* [Qemu-devel] [PATCH] migration: ram_handle_compressed
@ 2013-09-18 10:07 Isaku Yamahata
2013-09-18 12:08 ` Juan Quintela
2013-09-18 12:30 ` Michael R. Hines
0 siblings, 2 replies; 4+ messages in thread
From: Isaku Yamahata @ 2013-09-18 10:07 UTC (permalink / raw)
To: qemu-devel; +Cc: owasserm, pbonzini, chegu_vinod, mrhines, quintela
ram_handle_compressed() should be aware size > TARGET_PAGE_SIZE
migration-rdma can call it with larger size.
Signed-off-by: Isaku Yamahata <yamahata@private.email.ne.jp>
---
arch_init.c | 21 ++++++++++++++-------
1 file changed, 14 insertions(+), 7 deletions(-)
diff --git a/arch_init.c b/arch_init.c
index e47e139..64c81b0 100644
--- a/arch_init.c
+++ b/arch_init.c
@@ -844,15 +844,22 @@ static inline void *host_from_stream_offset(QEMUFile *f,
*/
void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
{
- if (ch != 0 || !is_zero_page(host)) {
- memset(host, ch, size);
+ uint64_t pagesize = getpagesize();
+ while (size > 0) {
+ uint64_t length = MIN(pagesize, size);
+
+ if (ch !=0 || buffer_find_nonzero_offset(host, length) != length) {
+ memset(host, ch, length);
#ifndef _WIN32
- if (ch == 0 &&
- (!kvm_enabled() || kvm_has_sync_mmu()) &&
- getpagesize() <= TARGET_PAGE_SIZE) {
- qemu_madvise(host, TARGET_PAGE_SIZE, QEMU_MADV_DONTNEED);
- }
+ if (ch == 0 &&
+ (!kvm_enabled() || kvm_has_sync_mmu()) && pagesize <= length) {
+ qemu_madvise(host, size, QEMU_MADV_DONTNEED);
+ }
#endif
+ }
+
+ size -= length;
+ host += length;
}
}
--
1.7.10.4
^ permalink raw reply related [flat|nested] 4+ messages in thread* Re: [Qemu-devel] [PATCH] migration: ram_handle_compressed
2013-09-18 10:07 [Qemu-devel] [PATCH] migration: ram_handle_compressed Isaku Yamahata
@ 2013-09-18 12:08 ` Juan Quintela
2013-09-20 7:43 ` Isaku Yamahata
2013-09-18 12:30 ` Michael R. Hines
1 sibling, 1 reply; 4+ messages in thread
From: Juan Quintela @ 2013-09-18 12:08 UTC (permalink / raw)
To: Isaku Yamahata; +Cc: owasserm, chegu_vinod, mrhines, qemu-devel, pbonzini
Isaku Yamahata <yamahata@private.email.ne.jp> wrote:
> ram_handle_compressed() should be aware size > TARGET_PAGE_SIZE
> migration-rdma can call it with larger size.
>
> Signed-off-by: Isaku Yamahata <yamahata@private.email.ne.jp>
> ---
> arch_init.c | 21 ++++++++++++++-------
> 1 file changed, 14 insertions(+), 7 deletions(-)
>
> diff --git a/arch_init.c b/arch_init.c
> index e47e139..64c81b0 100644
> --- a/arch_init.c
> +++ b/arch_init.c
> @@ -844,15 +844,22 @@ static inline void *host_from_stream_offset(QEMUFile *f,
> */
> void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
> {
> - if (ch != 0 || !is_zero_page(host)) {
> - memset(host, ch, size);
I assume that size is always a multiple of pagesize, right?
> + uint64_t pagesize = getpagesize();
> + while (size > 0) {
> + uint64_t length = MIN(pagesize, size);
> +
> + if (ch !=0 || buffer_find_nonzero_offset(host, length) != length) {
This is a different optimization, we were't doing this before.
> + memset(host, ch, length);
> #ifndef _WIN32
> - if (ch == 0 &&
> - (!kvm_enabled() || kvm_has_sync_mmu()) &&
> - getpagesize() <= TARGET_PAGE_SIZE) {
> - qemu_madvise(host, TARGET_PAGE_SIZE, QEMU_MADV_DONTNEED);
> - }
> + if (ch == 0 &&
> + (!kvm_enabled() || kvm_has_sync_mmu()) && pagesize <= length) {
> + qemu_madvise(host, size, QEMU_MADV_DONTNEED);
Here don't we need s/size/length/?
> + }
> #endif
> + }
> +
> + size -= length;
> + host += length;
> }
> }
^ permalink raw reply [flat|nested] 4+ messages in thread* Re: [Qemu-devel] [PATCH] migration: ram_handle_compressed
2013-09-18 12:08 ` Juan Quintela
@ 2013-09-20 7:43 ` Isaku Yamahata
0 siblings, 0 replies; 4+ messages in thread
From: Isaku Yamahata @ 2013-09-20 7:43 UTC (permalink / raw)
To: Juan Quintela
Cc: Isaku Yamahata, qemu-devel, owasserm, mrhines, pbonzini,
chegu_vinod
On Wed, Sep 18, 2013 at 02:08:40PM +0200, Juan Quintela wrote:
> Isaku Yamahata <yamahata@private.email.ne.jp> wrote:
> > ram_handle_compressed() should be aware size > TARGET_PAGE_SIZE
> > migration-rdma can call it with larger size.
> >
> > Signed-off-by: Isaku Yamahata <yamahata@private.email.ne.jp>
> > ---
> > arch_init.c | 21 ++++++++++++++-------
> > 1 file changed, 14 insertions(+), 7 deletions(-)
> >
> > diff --git a/arch_init.c b/arch_init.c
> > index e47e139..64c81b0 100644
> > --- a/arch_init.c
> > +++ b/arch_init.c
> > @@ -844,15 +844,22 @@ static inline void *host_from_stream_offset(QEMUFile *f,
> > */
> > void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
> > {
> > - if (ch != 0 || !is_zero_page(host)) {
> > - memset(host, ch, size);
>
> I assume that size is always a multiple of pagesize, right?
Right. The issues is that is_zero_page() checks only first page.
The following page needs to be checked.
>
> > + uint64_t pagesize = getpagesize();
> > + while (size > 0) {
> > + uint64_t length = MIN(pagesize, size);
> > +
> > + if (ch !=0 || buffer_find_nonzero_offset(host, length) != length) {
>
> This is a different optimization, we were't doing this before.
Okay, I'll drop loop and use size.
thanks,
> > + memset(host, ch, length);
> > #ifndef _WIN32
> > - if (ch == 0 &&
> > - (!kvm_enabled() || kvm_has_sync_mmu()) &&
> > - getpagesize() <= TARGET_PAGE_SIZE) {
> > - qemu_madvise(host, TARGET_PAGE_SIZE, QEMU_MADV_DONTNEED);
> > - }
> > + if (ch == 0 &&
> > + (!kvm_enabled() || kvm_has_sync_mmu()) && pagesize <= length) {
> > + qemu_madvise(host, size, QEMU_MADV_DONTNEED);
>
> Here don't we need s/size/length/?
>
> > + }
> > #endif
> > + }
> > +
> > + size -= length;
> > + host += length;
> > }
> > }
>
--
yamahata
^ permalink raw reply [flat|nested] 4+ messages in thread
* Re: [Qemu-devel] [PATCH] migration: ram_handle_compressed
2013-09-18 10:07 [Qemu-devel] [PATCH] migration: ram_handle_compressed Isaku Yamahata
2013-09-18 12:08 ` Juan Quintela
@ 2013-09-18 12:30 ` Michael R. Hines
1 sibling, 0 replies; 4+ messages in thread
From: Michael R. Hines @ 2013-09-18 12:30 UTC (permalink / raw)
To: Isaku Yamahata
Cc: quintela, qemu-devel, owasserm, mrhines, pbonzini, chegu_vinod
On 09/18/2013 06:07 AM, Isaku Yamahata wrote:
> ram_handle_compressed() should be aware size > TARGET_PAGE_SIZE
> migration-rdma can call it with larger size.
>
> Signed-off-by: Isaku Yamahata <yamahata@private.email.ne.jp>
> ---
> arch_init.c | 21 ++++++++++++++-------
> 1 file changed, 14 insertions(+), 7 deletions(-)
>
> diff --git a/arch_init.c b/arch_init.c
> index e47e139..64c81b0 100644
> --- a/arch_init.c
> +++ b/arch_init.c
> @@ -844,15 +844,22 @@ static inline void *host_from_stream_offset(QEMUFile *f,
> */
> void ram_handle_compressed(void *host, uint8_t ch, uint64_t size)
> {
> - if (ch != 0 || !is_zero_page(host)) {
> - memset(host, ch, size);
> + uint64_t pagesize = getpagesize();
> + while (size > 0) {
> + uint64_t length = MIN(pagesize, size);
> +
> + if (ch !=0 || buffer_find_nonzero_offset(host, length) != length) {
> + memset(host, ch, length);
> #ifndef _WIN32
> - if (ch == 0 &&
> - (!kvm_enabled() || kvm_has_sync_mmu()) &&
> - getpagesize() <= TARGET_PAGE_SIZE) {
> - qemu_madvise(host, TARGET_PAGE_SIZE, QEMU_MADV_DONTNEED);
> - }
> + if (ch == 0 &&
> + (!kvm_enabled() || kvm_has_sync_mmu()) && pagesize <= length) {
> + qemu_madvise(host, size, QEMU_MADV_DONTNEED);
> + }
> #endif
> + }
> +
> + size -= length;
> + host += length;
> }
> }
>
What is the reason for breaking the madvise() into smaller pieces?
- Michael
^ permalink raw reply [flat|nested] 4+ messages in thread
end of thread, other threads:[~2013-09-20 7:43 UTC | newest]
Thread overview: 4+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2013-09-18 10:07 [Qemu-devel] [PATCH] migration: ram_handle_compressed Isaku Yamahata
2013-09-18 12:08 ` Juan Quintela
2013-09-20 7:43 ` Isaku Yamahata
2013-09-18 12:30 ` Michael R. Hines
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).