Re: [PATCH 3/5] sh: fix DMA driver's descriptor chaining and cookie assignment

From: Dan Williams
Date: Mon Dec 07 2009 - 19:58:12 EST


I would like an acked-by from Nobuhiro on this one.

[ full patch quoted below ]

Thanks,
Dan

On Fri, Dec 4, 2009 at 11:44 AM, Guennadi Liakhovetski
<g.liakhovetski@xxxxxx> wrote:
> The SH DMA driver wrongly assigns negative cookies to transfer descriptors,
> also, its chaining of partial descriptors is broken. The latter problem is
> usually invisible, because maximum transfer size per chunk is 16M, but if you
> artificially set this limit lower, the driver fails. Since cookies are also
> used in chunk management, both these problems are fixed in one patch.
>
> Signed-off-by: Guennadi Liakhovetski <g.liakhovetski@xxxxxx>
> ---
>  drivers/dma/shdma.c |  246 ++++++++++++++++++++++++++++++++-------------------
>  drivers/dma/shdma.h |    1 -
>  2 files changed, 153 insertions(+), 94 deletions(-)
>
> diff --git a/drivers/dma/shdma.c b/drivers/dma/shdma.c
> index f5fae12..aac8f78 100644
> --- a/drivers/dma/shdma.c
> +++ b/drivers/dma/shdma.c
> @@ -30,9 +30,12 @@
>  #include "shdma.h"
>
>  /* DMA descriptor control */
> -#define DESC_LAST      (-1)
> -#define DESC_COMP      (1)
> -#define DESC_NCOMP     (0)
> +enum sh_dmae_desc_status {
> +       DESC_PREPARED,
> +       DESC_SUBMITTED,
> +       DESC_COMPLETED, /* completed, have to call callback */
> +       DESC_WAITING,   /* callback called, waiting for ack / re-submit */
> +};
>
>  #define NR_DESCS_PER_CHANNEL 32
>  /*
> @@ -45,6 +48,8 @@
>  */
>  #define RS_DEFAULT  (RS_DUAL)
>
> +static void sh_dmae_chan_ld_cleanup(struct sh_dmae_chan *sh_chan, bool all);
> +
>  #define SH_DMAC_CHAN_BASE(id) (dma_base_addr[id])
>  static void sh_dmae_writel(struct sh_dmae_chan *sh_dc, u32 data, u32 reg)
>  {
> @@ -185,7 +190,7 @@ static int dmae_set_dmars(struct sh_dmae_chan *sh_chan, u16 val)
>
>  static dma_cookie_t sh_dmae_tx_submit(struct dma_async_tx_descriptor *tx)
>  {
> -       struct sh_desc *desc = tx_to_sh_desc(tx);
> +       struct sh_desc *desc = tx_to_sh_desc(tx), *chunk;
>        struct sh_dmae_chan *sh_chan = to_sh_chan(tx->chan);
>        dma_cookie_t cookie;
>
> @@ -196,45 +201,40 @@ static dma_cookie_t sh_dmae_tx_submit(struct dma_async_tx_descriptor *tx)
>        if (cookie < 0)
>                cookie = 1;
>
> -       /* If desc only in the case of 1 */
> -       if (desc->async_tx.cookie != -EBUSY)
> -               desc->async_tx.cookie = cookie;
> -       sh_chan->common.cookie = desc->async_tx.cookie;
> -
> -       list_splice_init(&desc->tx_list, sh_chan->ld_queue.prev);
> +       tx->cookie = cookie;
> +       dev_dbg(sh_chan->dev, "submit %p #%d start %u\n",
> +               tx, tx->cookie, desc->hw.sar);
> +       sh_chan->common.cookie = tx->cookie;
> +
> +       /* Mark all chunks of this descriptor as submitted */
> +       list_for_each_entry(chunk, desc->node.prev, node) {
> +               /*
> +                * All chunks are on the global ld_queue, so, we have to find
> +                * the end of the chain ourselves
> +                */
> +               if (chunk != desc && (chunk->async_tx.cookie > 0 ||
> +                                     chunk->async_tx.cookie == -EBUSY))
> +                       break;
> +               chunk->mark = DESC_SUBMITTED;
> +       }
>
>        spin_unlock_bh(&sh_chan->desc_lock);
>
>        return cookie;
>  }
>
> +/* Called with desc_lock held */
>  static struct sh_desc *sh_dmae_get_desc(struct sh_dmae_chan *sh_chan)
>  {
> -       struct sh_desc *desc, *_desc, *ret = NULL;
> -
> -       spin_lock_bh(&sh_chan->desc_lock);
> -       list_for_each_entry_safe(desc, _desc, &sh_chan->ld_free, node) {
> -               if (async_tx_test_ack(&desc->async_tx)) {
> -                       list_del(&desc->node);
> -                       ret = desc;
> -                       break;
> -               }
> -       }
> -       spin_unlock_bh(&sh_chan->desc_lock);
> -
> -       return ret;
> -}
> +       struct sh_desc *desc;
>
> -static void sh_dmae_put_desc(struct sh_dmae_chan *sh_chan, struct sh_desc *desc)
> -{
> -       if (desc) {
> -               spin_lock_bh(&sh_chan->desc_lock);
> +       if (list_empty(&sh_chan->ld_free))
> +               return NULL;
>
> -               list_splice_init(&desc->tx_list, &sh_chan->ld_free);
> -               list_add(&desc->node, &sh_chan->ld_free);
> +       desc = list_entry(sh_chan->ld_free.next, struct sh_desc, node);
> +       list_del(&desc->node);
>
> -               spin_unlock_bh(&sh_chan->desc_lock);
> -       }
> +       return desc;
>  }
>
>  static int sh_dmae_alloc_chan_resources(struct dma_chan *chan)
> @@ -254,10 +254,9 @@ static int sh_dmae_alloc_chan_resources(struct dma_chan *chan)
>                                        &sh_chan->common);
>                desc->async_tx.tx_submit = sh_dmae_tx_submit;
>                desc->async_tx.flags = DMA_CTRL_ACK;
> -               INIT_LIST_HEAD(&desc->tx_list);
> -               sh_dmae_put_desc(sh_chan, desc);
>
>                spin_lock_bh(&sh_chan->desc_lock);
> +               list_add(&desc->node, &sh_chan->ld_free);
>                sh_chan->descs_allocated++;
>        }
>        spin_unlock_bh(&sh_chan->desc_lock);
> @@ -274,7 +273,10 @@ static void sh_dmae_free_chan_resources(struct dma_chan *chan)
>        struct sh_desc *desc, *_desc;
>        LIST_HEAD(list);
>
> -       BUG_ON(!list_empty(&sh_chan->ld_queue));
> +       /* Prepared and not submitted descriptors can still be on the queue */
> +       if (!list_empty(&sh_chan->ld_queue))
> +               sh_dmae_chan_ld_cleanup(sh_chan, true);
> +
>        spin_lock_bh(&sh_chan->desc_lock);
>
>        list_splice_init(&sh_chan->ld_free, &list);
> @@ -293,6 +295,7 @@ static struct dma_async_tx_descriptor *sh_dmae_prep_memcpy(
>        struct sh_dmae_chan *sh_chan;
>        struct sh_desc *first = NULL, *prev = NULL, *new;
>        size_t copy_size;
> +       LIST_HEAD(tx_list);
>
>        if (!chan)
>                return NULL;
> @@ -302,43 +305,70 @@ static struct dma_async_tx_descriptor *sh_dmae_prep_memcpy(
>
>        sh_chan = to_sh_chan(chan);
>
> +       /* Have to lock the whole loop to protect against concurrent release */
> +       spin_lock_bh(&sh_chan->desc_lock);
> +
> +       /*
> +        * Chaining:
> +        * first descriptor is what user is dealing with in all API calls, its
> +        *      cookie is at first set to -EBUSY, at tx-submit to a positive
> +        *      number
> +        * if more than one chunk is needed further chunks have cookie = -EINVAL
> +        * the last chunk, if not equal to the first, has cookie = -ENOSPC
> +        * all chunks are linked onto the tx_list head with their .node heads
> +        *      only during this function, then they are immediately spliced
> +        *      onto the queue
> +        */
>        do {
>                /* Allocate the link descriptor from DMA pool */
>                new = sh_dmae_get_desc(sh_chan);
>                if (!new) {
>                        dev_err(sh_chan->dev,
>                                        "No free memory for link descriptor\n");
> -                       goto err_get_desc;
> +                       list_splice(&tx_list, &sh_chan->ld_free);
> +                       spin_unlock_bh(&sh_chan->desc_lock);
> +                       return NULL;
>                }
>
> -               copy_size = min(len, (size_t)SH_DMA_TCR_MAX);
> +               copy_size = min(len, (size_t)SH_DMA_TCR_MAX + 1);
>
>                new->hw.sar = dma_src;
>                new->hw.dar = dma_dest;
>                new->hw.tcr = copy_size;
> -               if (!first)
> +               if (!first) {
> +                       /* First desc */
> +                       new->async_tx.cookie = -EBUSY;
>                        first = new;
> +               } else {
> +                       /* Other desc - invisible to the user */
> +                       new->async_tx.cookie = -EINVAL;
> +               }
> +
> +               dev_dbg(sh_chan->dev,
> +                       "chaining %u of %u with %p, start %u, cookie %d\n",
> +                       copy_size, len, &new->async_tx, dma_src,
> +                       new->async_tx.cookie);
>
> -               new->mark = DESC_NCOMP;
> -               async_tx_ack(&new->async_tx);
> +               new->mark = DESC_PREPARED;
> +               new->async_tx.flags = flags;
>
>                prev = new;
>                len -= copy_size;
>                dma_src += copy_size;
>                dma_dest += copy_size;
>                /* Insert the link descriptor to the LD ring */
> -               list_add_tail(&new->node, &first->tx_list);
> +               list_add_tail(&new->node, &tx_list);
>        } while (len);
>
> -       new->async_tx.flags = flags; /* client is in control of this ack */
> -       new->async_tx.cookie = -EBUSY; /* Last desc */
> +       if (new != first)
> +               new->async_tx.cookie = -ENOSPC;
>
> -       return &first->async_tx;
> +       /* Put them immediately on the queue, so, they don't get lost */
> +       list_splice_tail(&tx_list, sh_chan->ld_queue.prev);
>
> -err_get_desc:
> -       sh_dmae_put_desc(sh_chan, first);
> -       return NULL;
> +       spin_unlock_bh(&sh_chan->desc_lock);
>
> +       return &first->async_tx;
>  }
>
>  /*
> @@ -346,64 +376,87 @@ err_get_desc:
>  *
>  * This function clean up the ld_queue of DMA channel.
>  */
> -static void sh_dmae_chan_ld_cleanup(struct sh_dmae_chan *sh_chan)
> +static void sh_dmae_chan_ld_cleanup(struct sh_dmae_chan *sh_chan, bool all)
>  {
>        struct sh_desc *desc, *_desc;
>
>        spin_lock_bh(&sh_chan->desc_lock);
>        list_for_each_entry_safe(desc, _desc, &sh_chan->ld_queue, node) {
> -               dma_async_tx_callback callback;
> -               void *callback_param;
> +               struct dma_async_tx_descriptor *tx = &desc->async_tx;
>
> -               /* non send data */
> -               if (desc->mark == DESC_NCOMP)
> +               /* unsent data */
> +               if (desc->mark == DESC_SUBMITTED && !all)
>                        break;
>
> -               /* send data sesc */
> -               callback = desc->async_tx.callback;
> -               callback_param = desc->async_tx.callback_param;
> +               if (desc->mark == DESC_PREPARED && !all)
> +                       continue;
> +
> +               /* Call callback on the "exposed" descriptor (cookie > 0) */
> +               if (tx->cookie > 0 && (desc->mark == DESC_COMPLETED ||
> +                                      desc->mark == DESC_WAITING)) {
> +                       struct sh_desc *chunk;
> +                       bool head_acked = async_tx_test_ack(tx);
> +                       /* sent data desc */
> +                       dma_async_tx_callback callback = tx->callback;
> +
> +                       /* Run the link descriptor callback function */
> +                       if (callback && desc->mark == DESC_COMPLETED) {
> +                               spin_unlock_bh(&sh_chan->desc_lock);
> +                               dev_dbg(sh_chan->dev,
> +                                       "descriptor %p callback\n", tx);
> +                               callback(tx->callback_param);
> +                               spin_lock_bh(&sh_chan->desc_lock);
> +                       }
>
> -               /* Remove from ld_queue list */
> -               list_splice_init(&desc->tx_list, &sh_chan->ld_free);
> +                       list_for_each_entry(chunk, desc->node.prev, node) {
> +                               /*
> +                                * All chunks are on the global ld_queue, so, we
> +                                * have to find the end of the chain ourselves
> +                                */
> +                               if (chunk != desc &&
> +                                   (chunk->async_tx.cookie > 0 ||
> +                                    chunk->async_tx.cookie == -EBUSY))
> +                                       break;
> +
> +                               if (head_acked)
> +                                       async_tx_ack(&chunk->async_tx);
> +                               else
> +                                       chunk->mark = DESC_WAITING;
> +                       }
> +               }
>
> -               dev_dbg(sh_chan->dev, "link descriptor %p will be recycle.\n",
> -                               desc);
> +               dev_dbg(sh_chan->dev, "descriptor %p #%d completed.\n",
> +                       tx, tx->cookie);
>
> -               list_move(&desc->node, &sh_chan->ld_free);
> -               /* Run the link descriptor callback function */
> -               if (callback) {
> -                       spin_unlock_bh(&sh_chan->desc_lock);
> -                       dev_dbg(sh_chan->dev, "link descriptor %p callback\n",
> -                                       desc);
> -                       callback(callback_param);
> -                       spin_lock_bh(&sh_chan->desc_lock);
> -               }
> +               if (((desc->mark == DESC_COMPLETED ||
> +                     desc->mark == DESC_WAITING) &&
> +                    async_tx_test_ack(&desc->async_tx)) || all)
> +                       /* Remove from ld_queue list */
> +                       list_move(&desc->node, &sh_chan->ld_free);
>        }
>        spin_unlock_bh(&sh_chan->desc_lock);
>  }
>
>  static void sh_chan_xfer_ld_queue(struct sh_dmae_chan *sh_chan)
>  {
> -       struct list_head *ld_node;
>        struct sh_dmae_regs hw;
> +       struct sh_desc *sd;
>
>        /* DMA work check */
>        if (dmae_is_idle(sh_chan))
>                return;
>
> +       spin_lock_bh(&sh_chan->desc_lock);
>        /* Find the first un-transfer desciptor */
> -       for (ld_node = sh_chan->ld_queue.next;
> -               (ld_node != &sh_chan->ld_queue)
> -                       && (to_sh_desc(ld_node)->mark == DESC_COMP);
> -               ld_node = ld_node->next)
> -               cpu_relax();
> -
> -       if (ld_node != &sh_chan->ld_queue) {
> -               /* Get the ld start address from ld_queue */
> -               hw = to_sh_desc(ld_node)->hw;
> -               dmae_set_reg(sh_chan, hw);
> -               dmae_start(sh_chan);
> -       }
> +       list_for_each_entry(sd, &sh_chan->ld_queue, node)
> +               if (sd->mark == DESC_SUBMITTED) {
> +                       /* Get the ld start address from ld_queue */
> +                       hw = sd->hw;
> +                       dmae_set_reg(sh_chan, hw);
> +                       dmae_start(sh_chan);
> +                       break;
> +               }
> +       spin_unlock_bh(&sh_chan->desc_lock);
>  }
>
>  static void sh_dmae_memcpy_issue_pending(struct dma_chan *chan)
> @@ -421,12 +474,11 @@ static enum dma_status sh_dmae_is_complete(struct dma_chan *chan,
>        dma_cookie_t last_used;
>        dma_cookie_t last_complete;
>
> -       sh_dmae_chan_ld_cleanup(sh_chan);
> +       sh_dmae_chan_ld_cleanup(sh_chan, false);
>
>        last_used = chan->cookie;
>        last_complete = sh_chan->completed_cookie;
> -       if (last_complete == -EBUSY)
> -               last_complete = last_used;
> +       BUG_ON(last_complete < 0);
>
>        if (done)
>                *done = last_complete;
> @@ -481,11 +533,13 @@ static irqreturn_t sh_dmae_err(int irq, void *data)
>                err = sh_dmae_rst(0);
>                if (err)
>                        return err;
> +#ifdef SH_DMAC_BASE1
>                if (shdev->pdata.mode & SHDMA_DMAOR1) {
>                        err = sh_dmae_rst(1);
>                        if (err)
>                                return err;
>                }
> +#endif
>                disable_irq(irq);
>                return IRQ_HANDLED;
>        }
> @@ -499,30 +553,36 @@ static void dmae_do_tasklet(unsigned long data)
>        u32 sar_buf = sh_dmae_readl(sh_chan, SAR);
>        list_for_each_entry_safe(desc, _desc,
>                                        &sh_chan->ld_queue, node) {
> -               if ((desc->hw.sar + desc->hw.tcr) == sar_buf) {
> +               if ((desc->hw.sar + desc->hw.tcr) == sar_buf &&
> +                   desc->mark == DESC_SUBMITTED) {
>                        cur_desc = desc;
>                        break;
>                }
>        }
>
>        if (cur_desc) {
> +               dev_dbg(sh_chan->dev, "done %p #%d start %u\n",
> +                       &cur_desc->async_tx, cur_desc->async_tx.cookie,
> +                       cur_desc->hw.sar);
> +
>                switch (cur_desc->async_tx.cookie) {
> -               case 0: /* other desc data */
> +               case -EINVAL:
> +               case -ENOSPC:
> +                       /* other desc data */
>                        break;
> -               case -EBUSY: /* last desc */
> -               sh_chan->completed_cookie =
> -                               cur_desc->async_tx.cookie;
> +               case -EBUSY: /* Cannot be */
> +                       BUG_ON(1);
>                        break;
> -               default: /* first desc ( 0 < )*/
> +               default: /* first desc: cookie > 0 */
>                        sh_chan->completed_cookie =
> -                               cur_desc->async_tx.cookie - 1;
> +                               cur_desc->async_tx.cookie;
>                        break;
>                }
> -               cur_desc->mark = DESC_COMP;
> +               cur_desc->mark = DESC_COMPLETED;
>        }
>        /* Next desc */
>        sh_chan_xfer_ld_queue(sh_chan);
> -       sh_dmae_chan_ld_cleanup(sh_chan);
> +       sh_dmae_chan_ld_cleanup(sh_chan, false);
>  }
>
>  static unsigned int get_dmae_irq(unsigned int id)
> diff --git a/drivers/dma/shdma.h b/drivers/dma/shdma.h
> index 2b4bc15..c93555c 100644
> --- a/drivers/dma/shdma.h
> +++ b/drivers/dma/shdma.h
> @@ -26,7 +26,6 @@ struct sh_dmae_regs {
>  };
>
>  struct sh_desc {
> -       struct list_head tx_list;
>        struct sh_dmae_regs hw;
>        struct list_head node;
>        struct dma_async_tx_descriptor async_tx;
> --
> 1.6.2.4
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> the body of a message to majordomo@xxxxxxxxxxxxxxx
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
> Please read the FAQ at  http://www.tux.org/lkml/
>
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/