Linux PCI Non-Transparent Bridge framework and drivers
 help / color / mirror / Atom feed
From: "Jiang, Dave" <dave.jiang@intel.com>
To: "allenbh@gmail.com" <allenbh@gmail.com>
Cc: "allen.hubbe@emc.com" <allen.hubbe@emc.com>,
	"dmaengine@vger.kernel.org" <dmaengine@vger.kernel.org>,
	"Williams, Dan J" <dan.j.williams@intel.com>,
	"jdmason@kudzu.us" <jdmason@kudzu.us>,
	"Koul, Vinod" <vinod.koul@intel.com>,
	"linux-ntb@googlegroups.com" <linux-ntb@googlegroups.com>
Subject: Re: [PATCH 2/5] dmaengine: IOATDMA: Add error handling to driver
Date: Tue, 28 Jun 2016 16:11:52 +0000	[thread overview]
Message-ID: <1467130291.3754.8.camel@intel.com> (raw)
In-Reply-To: <CAJ80satySmQk_O9hu6E7ytEovSq7v=DMFo9jawU-Sizx2Or8Lg@mail.gmail.com>

On Tue, 2016-06-28 at 01:08 +0000, Allen Hubbe wrote:
> On Mon, Jun 27, 2016 at 7:09 PM, Dave Jiang <dave.jiang@intel.com>
> wrote:
> > Signed-off-by: Dave Jiang <dave.jiang@intel.com>
> > ---
> >  drivers/dma/ioat/dma.c       |  136
> > ++++++++++++++++++++++++++++++++++++------
> >  drivers/dma/ioat/registers.h |    2 +
> >  2 files changed, 120 insertions(+), 18 deletions(-)
> > 
> > diff --git a/drivers/dma/ioat/dma.c b/drivers/dma/ioat/dma.c
> > index bd09961..1bbc005 100644
> > --- a/drivers/dma/ioat/dma.c
> > +++ b/drivers/dma/ioat/dma.c
> > @@ -622,7 +622,8 @@ static void ioat_cleanup(struct ioatdma_chan
> > *ioat_chan)
> >         if (is_ioat_halted(*ioat_chan->completion)) {
> >                 u32 chanerr = readl(ioat_chan->reg_base +
> > IOAT_CHANERR_OFFSET);
> > 
> > -               if (chanerr & IOAT_CHANERR_HANDLE_MASK) {
> > +               if (chanerr &
> > +                   (IOAT_CHANERR_HANDLE_MASK |
> > IOAT_CHANERR_RECOVER_MASK)) {
> >                         mod_timer(&ioat_chan->timer, jiffies +
> > IDLE_TIMEOUT);
> >                         ioat_eh(ioat_chan);
> >                 }
> > @@ -652,6 +653,60 @@ static void ioat_restart_channel(struct
> > ioatdma_chan *ioat_chan)
> >         __ioat_restart_chan(ioat_chan);
> >  }
> > 
> > +
> > +static void ioat_abort_descs(struct ioatdma_chan *ioat_chan)
> > +{
> > +       struct ioatdma_device *ioat_dma = ioat_chan->ioat_dma;
> > +       struct ioat_ring_ent *desc;
> > +       u16 active;
> > +       int idx = ioat_chan->tail, i;
> > +
> > +       /*
> > +        * We assume that the failed descriptor has been processed.
> > +        * Now we are just returning all the remaining submitted
> > +        * descriptors to abort.
> > +        */
> > +       active = ioat_ring_active(ioat_chan);
> > +
> > +       /* we skip the failed descriptor that tail points to */
> > +       for (i = 1; i < active; i++) {
> > +               struct dma_async_tx_descriptor *tx;
> > +
> > +               smp_read_barrier_depends();
> > +               prefetch(ioat_get_ring_ent(ioat_chan, idx + i +
> > 1));
> > +               desc = ioat_get_ring_ent(ioat_chan, idx + i);
> > +               desc->txd.result |= ERR_DMA_ABORT;
> 
> See below, txd.result doesn't seem to be a bit mask.

I'll fix that. It shouldn't be a bit mask.

> 
> > +
> > +               tx = &desc->txd;
> > +               if (tx->cookie) {
> > +                       dma_cookie_complete(tx);
> > +                       dma_descriptor_unmap(tx);
> > +                       if (tx->callback) {
> > +                               tx->callback(tx->callback_param);
> > +                               tx->callback = NULL;
> > +                       }
> > +               }
> > +
> > +               /* skip extended descriptors */
> > +               if (desc_has_ext(desc)) {
> > +                       WARN_ON(i + 1 >= active);
> > +                       i++;
> 
> Is it always true that extended descriptors (note: descriptors is
> plural) means "exactly one?"  Otherwise, ++ aught to be += how how
> many.  I didn't notice anything in dma_prep.c that would use more
> than
> one extra.

There can only be 1 extended descriptor always.

> 
> > +               }
> > +
> > +               /* cleanup super extended descriptors */
> > +               if (desc->sed) {
> > +                       ioat_free_sed(ioat_dma, desc->sed);
> > +                       desc->sed = NULL;
> > +               }
> > +       }
> > +
> > +       smp_mb(); /* finish all descriptor reads before
> > incrementing tail */
> > +       ioat_chan->tail = idx + i;
> 
> Would it also be correct to say `idx + active` here?  If so, that
> might be more clear.

ok

> 
> > +
> > +       desc = ioat_get_ring_ent(ioat_chan, ioat_chan->tail);
> > +       ioat_chan->last_completion = *ioat_chan->completion = desc-
> > >txd.phys;
> > +}
> > +
> >  static void ioat_eh(struct ioatdma_chan *ioat_chan)
> >  {
> >         struct pci_dev *pdev = to_pdev(ioat_chan);
> > @@ -662,6 +717,7 @@ static void ioat_eh(struct ioatdma_chan
> > *ioat_chan)
> >         u32 err_handled = 0;
> >         u32 chanerr_int;
> >         u32 chanerr;
> > +       bool abort = false;
> > 
> >         /* cleanup so tail points to descriptor that caused the
> > error */
> >         if (ioat_cleanup_preamble(ioat_chan, &phys_complete))
> > @@ -697,30 +753,50 @@ static void ioat_eh(struct ioatdma_chan
> > *ioat_chan)
> >                 break;
> >         }
> > 
> > +       if (chanerr & IOAT_CHANERR_RECOVER_MASK) {
> > +               if (chanerr & IOAT_CHANERR_READ_DATA_ERR) {
> > +                       desc->txd.result |= ERR_DMA_READ;
> 
> In [PATCH 1/5] ERR_DMA_XYZ values do not look appropriate for use as
> a bit mask.
> 
> +enum err_result_flags {
> +       ERR_DMA_NONE = 0,
> +       ERR_DMA_READ,
> +       ERR_DMA_WRITE,
> +       ERR_DMA_ABORT,
> +};
> 
> And in [PATCH 4/5] and [PATCH 5/5] txd.result is not used as a bit
> mask.
> 
> +               switch (txd->result) {
> +               case ERR_DMA_READ:
> +               case ERR_DMA_WRITE:
> +                       entry->errors++;
> +               case ERR_DMA_ABORT:
> +                       flags = DESC_ABORT_FLAG;
> +                       break;
> +               case ERR_DMA_NONE:
> +               default:
> +                       break;
> +               }
> 
> > +                       err_handled |= IOAT_CHANERR_READ_DATA_ERR;
> > +               } else if (chanerr & IOAT_CHANERR_WRITE_DATA_ERR) {
> > +                       desc->txd.result |= ERR_DMA_WRITE;
> > +                       err_handled |= IOAT_CHANERR_WRITE_DATA_ERR;
> > +               }
> > +
> > +               abort = true;
> > +       }
> > +
> >         /* fault on unhandled error or spurious halt */
> >         if (chanerr ^ err_handled || chanerr == 0) {
> 
> Not part of your change, but chanerr == 0 is confusing.  Doesn't that
> mean "no errors?"  No errors would make it hard to justify BUG() that
> follows.

Legacy code. This is the beginning of me cleaning out all the BUG() in
the driver and put in proper error handling. 

> 
> >                 dev_err(to_dev(ioat_chan), "%s: fatal error
> > (%x:%x)\n",
> >                         __func__, chanerr, err_handled);
> >                 BUG();
> > -       } else { /* cleanup the faulty descriptor */
> > -               tx = &desc->txd;
> > -               if (tx->cookie) {
> > -                       dma_cookie_complete(tx);
> > -                       dma_descriptor_unmap(tx);
> > -                       if (tx->callback) {
> > -                               tx->callback(tx->callback_param);
> > -                               tx->callback = NULL;
> > -                       }
> > -               }
> >         }
> > 
> > -       writel(chanerr, ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
> > -       pci_write_config_dword(pdev, IOAT_PCI_CHANERR_INT_OFFSET,
> > chanerr_int);
> > +       /* cleanup the faulty descriptor since we are continuing */
> > +       tx = &desc->txd;
> > +       if (tx->cookie) {
> > +               dma_cookie_complete(tx);
> > +               dma_descriptor_unmap(tx);
> > +               if (tx->callback) {
> > +                       tx->callback(tx->callback_param);
> > +                       tx->callback = NULL;
> > +               }
> > +       }
> > 
> >         /* mark faulting descriptor as complete */
> >         *ioat_chan->completion = desc->txd.phys;
> > 
> >         spin_lock_bh(&ioat_chan->prep_lock);
> > +       /* we need abort all descriptors */
> > +       if (abort) {
> > +               ioat_abort_descs(ioat_chan);
> > +               /* clean up the channel, we could be in weird state
> > */
> > +               ioat_reset_hw(ioat_chan);
> > +       }
> > +
> > +       writel(chanerr, ioat_chan->reg_base + IOAT_CHANERR_OFFSET);
> > +       pci_write_config_dword(pdev, IOAT_PCI_CHANERR_INT_OFFSET,
> > chanerr_int);
> > +
> >         ioat_restart_channel(ioat_chan);
> >         spin_unlock_bh(&ioat_chan->prep_lock);
> >  }
> > @@ -753,10 +829,25 @@ void ioat_timer_event(unsigned long data)
> >                 chanerr = readl(ioat_chan->reg_base +
> > IOAT_CHANERR_OFFSET);
> >                 dev_err(to_dev(ioat_chan), "%s: Channel halted
> > (%x)\n",
> >                         __func__, chanerr);
> > -               if (test_bit(IOAT_RUN, &ioat_chan->state))
> > -                       BUG_ON(is_ioat_bug(chanerr));
> > -               else /* we never got off the ground */
> > -                       return;
> > +               if (test_bit(IOAT_RUN, &ioat_chan->state)) {
> > +                       spin_lock_bh(&ioat_chan->cleanup_lock);
> > +                       spin_lock_bh(&ioat_chan->prep_lock);
> > +                       set_bit(IOAT_CHAN_DOWN, &ioat_chan->state);
> > +                       spin_unlock_bh(&ioat_chan->prep_lock);
> > +
> > +                       ioat_abort_descs(ioat_chan);
> > +                       dev_warn(to_dev(ioat_chan), "Reset
> > channel...\n");
> > +                       ioat_reset_hw(ioat_chan);
> > +                       dev_warn(to_dev(ioat_chan), "Restart
> > channel...\n");
> > +                       ioat_restart_channel(ioat_chan);
> > +
> > +                       spin_lock_bh(&ioat_chan->prep_lock);
> > +                       clear_bit(IOAT_CHAN_DOWN, &ioat_chan-
> > >state);
> > +                       spin_unlock_bh(&ioat_chan->prep_lock);
> > +                       spin_unlock_bh(&ioat_chan->cleanup_lock);
> > +               }
> > +
> > +               return;
> >         }
> > 
> >         spin_lock_bh(&ioat_chan->cleanup_lock);
> > @@ -780,14 +871,23 @@ void ioat_timer_event(unsigned long data)
> >                 u32 chanerr;
> > 
> >                 chanerr = readl(ioat_chan->reg_base +
> > IOAT_CHANERR_OFFSET);
> > -               dev_warn(to_dev(ioat_chan), "Restarting
> > channel...\n");
> >                 dev_warn(to_dev(ioat_chan), "CHANSTS: %#Lx CHANERR:
> > %#x\n",
> >                          status, chanerr);
> >                 dev_warn(to_dev(ioat_chan), "Active descriptors:
> > %d\n",
> >                          ioat_ring_active(ioat_chan));
> > 
> >                 spin_lock_bh(&ioat_chan->prep_lock);
> > +               set_bit(IOAT_CHAN_DOWN, &ioat_chan->state);
> > +               spin_unlock_bh(&ioat_chan->prep_lock);
> > +
> > +               ioat_abort_descs(ioat_chan);
> > +               dev_warn(to_dev(ioat_chan), "Resetting
> > channel...\n");
> > +               ioat_reset_hw(ioat_chan);
> > +               dev_warn(to_dev(ioat_chan), "Restarting
> > channel...\n");
> >                 ioat_restart_channel(ioat_chan);
> > +
> > +               spin_lock_bh(&ioat_chan->prep_lock);
> > +               clear_bit(IOAT_CHAN_DOWN, &ioat_chan->state);
> >                 spin_unlock_bh(&ioat_chan->prep_lock);
> >                 spin_unlock_bh(&ioat_chan->cleanup_lock);
> >                 return;
> > diff --git a/drivers/dma/ioat/registers.h
> > b/drivers/dma/ioat/registers.h
> > index 4994a36..faf20fa 100644
> > --- a/drivers/dma/ioat/registers.h
> > +++ b/drivers/dma/ioat/registers.h
> > @@ -233,6 +233,8 @@
> >  #define IOAT_CHANERR_DESCRIPTOR_COUNT_ERR      0x40000
> > 
> >  #define IOAT_CHANERR_HANDLE_MASK (IOAT_CHANERR_XOR_P_OR_CRC_ERR |
> > IOAT_CHANERR_XOR_Q_ERR)
> > +#define IOAT_CHANERR_RECOVER_MASK (IOAT_CHANERR_READ_DATA_ERR | \
> > +                                  IOAT_CHANERR_WRITE_DATA_ERR)
> > 
> >  #define IOAT_CHANERR_MASK_OFFSET               0x2C    /* 32-bit
> > Channel Error Register */
> > 
> > 
> 
> -- 

  reply	other threads:[~2016-06-28 16:12 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-06-27 23:09 [PATCH 0/5] dmaengine: ioatdma: DMA error reporting Dave Jiang
2016-06-27 23:09 ` [PATCH 1/5] dmaengine: Adding error handling flag Dave Jiang
2016-06-27 23:09 ` [PATCH 2/5] dmaengine: IOATDMA: Add error handling to driver Dave Jiang
2016-06-28  1:08   ` Allen Hubbe
2016-06-28 16:11     ` Jiang, Dave [this message]
2016-06-27 23:09 ` [PATCH 3/5] dmaengine: ioatdma: add error strings to chanerr output Dave Jiang
2016-06-27 23:09 ` [PATCH 4/5] ntb: add DMA error handling for TX DMA Dave Jiang
2016-06-28  1:53   ` Allen Hubbe
2016-06-28 16:15     ` Jiang, Dave
2016-06-27 23:10 ` [PATCH 5/5] ntb: add DMA error handling for RX DMA Dave Jiang

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=1467130291.3754.8.camel@intel.com \
    --to=dave.jiang@intel.com \
    --cc=allen.hubbe@emc.com \
    --cc=allenbh@gmail.com \
    --cc=dan.j.williams@intel.com \
    --cc=dmaengine@vger.kernel.org \
    --cc=jdmason@kudzu.us \
    --cc=linux-ntb@googlegroups.com \
    --cc=vinod.koul@intel.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox