[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Qemu-devel] [PATCH V4] spapr: Fix stale HTAB during live migration
From: |
Alexey Kardashevskiy |
Subject: |
Re: [Qemu-devel] [PATCH V4] spapr: Fix stale HTAB during live migration |
Date: |
Wed, 20 Aug 2014 16:45:49 +1000 |
User-agent: |
Mozilla/5.0 (X11; Linux i686 on x86_64; rv:31.0) Gecko/20100101 Thunderbird/31.0 |
On 08/20/2014 12:13 PM, Samuel Mendoza-Jonas wrote:
> If a guest reboots during a running migration, changes to the
> hash page table are not necessarily updated on the destination.
> Opening a new file descriptor to the HTAB forces the migration
> handler to resend the entire table.
>
> Signed-off-by: Samuel Mendoza-Jonas <address@hidden>
> ---
> Changes in v4: Readability: need_reset to htab_fd_stale
> Add spapr_check_htab_fd() and use error_report()
> Changes in v3: Pointed out by David, htab_save_iterate could
> potentially try to read before htab_fd is open again.
> Leave opening the fd to the functions trying to read.
> Changes in v2: Forgot check on kvmppc_get_htab_fd return value
>
> hw/ppc/spapr.c | 37 +++++++++++++++++++++++++++++++++++++
> include/hw/ppc/spapr.h | 1 +
> 2 files changed, 38 insertions(+)
>
> diff --git a/hw/ppc/spapr.c b/hw/ppc/spapr.c
> index 3a6d26d..68f97a9 100644
> --- a/hw/ppc/spapr.c
> +++ b/hw/ppc/spapr.c
> @@ -997,6 +997,11 @@ static void spapr_reset_htab(sPAPREnvironment *spapr)
> /* Kernel handles htab, we don't need to allocate one */
> spapr->htab_shift = shift;
> kvmppc_kern_htab = true;
> +
> + /* Check if we are overlapping a migration */
> + if (spapr->htab_fd > 0) {
> + spapr->htab_fd_stale = true;
> + }
> } else {
> if (!spapr->htab) {
> /* Allocate an htab if we don't yet have one */
> @@ -1014,6 +1019,27 @@ static void spapr_reset_htab(sPAPREnvironment *spapr)
> }
> }
>
> +/* A guest reset will cause spapr->htab_fd to become stale if being used.
> + * Reopen the file descriptor to make sure the whole HTAB is properly read.
> + */
> +static int spapr_check_htab_fd(sPAPREnvironment *spapr)
> +{
> + int rc = 0;
> +
> + if (atomic_cmpxchg(&spapr->htab_fd_stale, true, false) == true) {
I was wrong about xchg :)
You have to loop atomic_cmpxchg() if it returns false. Second, "if" and "==
true" does not look nice, I'd drop "== true". Also this is (probably?)
missing a memory barrier between setting and clearing @htab_fd_stale.
Paul suggested using pthread_mutex_lock() + pthread_mutex_unlock() where we
set and clear the flag as we do not expect millions of those and this is
way simpler.
> + close(spapr->htab_fd);
> + spapr->htab_fd = kvmppc_get_htab_fd(false);
> +
> + if (spapr->htab_fd < 0) {
> + error_report("Unable to open fd for reading hash table from KVM:
> "
> + "%s", strerror(errno));
> + rc = -1;;
Double semicolon.
> + }
> + }
> +
> + return rc;
> +}
> +
> static void ppc_spapr_reset(void)
> {
> PowerPCCPU *first_ppc_cpu;
> @@ -1156,6 +1182,7 @@ static int htab_save_setup(QEMUFile *f, void *opaque)
> } else {
> assert(kvm_enabled());
>
> + spapr->htab_fd_stale = false;
> spapr->htab_fd = kvmppc_get_htab_fd(false);
> if (spapr->htab_fd < 0) {
> fprintf(stderr, "Unable to open fd for reading hash table from
> KVM: %s\n",
> @@ -1309,6 +1336,11 @@ static int htab_save_iterate(QEMUFile *f, void *opaque)
> if (!spapr->htab) {
> assert(kvm_enabled());
>
> + rc = spapr_check_htab_fd(spapr);
> + if (rc < 0) {
> + return rc;
> + }
> +
> rc = kvmppc_save_htab(f, spapr->htab_fd,
> MAX_KVM_BUF_SIZE, MAX_ITERATION_NS);
> if (rc < 0) {
> @@ -1340,6 +1372,11 @@ static int htab_save_complete(QEMUFile *f, void
> *opaque)
>
> assert(kvm_enabled());
>
> + rc = spapr_check_htab_fd(spapr);
> + if (rc < 0) {
> + return rc;
> + }
> +
> rc = kvmppc_save_htab(f, spapr->htab_fd, MAX_KVM_BUF_SIZE, -1);
> if (rc < 0) {
> return rc;
> diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
> index 0c2e3c5..0421d9a 100644
> --- a/include/hw/ppc/spapr.h
> +++ b/include/hw/ppc/spapr.h
> @@ -71,6 +71,7 @@ typedef struct sPAPREnvironment {
> int htab_save_index;
> bool htab_first_pass;
> int htab_fd;
> + bool htab_fd_stale;
>
> /* state for Dynamic Reconfiguration Connectors */
> sPAPRDrcEntry drc_table[SPAPR_DRC_TABLE_SIZE];
>
--
Alexey