_AUTHOR("Rameshwar Prasad Sahu <rsahu at apm.com>");
>> +MODULE_AUTHOR("Loc Ho <lho at apm.com>");
>> +MODULE_LICENSE("GPL");
>> +MODULE_VERSION("1.0");
> why do you need this?
I don't see really use of this, but this will give us idea which
version of dma driver support what offload functionality for later.
>> --
>> 1.8.2.1
>>
>
> --
More information about the linux-arm-kernel
mailing list