grub-devel.gnu.org archive mirror
 help / color / mirror / Atom feed
From: Andrei Borzenkov <arvidjaar@gmail.com>
To: The development of GNU GRUB <grub-devel@gnu.org>, kernel-team@fb.com
Subject: Re: [PATCH 3/3] pxenet: process transmit interrupts when out of resources
Date: Sat, 12 Mar 2016 09:20:49 +0300	[thread overview]
Message-ID: <56E3B541.3050907@gmail.com> (raw)
In-Reply-To: <1457713715-31708-3-git-send-email-jbacik@fb.com>

11.03.2016 19:28, Josef Bacik пишет:
> We were hitting a problem in production where our bios based machines would
> sometimes timeout while pulling down either the kernel/initrd.  It turned out
> that sometimes we'd run out of transmit buffers and would then error out while
> sending packets.  This is because we don't actually have an interrupt handler in
> PXE, we just poll the card when we want to receive.  This works most of the
> time, but sometimes we end up with too many transmit interrupts pending and then
> we can't add new packets to the transmit buffers.
> 
> So rework the whole ISR logic to be able to be called from both transmit and
> receive.  If we get OUT_OF_RESOURCES while trying to transmit then we can go
> through and process the interrupts and hope that leaves us with space to retry
> the transmit.  Unfortunately this puts us in a place where we can trip over a
> RECEIVE interrupt, so we have to process that interrupt and leave a netbuff
> behind for the next call into recv.
> 
> With this patch we are now able to properly provision boxes suffering from this
> problem.  Thanks,
> 
> Signed-off-by: Josef Bacik <jbacik@fb.com>
> ---
>  grub-core/net/drivers/i386/pc/pxe.c | 155 +++++++++++++++++++++++++-----------
>  1 file changed, 108 insertions(+), 47 deletions(-)
> 
> diff --git a/grub-core/net/drivers/i386/pc/pxe.c b/grub-core/net/drivers/i386/pc/pxe.c
> index 3f4152d..57445b7 100644
> --- a/grub-core/net/drivers/i386/pc/pxe.c
> +++ b/grub-core/net/drivers/i386/pc/pxe.c
> @@ -166,16 +166,11 @@ grub_pxe_scan (void)
>    return bangpxe;
>  }
>  
> -static struct grub_net_buff *
> -grub_pxe_recv (struct grub_net_card *dev __attribute__ ((unused)))
> -{
> -  struct grub_pxe_undi_isr *isr;
> -  static int in_progress = 0;
> -  grub_uint8_t *ptr, *end;
> -  struct grub_net_buff *buf;
> -
> -  isr = (void *) GRUB_MEMORY_MACHINE_SCRATCH_ADDR;
> +static int in_progress = 0;
>  

You need to reset it in ->open or ->close.

> +static void
> +grub_pxe_process_isr (struct grub_pxe_undi_isr *isr)
> +{
>    if (!in_progress)
>      {
>        grub_memset (isr, 0, sizeof (*isr));
> @@ -186,13 +181,14 @@ grub_pxe_recv (struct grub_net_card *dev __attribute__ ((unused)))
>  	 breaks on intel cards.
>         */
>        if (isr->status)
> -	{
> -	  in_progress = 0;
> -	  return NULL;
> +        {
> +	  grub_dprintf("net", "problem pulling isr %d\n", (int)isr->status);

"pxe" would be better for dedicated debugging, "net" is too generic. We
can also set debug="pxe net" if needed. Also PXE spec lists status
values as hex, would be better to print it this way too.

> +	  return;
>  	}
>        grub_memset (isr, 0, sizeof (*isr));
>        isr->func_flag = GRUB_PXE_ISR_IN_PROCESS;
>        grub_pxe_call (GRUB_PXENV_UNDI_ISR, isr, pxe_rm_entry);
> +      in_progress = 1;
>      }
>    else
>      {
> @@ -200,18 +196,13 @@ grub_pxe_recv (struct grub_net_card *dev __attribute__ ((unused)))
>        isr->func_flag = GRUB_PXE_ISR_IN_GET_NEXT;
>        grub_pxe_call (GRUB_PXENV_UNDI_ISR, isr, pxe_rm_entry);
>      }
> +}
>  
> -  while (isr->func_flag != GRUB_PXE_ISR_OUT_RECEIVE)
> -    {
> -      if (isr->status || isr->func_flag == GRUB_PXE_ISR_OUT_DONE)
> -	{
> -	  in_progress = 0;
> -	  return NULL;
> -	}
> -      grub_memset (isr, 0, sizeof (*isr));
> -      isr->func_flag = GRUB_PXE_ISR_IN_GET_NEXT;
> -      grub_pxe_call (GRUB_PXENV_UNDI_ISR, isr, pxe_rm_entry);
> -    }
> +static struct grub_net_buff *
> +grub_pxe_make_net_buff (struct grub_pxe_undi_isr *isr)
> +{
> +  struct grub_net_buff *buf;
> +  grub_uint8_t *ptr, *end;
>  
>    buf = grub_netbuff_alloc (isr->frame_len + 2);
>    if (!buf)
> @@ -230,48 +221,118 @@ grub_pxe_recv (struct grub_net_card *dev __attribute__ ((unused)))
>    ptr += isr->buffer_len;
>    while (ptr < end)
>      {
> -      grub_memset (isr, 0, sizeof (*isr));
> -      isr->func_flag = GRUB_PXE_ISR_IN_GET_NEXT;
> -      grub_pxe_call (GRUB_PXENV_UNDI_ISR, isr, pxe_rm_entry);
> +      grub_pxe_process_isr (isr);
>        if (isr->status || isr->func_flag != GRUB_PXE_ISR_OUT_RECEIVE)
>  	{
> -	  in_progress = 1;
> +	  grub_dprintf("net", "half processed packet\n");
>  	  grub_netbuff_free (buf);
>  	  return NULL;
>  	}
> -
>        grub_memcpy (ptr, LINEAR (isr->buffer), isr->buffer_len);
>        ptr += isr->buffer_len;
>      }
> -  in_progress = 1;
> -
>    return buf;
>  }
>  
> +static struct grub_net_buff *
> +grub_pxe_recv (struct grub_net_card *dev)
> +{
> +  struct grub_pxe_undi_isr *isr;
> +
> +  if (dev->data)
> +    {
> +      struct grub_net_buff *buf = dev->data;
> +      grub_dprintf("net", "Pulling existing receive packet\n");
> +      dev->data = NULL;
> +      return buf;
> +    }
> +
> +  isr = (void *) GRUB_MEMORY_MACHINE_SCRATCH_ADDR;
> +  grub_pxe_process_isr (isr);
> +  while (isr->func_flag != GRUB_PXE_ISR_OUT_RECEIVE)
> +    {
> +      if (isr->status || isr->func_flag == GRUB_PXE_ISR_OUT_DONE)
> +	{
> +	  if (isr->status)
> +	    grub_dprintf("net", "error pulling next %d\n", (int)isr->status);
> +	  in_progress = 0;
> +	  return 0;
> +	}
> +      grub_memset (isr, 0, sizeof (*isr));
> +      isr->func_flag = GRUB_PXE_ISR_IN_GET_NEXT;
> +      grub_pxe_call (GRUB_PXENV_UNDI_ISR, isr, pxe_rm_entry);
> +    }
> +  return grub_pxe_make_net_buff (isr);
> +}
> +
>  static grub_err_t 
> -grub_pxe_send (struct grub_net_card *dev __attribute__ ((unused)),
> -	       struct grub_net_buff *pack)
> +grub_pxe_send (struct grub_net_card *dev, struct grub_net_buff *pack)
>  {
>    struct grub_pxe_undi_transmit *trans;
>    struct grub_pxe_undi_tbd *tbd;
>    char *buf;
>  
> -  trans = (void *) GRUB_MEMORY_MACHINE_SCRATCH_ADDR;
> -  grub_memset (trans, 0, sizeof (*trans));
> -  tbd = (void *) (GRUB_MEMORY_MACHINE_SCRATCH_ADDR + 128);
> -  grub_memset (tbd, 0, sizeof (*tbd));
> -  buf = (void *) (GRUB_MEMORY_MACHINE_SCRATCH_ADDR + 256);
> -  grub_memcpy (buf, pack->data, pack->tail - pack->data);
> -
> -  trans->tbd = SEGOFS ((grub_addr_t) tbd);
> -  trans->protocol = 0;
> -  tbd->len = pack->tail - pack->data;
> -  tbd->buf = SEGOFS ((grub_addr_t) buf);
> -
> -  grub_pxe_call (GRUB_PXENV_UNDI_TRANSMIT, trans, pxe_rm_entry);
> -  if (trans->status)
> -    return grub_error (GRUB_ERR_IO, N_("couldn't send network packet"));
> +  while (1)
> +    {
> +      int made_progress = 0;
> +
> +      trans = (void *) GRUB_MEMORY_MACHINE_SCRATCH_ADDR;
> +      grub_memset (trans, 0, sizeof (*trans));
> +      tbd = (void *) (GRUB_MEMORY_MACHINE_SCRATCH_ADDR + 128);
> +      grub_memset (tbd, 0, sizeof (*tbd));
> +      buf = (void *) (GRUB_MEMORY_MACHINE_SCRATCH_ADDR + 256);
> +      grub_memcpy (buf, pack->data, pack->tail - pack->data);
> +
> +      trans->tbd = SEGOFS ((grub_addr_t) tbd);
> +      trans->protocol = 0;
> +      tbd->len = pack->tail - pack->data;
> +      tbd->buf = SEGOFS ((grub_addr_t) buf);
> +
> +      grub_pxe_call (GRUB_PXENV_UNDI_TRANSMIT, trans, pxe_rm_entry);
> +      if (!trans->status)
> +	break;

This looks like the only terminating condition. What if hardware is
stuck and cannot make progress anymore and continues to return error here?

> +      if (trans->status == GRUB_PXENV_STATUS_OUT_OF_RESOURCES)
> +	{
> +	  struct grub_pxe_undi_isr *isr;
> +
> +	  grub_dprintf("net", "Out of transmit buffers, processing "
> +		       "interrupts\n");
> +	  /* Process any outstanding transmit interrupts. */
> +	  isr = (void *) GRUB_MEMORY_MACHINE_SCRATCH_ADDR;
> +	  do
> +	    {
> +	      grub_pxe_process_isr (isr);
> +	      if (isr->status)
> +		{
> +		  grub_dprintf("net", "process interrupts errored %d,"
> +			       "made_progress %d\n", (int)isr->status, made_progress);
> +		  if (made_progress)
> +		    break;
> +		  else
> +		    goto out_err;
> +		}
> +	      if (isr->func_flag == GRUB_PXE_ISR_OUT_DONE)
> +		in_progress = 0;
> +	      made_progress = 1;
> +	    } while (isr->func_flag == GRUB_PXE_ISR_OUT_TRANSMIT);
> +
> +	  /* If we had a receive interrupt in the queue we need to copy this
> +	     buffer out otherwise we'll lose it. */
> +	  if (isr->func_flag == GRUB_PXE_ISR_OUT_RECEIVE)

It probably should check isr->status, we come here also if we got an
error previously. Such packets are ignored everywhere else.

> +	    {
> +	      if (dev->data)
> +		grub_dprintf("net", "dropping packet, already have a "
> +			     "pending packet.\n");
> +	      else
> +	        dev->data = grub_pxe_make_net_buff (isr);
> +	    }
> +	}
> +      else
> +        goto out_err;
> +    }
>    return 0;
> +out_err:
> +  return grub_error (GRUB_ERR_IO, N_("couldn't send network packet"));
>  }
>  
>  static void
> 



  reply	other threads:[~2016-03-12  6:21 UTC|newest]

Thread overview: 13+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2016-03-11 16:28 [PATCH 1/3] push/pop errno in initrd read file path Josef Bacik
2016-03-11 16:28 ` [PATCH 2/3] tcp: add a dprintf for opening tcp connections Josef Bacik
2016-03-11 16:28 ` [PATCH 3/3] pxenet: process transmit interrupts when out of resources Josef Bacik
2016-03-12  6:20   ` Andrei Borzenkov [this message]
2016-03-11 17:23 ` [PATCH 1/3] push/pop errno in initrd read file path Vladimir 'phcoder' Serbinenko
2016-03-11 18:13   ` Josef Bacik
2016-03-11 19:34     ` Vladimir 'phcoder' Serbinenko
2016-03-11 20:34       ` Josef Bacik
2016-03-11 22:00         ` Vladimir 'phcoder' Serbinenko
2016-03-12  7:01           ` Andrei Borzenkov
2016-03-15 17:26             ` Vladimir 'phcoder' Serbinenko
2016-11-10 12:49 ` Daniel Kiper
2016-11-18 14:51   ` Josef Bacik

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=56E3B541.3050907@gmail.com \
    --to=arvidjaar@gmail.com \
    --cc=grub-devel@gnu.org \
    --cc=kernel-team@fb.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;
as well as URLs for NNTP newsgroup(s).