qemu-devel.nongnu.org archive mirror
 help / color / mirror / Atom feed
From: Alexey Kardashevskiy <aik@ozlabs.ru>
To: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>,
	qemu-ppc@nongnu.org, qemu-devel@nongnu.org
Subject: Re: [Qemu-devel] [PATCH 2/2] spapr: Fix stale HTAB during live migration (TCG)
Date: Thu, 13 Nov 2014 18:12:22 +1100	[thread overview]
Message-ID: <546459D6.1040706@ozlabs.ru> (raw)
In-Reply-To: <1415168221-2324-3-git-send-email-sam.mj@au1.ibm.com>

On 11/05/2014 05:17 PM, Samuel Mendoza-Jonas wrote:
> If a TCG guest reboots during a running migration HTAB entries are not
> marked dirty, and the destination boots with an invalid HTAB.
> 
> When a reboot occurs reset the state of HTAB migration, and explicitly
> inform the destination of invalid entries.
> 
> Signed-off-by: Samuel Mendoza-Jonas <sam.mj@au1.ibm.com>
> ---
>  hw/ppc/spapr.c         | 59 +++++++++++++++++++++++++++++++++++---------------
>  include/hw/ppc/spapr.h |  1 +
>  2 files changed, 42 insertions(+), 18 deletions(-)
> 
> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> index 1610c28..9f419e8 100644
> --- a/hw/ppc/spapr.c
> +++ b/hw/ppc/spapr.c
> @@ -829,26 +829,30 @@ static void spapr_reset_htab(sPAPREnvironment *spapr)
>  
>      shift = kvmppc_reset_htab(spapr->htab_shift);
>  
> +    pthread_mutex_lock(&spapr->htab_mutex);
>      if (shift > 0) {
>          /* Kernel handles htab, we don't need to allocate one */
>          spapr->htab_shift = shift;
>          kvmppc_kern_htab = true;
>  
>          /* Tell readers to update their file descriptor */
> -        pthread_mutex_lock(&spapr->htab_mutex);
>          if (spapr->htab_fd > 0) {
>              spapr->htab_fd_stale = true;
>          }
> -        pthread_mutex_unlock(&spapr->htab_mutex);
>      } else {
>          if (!spapr->htab) {
>              /* Allocate an htab if we don't yet have one */
>              spapr->htab = qemu_memalign(HTAB_SIZE(spapr), HTAB_SIZE(spapr));
> +        } else {
> +            spapr->htab_mig_full = true;
> +            spapr->htab_first_pass = true;
> +            spapr->htab_save_index = 0;
>          }
>  
>          /* And clear it */
>          memset(spapr->htab, 0, HTAB_SIZE(spapr));
>      }
> +    pthread_mutex_unlock(&spapr->htab_mutex);


The pthread_mutex_(un)lock things from the chunk above should go to the
previous patch.


>  
>      /* Update the RMA size if necessary */
>      if (spapr->vrma_adjust) {
> @@ -1019,6 +1023,7 @@ static int htab_save_setup(QEMUFile *f, void *opaque)
>          pthread_mutex_lock(&spapr->htab_mutex);
>          spapr->htab_fd = kvmppc_get_htab_fd(false);
>          spapr->htab_fd_stale = false;
> +        spapr->htab_mig_full = false;
>          pthread_mutex_unlock(&spapr->htab_mutex);
>          if (spapr->htab_fd < 0) {
>              fprintf(stderr, "Unable to open fd for reading hash table from KVM: %s\n",
> @@ -1034,6 +1039,7 @@ static int htab_save_setup(QEMUFile *f, void *opaque)
>  static void htab_save_first_pass(QEMUFile *f, sPAPREnvironment *spapr,
>                                   int64_t max_ns)
>  {
> +    bool final = max_ns < 0;
>      int htabslots = HTAB_SIZE(spapr) / HASH_PTE_SIZE_64;
>      int index = spapr->htab_save_index;
>      int64_t starttime = qemu_clock_get_ns(QEMU_CLOCK_REALTIME);
> @@ -1041,33 +1047,40 @@ static void htab_save_first_pass(QEMUFile *f, sPAPREnvironment *spapr,
>      assert(spapr->htab_first_pass);
>  
>      do {
> -        int chunkstart;
> +        int chunkstart, invalidstart;
>  
> -        /* Consume invalid HPTEs */
> -        while ((index < htabslots)
> -               && !HPTE_VALID(HPTE(spapr->htab, index))) {


Nit: in most places in qemu "&&" would be in the end of the previos line :)


> +        chunkstart = index;
> +        /* Consume valid HPTEs */
> +        while ((index < htabslots && (index - chunkstart < USHRT_MAX))
> +               && HPTE_VALID(HPTE(spapr->htab, index))) {
>              index++;
>              CLEAN_HPTE(HPTE(spapr->htab, index));
>          }
>  
> -        /* Consume valid HPTEs */
> -        chunkstart = index;
> -        while ((index < htabslots)
> -               && HPTE_VALID(HPTE(spapr->htab, index))) {
> +        invalidstart = index;
> +        /* Consume invalid HPTEs */
> +        while ((index < htabslots && (index - invalidstart < USHRT_MAX))
> +               && !HPTE_VALID(HPTE(spapr->htab, index))) {
>              index++;
>              CLEAN_HPTE(HPTE(spapr->htab, index));
>          }
>  
> -        if (index > chunkstart) {
> -            int n_valid = index - chunkstart;
> +        /* Avoid writing an end marker (0,0,0) */
> +        if (index > chunkstart
> +               && !(chunkstart == invalidstart && !spapr->htab_mig_full)) {
> +            int n_valid = invalidstart - chunkstart;
> +            /* If a reset has occured we must explicitly overwrite the HTAB
> +             * of the destination */
> +            int n_invalid = spapr->htab_mig_full ? index - invalidstart : 0;
>  
>              qemu_put_be32(f, chunkstart);
>              qemu_put_be16(f, n_valid);
> -            qemu_put_be16(f, 0);
> +            qemu_put_be16(f, n_invalid);
>              qemu_put_buffer(f, HPTE(spapr->htab, chunkstart),
>                              HASH_PTE_SIZE_64 * n_valid);
>  
> -            if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns) {
> +            if ((qemu_clock_get_ns(QEMU_CLOCK_REALTIME) - starttime) > max_ns
> +                   && !final) {
>                  break;
>              }
>          }
> @@ -1182,10 +1195,14 @@ static int htab_save_iterate(QEMUFile *f, void *opaque)
>          if (rc < 0) {
>              return rc;
>          }
> -    } else  if (spapr->htab_first_pass) {
> -        htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
>      } else {
> -        rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
> +        pthread_mutex_lock(&spapr->htab_mutex);
> +        if (spapr->htab_first_pass) {
> +            htab_save_first_pass(f, spapr, MAX_ITERATION_NS);
> +        } else {
> +            rc = htab_save_later_pass(f, spapr, MAX_ITERATION_NS);
> +        }
> +        pthread_mutex_unlock(&spapr->htab_mutex);
>      }
>  
>      /* End marker */
> @@ -1220,7 +1237,13 @@ static int htab_save_complete(QEMUFile *f, void *opaque)
>          close(spapr->htab_fd);
>          spapr->htab_fd = -1;
>      } else {
> -        htab_save_later_pass(f, spapr, -1);
> +        pthread_mutex_lock(&spapr->htab_mutex);
> +        if (spapr->htab_first_pass) {
> +            htab_save_first_pass(f, spapr, -1);
> +        } else {
> +            htab_save_later_pass(f, spapr, -1);
> +        }
> +        pthread_mutex_unlock(&spapr->htab_mutex);
>      }
>  
>      /* End marker */
> diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
> index 5e29bec..ee95459 100644
> --- a/include/hw/ppc/spapr.h
> +++ b/include/hw/ppc/spapr.h
> @@ -38,6 +38,7 @@ typedef struct sPAPREnvironment {
>      bool htab_first_pass;
>      int htab_fd;
>      bool htab_fd_stale;
> +    bool htab_mig_full;

Not sure what @htab_mig_full stands for exactly. Please explain it in the
commit log.


>      pthread_mutex_t htab_mutex;
>  } sPAPREnvironment;
>  
> 


-- 
Alexey

      parent reply	other threads:[~2014-11-13  7:12 UTC|newest]

Thread overview: 10+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2014-11-05  6:16 [Qemu-devel] [PATCH 0/2] spapr: Fix stale HTAB during live migration Samuel Mendoza-Jonas
2014-11-05  6:17 ` [Qemu-devel] [PATCH 1/2] spapr: Fix stale HTAB during live migration (KVM) Samuel Mendoza-Jonas
2014-11-05  7:57   ` [Qemu-devel] [Qemu-ppc] " Alexander Graf
2014-11-05 21:56     ` Samuel Mendoza-Jonas
2014-11-13  6:59   ` [Qemu-devel] " Alexey Kardashevskiy
2014-11-05  6:17 ` [Qemu-devel] [PATCH 2/2] spapr: Fix stale HTAB during live migration (TCG) Samuel Mendoza-Jonas
2014-11-05  8:05   ` [Qemu-devel] [Qemu-ppc] " Alexander Graf
2014-11-05 22:00     ` Samuel Mendoza-Jonas
2014-11-05 22:04       ` Alexander Graf
2014-11-13  7:12   ` Alexey Kardashevskiy [this message]

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=546459D6.1040706@ozlabs.ru \
    --to=aik@ozlabs.ru \
    --cc=qemu-devel@nongnu.org \
    --cc=qemu-ppc@nongnu.org \
    --cc=sam.mj@au1.ibm.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).