| Message ID | 1465813009-21390-2-git-send-email-liang.z.li@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 13.06.2016 12:16, Liang Li wrote: > The implementation of the current virtio-balloon is not very efficient, > Bellow is test result of time spends on inflating the balloon to 3GB of > a 4GB idle guest: > > a. allocating pages (6.5%, 103ms) > b. sending PFNs to host (68.3%, 787ms) > c. address translation (6.1%, 96ms) > d. madvise (19%, 300ms) > > It takes about 1577ms for the whole inflating process to complete. The > test shows that the bottle neck is the stage b and stage d. > > If using a bitmap to send the page info instead of the PFNs, we can > reduce the overhead spends on stage b quite a lot. Furthermore, it's > possible to do the address translation and do the madvise with a bulk > of pages, instead of the current page per page way, so the overhead of > stage c and stage d can also be reduced a lot. > > This patch is the QEMU side implementation which is intended to speed > up the inflating & deflating process by adding a new feature to the > virtio-balloon device. And now, inflating the balloon to 3GB of a 4GB > idle guest only takes 210ms, it's about 8 times as fast as before. > > TODO: optimize stage a by allocating/freeing a chunk of pages instead > of a single page at a time. > > Signed-off-by: Liang Li <liang.z.li@intel.com> > --- > hw/virtio/virtio-balloon.c | 159 ++++++++++++++++++++---- > include/standard-headers/linux/virtio_balloon.h | 1 + > 2 files changed, 139 insertions(+), 21 deletions(-) > > diff --git a/hw/virtio/virtio-balloon.c b/hw/virtio/virtio-balloon.c > index 8c15e09..8cf74c2 100644 > --- a/hw/virtio/virtio-balloon.c > +++ b/hw/virtio/virtio-balloon.c > @@ -47,6 +47,76 @@ static void balloon_page(void *addr, int deflate) > #endif > } > > +static void do_balloon_bulk_pages(ram_addr_t base_pfn, int page_shift, > + unsigned long len, bool deflate) > +{ > + ram_addr_t size, processed, chunk, base; > + void *addr; > + MemoryRegionSection section = {.mr = NULL}; > + > + size = (len << page_shift); > + base = (base_pfn << page_shift); > + > + for (processed = 0; processed < size; processed += chunk) { > + chunk = size - processed; > + while (chunk >= TARGET_PAGE_SIZE) { > + section = memory_region_find(get_system_memory(), > + base + processed, chunk); > + if (!section.mr) { > + chunk = QEMU_ALIGN_DOWN(chunk / 2, TARGET_PAGE_SIZE); > + } else { > + break; > + } > + } > + > + if (section.mr && > + (int128_nz(section.size) && memory_region_is_ram(section.mr))) { > + addr = section.offset_within_region + > + memory_region_get_ram_ptr(section.mr); > + qemu_madvise(addr, chunk, > + deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED); > + } else { > + fprintf(stderr, "can't find the chunk, skip\n"); Please try to avoid new fprintf(stderr, ...) in the QEMU sources. Use error_report(...) or in this case maybe rather qemu_log_mask(LOG_GUEST_ERROR, ...) instead, and try to use a more reasonable error message (e.g. that it is clear that the error happened in the balloon code). > + chunk = TARGET_PAGE_SIZE; > + } > + } > +} > + > +static void balloon_bulk_pages(ram_addr_t base_pfn, unsigned long *bitmap, > + unsigned long len, int page_shift, bool deflate) > +{ > +#if defined(__linux__) Why do you need this #if here? > + unsigned long end = len * 8; > + unsigned long current = 0; > + > + if (!qemu_balloon_is_inhibited() && (!kvm_enabled() || > + kvm_has_sync_mmu())) { > + while (current < end) { > + unsigned long one = find_next_bit(bitmap, end, current); > + > + if (one < end) { > + unsigned long zero = find_next_zero_bit(bitmap, end, one + 1); > + unsigned long page_length; > + > + if (zero >= end) { > + page_length = end - one; > + } else { > + page_length = zero - one; > + } > + > + if (page_length) { > + do_balloon_bulk_pages(base_pfn + one, page_shift, > + page_length, deflate); > + } > + current = one + page_length; > + } else { > + current = one; > + } > + } > + } > +#endif > +} Thomas -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
> Subject: Re: [QEMU 1/7] balloon: speed up inflating & deflating process > > On 13.06.2016 12:16, Liang Li wrote: > > The implementation of the current virtio-balloon is not very > > efficient, Bellow is test result of time spends on inflating the > > balloon to 3GB of a 4GB idle guest: > > > > a. allocating pages (6.5%, 103ms) > > b. sending PFNs to host (68.3%, 787ms) c. address translation (6.1%, > > 96ms) d. madvise (19%, 300ms) > > > > It takes about 1577ms for the whole inflating process to complete. The > > test shows that the bottle neck is the stage b and stage d. > > > > If using a bitmap to send the page info instead of the PFNs, we can > > reduce the overhead spends on stage b quite a lot. Furthermore, it's > > possible to do the address translation and do the madvise with a bulk > > of pages, instead of the current page per page way, so the overhead of > > stage c and stage d can also be reduced a lot. > > > > This patch is the QEMU side implementation which is intended to speed > > up the inflating & deflating process by adding a new feature to the > > virtio-balloon device. And now, inflating the balloon to 3GB of a 4GB > > idle guest only takes 210ms, it's about 8 times as fast as before. > > > > TODO: optimize stage a by allocating/freeing a chunk of pages instead > > of a single page at a time. > > > > Signed-off-by: Liang Li <liang.z.li@intel.com> > > --- > > hw/virtio/virtio-balloon.c | 159 ++++++++++++++++++++---- > > include/standard-headers/linux/virtio_balloon.h | 1 + > > 2 files changed, 139 insertions(+), 21 deletions(-) > > > > diff --git a/hw/virtio/virtio-balloon.c b/hw/virtio/virtio-balloon.c > > index 8c15e09..8cf74c2 100644 > > --- a/hw/virtio/virtio-balloon.c > > +++ b/hw/virtio/virtio-balloon.c > > @@ -47,6 +47,76 @@ static void balloon_page(void *addr, int deflate) > > #endif } > > > > +static void do_balloon_bulk_pages(ram_addr_t base_pfn, int page_shift, > > + unsigned long len, bool deflate) { > > + ram_addr_t size, processed, chunk, base; > > + void *addr; > > + MemoryRegionSection section = {.mr = NULL}; > > + > > + size = (len << page_shift); > > + base = (base_pfn << page_shift); > > + > > + for (processed = 0; processed < size; processed += chunk) { > > + chunk = size - processed; > > + while (chunk >= TARGET_PAGE_SIZE) { > > + section = memory_region_find(get_system_memory(), > > + base + processed, chunk); > > + if (!section.mr) { > > + chunk = QEMU_ALIGN_DOWN(chunk / 2, TARGET_PAGE_SIZE); > > + } else { > > + break; > > + } > > + } > > + > > + if (section.mr && > > + (int128_nz(section.size) && memory_region_is_ram(section.mr))) { > > + addr = section.offset_within_region + > > + memory_region_get_ram_ptr(section.mr); > > + qemu_madvise(addr, chunk, > > + deflate ? QEMU_MADV_WILLNEED : > QEMU_MADV_DONTNEED); > > + } else { > > + fprintf(stderr, "can't find the chunk, skip\n"); > > Please try to avoid new fprintf(stderr, ...) in the QEMU sources. > Use error_report(...) or in this case maybe rather > qemu_log_mask(LOG_GUEST_ERROR, ...) instead, and try to use a more > reasonable error message (e.g. that it is clear that the error happened in the > balloon code). > Indeed, the error message is no good, will change in next version. > > + chunk = TARGET_PAGE_SIZE; > > + } > > + } > > +} > > + > > +static void balloon_bulk_pages(ram_addr_t base_pfn, unsigned long > *bitmap, > > + unsigned long len, int page_shift, > > +bool deflate) { #if defined(__linux__) > > Why do you need this #if here? > Ooh, it is wrong to add the '#if' here, will remove. Thanks a lot! Liang
> > On 13.06.2016 12:16, Liang Li wrote: > > > The implementation of the current virtio-balloon is not very > > > efficient, Bellow is test result of time spends on inflating the > > > balloon to 3GB of a 4GB idle guest: > > > > > > a. allocating pages (6.5%, 103ms) > > > b. sending PFNs to host (68.3%, 787ms) c. address translation (6.1%, > > > 96ms) d. madvise (19%, 300ms) > > > > > > It takes about 1577ms for the whole inflating process to complete. > > > The test shows that the bottle neck is the stage b and stage d. > > > > > > If using a bitmap to send the page info instead of the PFNs, we can > > > reduce the overhead spends on stage b quite a lot. Furthermore, it's > > > possible to do the address translation and do the madvise with a > > > bulk of pages, instead of the current page per page way, so the > > > overhead of stage c and stage d can also be reduced a lot. > > > > > > This patch is the QEMU side implementation which is intended to > > > speed up the inflating & deflating process by adding a new feature > > > to the virtio-balloon device. And now, inflating the balloon to 3GB > > > of a 4GB idle guest only takes 210ms, it's about 8 times as fast as before. > > > > > > TODO: optimize stage a by allocating/freeing a chunk of pages > > > instead of a single page at a time. > > > > > > Signed-off-by: Liang Li <liang.z.li@intel.com> > > > --- > > > hw/virtio/virtio-balloon.c | 159 ++++++++++++++++++++---- > > > include/standard-headers/linux/virtio_balloon.h | 1 + > > > 2 files changed, 139 insertions(+), 21 deletions(-) > > > > > > diff --git a/hw/virtio/virtio-balloon.c b/hw/virtio/virtio-balloon.c > > > index 8c15e09..8cf74c2 100644 > > > --- a/hw/virtio/virtio-balloon.c > > > +++ b/hw/virtio/virtio-balloon.c > > > @@ -47,6 +47,76 @@ static void balloon_page(void *addr, int deflate) > > > #endif } > > > > > > +static void do_balloon_bulk_pages(ram_addr_t base_pfn, int > page_shift, > > > + unsigned long len, bool deflate) { > > > + ram_addr_t size, processed, chunk, base; > > > + void *addr; > > > + MemoryRegionSection section = {.mr = NULL}; > > > + > > > + size = (len << page_shift); > > > + base = (base_pfn << page_shift); > > > + > > > + for (processed = 0; processed < size; processed += chunk) { > > > + chunk = size - processed; > > > + while (chunk >= TARGET_PAGE_SIZE) { > > > + section = memory_region_find(get_system_memory(), > > > + base + processed, chunk); > > > + if (!section.mr) { > > > + chunk = QEMU_ALIGN_DOWN(chunk / 2, TARGET_PAGE_SIZE); > > > + } else { > > > + break; > > > + } > > > + } > > > + > > > + if (section.mr && > > > + (int128_nz(section.size) && memory_region_is_ram(section.mr))) > { > > > + addr = section.offset_within_region + > > > + memory_region_get_ram_ptr(section.mr); > > > + qemu_madvise(addr, chunk, > > > + deflate ? QEMU_MADV_WILLNEED : > > QEMU_MADV_DONTNEED); > > > + } else { > > > + fprintf(stderr, "can't find the chunk, skip\n"); > > > > Please try to avoid new fprintf(stderr, ...) in the QEMU sources. > > Use error_report(...) or in this case maybe rather > > qemu_log_mask(LOG_GUEST_ERROR, ...) instead, and try to use a more > > reasonable error message (e.g. that it is clear that the error > > happened in the balloon code). > > > > Indeed, the error message is no good, will change in next version. > > > > + chunk = TARGET_PAGE_SIZE; > > > + } > > > + } > > > +} > > > + > > > +static void balloon_bulk_pages(ram_addr_t base_pfn, unsigned long > > *bitmap, > > > + unsigned long len, int page_shift, > > > +bool deflate) { #if defined(__linux__) > > > > Why do you need this #if here? > > > > Ooh, it is wrong to add the '#if' here, will remove. No, it is needed, just follow the code in balloon_page(). only Linux support the madvise(). Liang
On 14.06.2016 16:41, Li, Liang Z wrote: >>> On 13.06.2016 12:16, Liang Li wrote: >>>> The implementation of the current virtio-balloon is not very >>>> efficient, Bellow is test result of time spends on inflating the >>>> balloon to 3GB of a 4GB idle guest: >>>> >>>> a. allocating pages (6.5%, 103ms) >>>> b. sending PFNs to host (68.3%, 787ms) c. address translation (6.1%, >>>> 96ms) d. madvise (19%, 300ms) >>>> >>>> It takes about 1577ms for the whole inflating process to complete. >>>> The test shows that the bottle neck is the stage b and stage d. >>>> >>>> If using a bitmap to send the page info instead of the PFNs, we can >>>> reduce the overhead spends on stage b quite a lot. Furthermore, it's >>>> possible to do the address translation and do the madvise with a >>>> bulk of pages, instead of the current page per page way, so the >>>> overhead of stage c and stage d can also be reduced a lot. >>>> >>>> This patch is the QEMU side implementation which is intended to >>>> speed up the inflating & deflating process by adding a new feature >>>> to the virtio-balloon device. And now, inflating the balloon to 3GB >>>> of a 4GB idle guest only takes 210ms, it's about 8 times as fast as before. [...] >>>> + chunk = TARGET_PAGE_SIZE; >>>> + } >>>> + } >>>> +} >>>> + >>>> +static void balloon_bulk_pages(ram_addr_t base_pfn, unsigned long >>> *bitmap, >>>> + unsigned long len, int page_shift, >>>> +bool deflate) { #if defined(__linux__) >>> >>> Why do you need this #if here? >>> >> Ooh, it is wrong to add the '#if' here, will remove. >> > No, it is needed, just follow the code in balloon_page(). > only Linux support the madvise(). I think it is not needed anymore today and the #if in balloon_page could be removed, too: As far as I can see, the #if there is from the early days, when there was no wrapper around madvise() yet. But nowadays, we've got the qemu_madvise() wrapper which takes care of either using madvise(), posix_madvise() or doing nothing, so the virtio-balloon code should be able to work without the #if now. Thomas -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
> >>>> + chunk = TARGET_PAGE_SIZE; > >>>> + } > >>>> + } > >>>> +} > >>>> + > >>>> +static void balloon_bulk_pages(ram_addr_t base_pfn, unsigned long > >>> *bitmap, > >>>> + unsigned long len, int page_shift, > >>>> +bool deflate) { #if defined(__linux__) > >>> > >>> Why do you need this #if here? > >>> > >> Ooh, it is wrong to add the '#if' here, will remove. > >> > > No, it is needed, just follow the code in balloon_page(). > > only Linux support the madvise(). > > I think it is not needed anymore today and the #if in balloon_page could be > removed, too: As far as I can see, the #if there is from the early days, when > there was no wrapper around madvise() yet. But nowadays, we've got the > qemu_madvise() wrapper which takes care of either using madvise(), > posix_madvise() or doing nothing, so the virtio-balloon code should be able > to work without the #if now. > > Thomas You are right! I will remove both of them. Thanks! Liang
On Mon, Jun 13, 2016 at 06:16:43PM +0800, Liang Li wrote: > The implementation of the current virtio-balloon is not very efficient, > Bellow is test result of time spends on inflating the balloon to 3GB of > a 4GB idle guest: > > a. allocating pages (6.5%, 103ms) > b. sending PFNs to host (68.3%, 787ms) > c. address translation (6.1%, 96ms) > d. madvise (19%, 300ms) > > It takes about 1577ms for the whole inflating process to complete. The > test shows that the bottle neck is the stage b and stage d. > > If using a bitmap to send the page info instead of the PFNs, we can > reduce the overhead spends on stage b quite a lot. Furthermore, it's > possible to do the address translation and do the madvise with a bulk > of pages, instead of the current page per page way, so the overhead of > stage c and stage d can also be reduced a lot. > > This patch is the QEMU side implementation which is intended to speed > up the inflating & deflating process by adding a new feature to the > virtio-balloon device. And now, inflating the balloon to 3GB of a 4GB > idle guest only takes 210ms, it's about 8 times as fast as before. > > TODO: optimize stage a by allocating/freeing a chunk of pages instead > of a single page at a time. > > Signed-off-by: Liang Li <liang.z.li@intel.com> > --- > hw/virtio/virtio-balloon.c | 159 ++++++++++++++++++++---- > include/standard-headers/linux/virtio_balloon.h | 1 + > 2 files changed, 139 insertions(+), 21 deletions(-) > > diff --git a/hw/virtio/virtio-balloon.c b/hw/virtio/virtio-balloon.c > index 8c15e09..8cf74c2 100644 > --- a/hw/virtio/virtio-balloon.c > +++ b/hw/virtio/virtio-balloon.c > @@ -47,6 +47,76 @@ static void balloon_page(void *addr, int deflate) > #endif > } > > +static void do_balloon_bulk_pages(ram_addr_t base_pfn, int page_shift, > + unsigned long len, bool deflate) > +{ > + ram_addr_t size, processed, chunk, base; > + void *addr; > + MemoryRegionSection section = {.mr = NULL}; > + > + size = (len << page_shift); > + base = (base_pfn << page_shift); > + > + for (processed = 0; processed < size; processed += chunk) { > + chunk = size - processed; > + while (chunk >= TARGET_PAGE_SIZE) { > + section = memory_region_find(get_system_memory(), > + base + processed, chunk); > + if (!section.mr) { > + chunk = QEMU_ALIGN_DOWN(chunk / 2, TARGET_PAGE_SIZE); > + } else { > + break; > + } > + } > + > + if (section.mr && > + (int128_nz(section.size) && memory_region_is_ram(section.mr))) { > + addr = section.offset_within_region + > + memory_region_get_ram_ptr(section.mr); > + qemu_madvise(addr, chunk, > + deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED); > + } else { > + fprintf(stderr, "can't find the chunk, skip\n"); > + chunk = TARGET_PAGE_SIZE; > + } > + } > +} > + > +static void balloon_bulk_pages(ram_addr_t base_pfn, unsigned long *bitmap, > + unsigned long len, int page_shift, bool deflate) > +{ > +#if defined(__linux__) > + unsigned long end = len * 8; > + unsigned long current = 0; > + > + if (!qemu_balloon_is_inhibited() && (!kvm_enabled() || > + kvm_has_sync_mmu())) { > + while (current < end) { > + unsigned long one = find_next_bit(bitmap, end, current); > + > + if (one < end) { > + unsigned long zero = find_next_zero_bit(bitmap, end, one + 1); > + unsigned long page_length; > + > + if (zero >= end) { > + page_length = end - one; > + } else { > + page_length = zero - one; > + } > + > + if (page_length) { > + do_balloon_bulk_pages(base_pfn + one, page_shift, > + page_length, deflate); > + } > + current = one + page_length; > + } else { > + current = one; > + } > + } > + } > +#endif > +} > + > static const char *balloon_stat_names[] = { > [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in", > [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out", > @@ -78,6 +148,12 @@ static bool balloon_stats_supported(const VirtIOBalloon *s) > return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ); > } > > +static bool balloon_page_bitmap_supported(const VirtIOBalloon *s) > +{ > + VirtIODevice *vdev = VIRTIO_DEVICE(s); > + return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_BITMAP); > +} > + > static bool balloon_stats_enabled(const VirtIOBalloon *s) > { > return s->stats_poll_interval > 0; > @@ -224,27 +300,66 @@ static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq) > return; > } > > - while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) { > - ram_addr_t pa; > - ram_addr_t addr; > - int p = virtio_ldl_p(vdev, &pfn); > - > - pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT; > - offset += 4; > - > - /* FIXME: remove get_system_memory(), but how? */ > - section = memory_region_find(get_system_memory(), pa, 1); > - if (!int128_nz(section.size) || !memory_region_is_ram(section.mr)) > - continue; > - > - trace_virtio_balloon_handle_output(memory_region_name(section.mr), > - pa); > - /* Using memory_region_get_ram_ptr is bending the rules a bit, but > - should be OK because we only want a single page. */ > - addr = section.offset_within_region; > - balloon_page(memory_region_get_ram_ptr(section.mr) + addr, > - !!(vq == s->dvq)); > - memory_region_unref(section.mr); > + if (balloon_page_bitmap_supported(s)) { > + uint64_t base_pfn, tmp64, bmap_len; > + uint32_t tmp32, page_shift, id; > + unsigned long *bitmap; > + > + iov_to_buf(elem->out_sg, elem->out_num, offset, > + &tmp32, sizeof(uint32_t)); > + id = virtio_ldl_p(vdev, &tmp32); > + offset += sizeof(uint32_t); > + /* to suppress build warning */ > + id = id; > + > + iov_to_buf(elem->out_sg, elem->out_num, offset, > + &tmp32, sizeof(uint32_t)); > + page_shift = virtio_ldl_p(vdev, &tmp32); > + offset += sizeof(uint32_t); > + > + iov_to_buf(elem->out_sg, elem->out_num, offset, > + &tmp64, sizeof(uint64_t)); > + base_pfn = virtio_ldq_p(vdev, &tmp64); > + offset += sizeof(uint64_t); > + > + iov_to_buf(elem->out_sg, elem->out_num, offset, > + &tmp64, sizeof(uint64_t)); > + bmap_len = virtio_ldq_p(vdev, &tmp64); > + offset += sizeof(uint64_t); > + > + bitmap = bitmap_new(bmap_len * BITS_PER_BYTE); > + iov_to_buf(elem->out_sg, elem->out_num, offset, > + bitmap, bmap_len); > + offset += bmap_len; > + > + balloon_bulk_pages(base_pfn, bitmap, bmap_len, > + page_shift, !!(vq == s->dvq)); > + g_free(bitmap); > + } else { > + while (iov_to_buf(elem->out_sg, elem->out_num, offset, > + &pfn, 4) == 4) { > + ram_addr_t pa; > + ram_addr_t addr; > + int p = virtio_ldl_p(vdev, &pfn); > + > + pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT; > + offset += 4; > + > + /* FIXME: remove get_system_memory(), but how? */ > + section = memory_region_find(get_system_memory(), pa, 1); > + if (!int128_nz(section.size) || > + !memory_region_is_ram(section.mr)) > + continue; > + > + trace_virtio_balloon_handle_output(memory_region_name( > + section.mr), pa); > + /* Using memory_region_get_ram_ptr is bending the rules a bit, > + * but should be OK because we only want a single page. */ > + addr = section.offset_within_region; > + balloon_page(memory_region_get_ram_ptr(section.mr) + addr, > + !!(vq == s->dvq)); > + memory_region_unref(section.mr); > + } > } > > virtqueue_push(vq, elem, offset); > @@ -374,6 +489,7 @@ static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f, > VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); > f |= dev->host_features; > virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ); > + virtio_add_feature(&f, VIRTIO_BALLOON_F_PAGE_BITMAP); > return f; > } > Pls add features to virtio_balloon_properties. You also need to handle compatibility by disabling for old machine types. > @@ -388,6 +504,7 @@ static void virtio_balloon_to_target(void *opaque, ram_addr_t target) > { > VirtIOBalloon *dev = VIRTIO_BALLOON(opaque); > VirtIODevice *vdev = VIRTIO_DEVICE(dev); > + > ram_addr_t vm_ram_size = get_current_ram_size(); > > if (target > vm_ram_size) { > diff --git a/include/standard-headers/linux/virtio_balloon.h b/include/standard-headers/linux/virtio_balloon.h > index 9d06ccd..7c9686c 100644 > --- a/include/standard-headers/linux/virtio_balloon.h > +++ b/include/standard-headers/linux/virtio_balloon.h > @@ -34,6 +34,7 @@ > #define VIRTIO_BALLOON_F_MUST_TELL_HOST 0 /* Tell before reclaiming pages */ > #define VIRTIO_BALLOON_F_STATS_VQ 1 /* Memory Stats virtqueue */ > #define VIRTIO_BALLOON_F_DEFLATE_ON_OOM 2 /* Deflate balloon on OOM */ > +#define VIRTIO_BALLOON_F_PAGE_BITMAP 3 /* Use page bitmap to send page info */ > > /* Size of a PFN in the balloon interface. */ > #define VIRTIO_BALLOON_PFN_SHIFT 12 We want to keep this in sync with Linux. Let's get a minimal patch to extend this header merged in linux, then update this one. > -- > 1.9.1 -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
> > > > virtqueue_push(vq, elem, offset); @@ -374,6 +489,7 @@ static > > uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f, > > VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); > > f |= dev->host_features; > > virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ); > > + virtio_add_feature(&f, VIRTIO_BALLOON_F_PAGE_BITMAP); > > return f; > > } > > > > Pls add features to virtio_balloon_properties. > You also need to handle compatibility by disabling for old machine types. > I forgot that, will add in next version. > > --- a/include/standard-headers/linux/virtio_balloon.h > > +++ b/include/standard-headers/linux/virtio_balloon.h > > @@ -34,6 +34,7 @@ > > #define VIRTIO_BALLOON_F_MUST_TELL_HOST 0 /* Tell before > reclaiming pages */ > > #define VIRTIO_BALLOON_F_STATS_VQ 1 /* Memory Stats virtqueue > */ > > #define VIRTIO_BALLOON_F_DEFLATE_ON_OOM 2 /* Deflate balloon > on OOM */ > > +#define VIRTIO_BALLOON_F_PAGE_BITMAP 3 /* Use page bitmap to > send > > +page info */ > > > > /* Size of a PFN in the balloon interface. */ #define > > VIRTIO_BALLOON_PFN_SHIFT 12 > > We want to keep this in sync with Linux. > Let's get a minimal patch to extend this header merged in linux, then update > this one. OK. Can this be independent of the virtio-balloon SPEC? As I understand it, it will not get merged before the SPEC is set? Thanks! Liang -- To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
diff --git a/hw/virtio/virtio-balloon.c b/hw/virtio/virtio-balloon.c index 8c15e09..8cf74c2 100644 --- a/hw/virtio/virtio-balloon.c +++ b/hw/virtio/virtio-balloon.c @@ -47,6 +47,76 @@ static void balloon_page(void *addr, int deflate) #endif } +static void do_balloon_bulk_pages(ram_addr_t base_pfn, int page_shift, + unsigned long len, bool deflate) +{ + ram_addr_t size, processed, chunk, base; + void *addr; + MemoryRegionSection section = {.mr = NULL}; + + size = (len << page_shift); + base = (base_pfn << page_shift); + + for (processed = 0; processed < size; processed += chunk) { + chunk = size - processed; + while (chunk >= TARGET_PAGE_SIZE) { + section = memory_region_find(get_system_memory(), + base + processed, chunk); + if (!section.mr) { + chunk = QEMU_ALIGN_DOWN(chunk / 2, TARGET_PAGE_SIZE); + } else { + break; + } + } + + if (section.mr && + (int128_nz(section.size) && memory_region_is_ram(section.mr))) { + addr = section.offset_within_region + + memory_region_get_ram_ptr(section.mr); + qemu_madvise(addr, chunk, + deflate ? QEMU_MADV_WILLNEED : QEMU_MADV_DONTNEED); + } else { + fprintf(stderr, "can't find the chunk, skip\n"); + chunk = TARGET_PAGE_SIZE; + } + } +} + +static void balloon_bulk_pages(ram_addr_t base_pfn, unsigned long *bitmap, + unsigned long len, int page_shift, bool deflate) +{ +#if defined(__linux__) + unsigned long end = len * 8; + unsigned long current = 0; + + if (!qemu_balloon_is_inhibited() && (!kvm_enabled() || + kvm_has_sync_mmu())) { + while (current < end) { + unsigned long one = find_next_bit(bitmap, end, current); + + if (one < end) { + unsigned long zero = find_next_zero_bit(bitmap, end, one + 1); + unsigned long page_length; + + if (zero >= end) { + page_length = end - one; + } else { + page_length = zero - one; + } + + if (page_length) { + do_balloon_bulk_pages(base_pfn + one, page_shift, + page_length, deflate); + } + current = one + page_length; + } else { + current = one; + } + } + } +#endif +} + static const char *balloon_stat_names[] = { [VIRTIO_BALLOON_S_SWAP_IN] = "stat-swap-in", [VIRTIO_BALLOON_S_SWAP_OUT] = "stat-swap-out", @@ -78,6 +148,12 @@ static bool balloon_stats_supported(const VirtIOBalloon *s) return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_STATS_VQ); } +static bool balloon_page_bitmap_supported(const VirtIOBalloon *s) +{ + VirtIODevice *vdev = VIRTIO_DEVICE(s); + return virtio_vdev_has_feature(vdev, VIRTIO_BALLOON_F_PAGE_BITMAP); +} + static bool balloon_stats_enabled(const VirtIOBalloon *s) { return s->stats_poll_interval > 0; @@ -224,27 +300,66 @@ static void virtio_balloon_handle_output(VirtIODevice *vdev, VirtQueue *vq) return; } - while (iov_to_buf(elem->out_sg, elem->out_num, offset, &pfn, 4) == 4) { - ram_addr_t pa; - ram_addr_t addr; - int p = virtio_ldl_p(vdev, &pfn); - - pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT; - offset += 4; - - /* FIXME: remove get_system_memory(), but how? */ - section = memory_region_find(get_system_memory(), pa, 1); - if (!int128_nz(section.size) || !memory_region_is_ram(section.mr)) - continue; - - trace_virtio_balloon_handle_output(memory_region_name(section.mr), - pa); - /* Using memory_region_get_ram_ptr is bending the rules a bit, but - should be OK because we only want a single page. */ - addr = section.offset_within_region; - balloon_page(memory_region_get_ram_ptr(section.mr) + addr, - !!(vq == s->dvq)); - memory_region_unref(section.mr); + if (balloon_page_bitmap_supported(s)) { + uint64_t base_pfn, tmp64, bmap_len; + uint32_t tmp32, page_shift, id; + unsigned long *bitmap; + + iov_to_buf(elem->out_sg, elem->out_num, offset, + &tmp32, sizeof(uint32_t)); + id = virtio_ldl_p(vdev, &tmp32); + offset += sizeof(uint32_t); + /* to suppress build warning */ + id = id; + + iov_to_buf(elem->out_sg, elem->out_num, offset, + &tmp32, sizeof(uint32_t)); + page_shift = virtio_ldl_p(vdev, &tmp32); + offset += sizeof(uint32_t); + + iov_to_buf(elem->out_sg, elem->out_num, offset, + &tmp64, sizeof(uint64_t)); + base_pfn = virtio_ldq_p(vdev, &tmp64); + offset += sizeof(uint64_t); + + iov_to_buf(elem->out_sg, elem->out_num, offset, + &tmp64, sizeof(uint64_t)); + bmap_len = virtio_ldq_p(vdev, &tmp64); + offset += sizeof(uint64_t); + + bitmap = bitmap_new(bmap_len * BITS_PER_BYTE); + iov_to_buf(elem->out_sg, elem->out_num, offset, + bitmap, bmap_len); + offset += bmap_len; + + balloon_bulk_pages(base_pfn, bitmap, bmap_len, + page_shift, !!(vq == s->dvq)); + g_free(bitmap); + } else { + while (iov_to_buf(elem->out_sg, elem->out_num, offset, + &pfn, 4) == 4) { + ram_addr_t pa; + ram_addr_t addr; + int p = virtio_ldl_p(vdev, &pfn); + + pa = (ram_addr_t) p << VIRTIO_BALLOON_PFN_SHIFT; + offset += 4; + + /* FIXME: remove get_system_memory(), but how? */ + section = memory_region_find(get_system_memory(), pa, 1); + if (!int128_nz(section.size) || + !memory_region_is_ram(section.mr)) + continue; + + trace_virtio_balloon_handle_output(memory_region_name( + section.mr), pa); + /* Using memory_region_get_ram_ptr is bending the rules a bit, + * but should be OK because we only want a single page. */ + addr = section.offset_within_region; + balloon_page(memory_region_get_ram_ptr(section.mr) + addr, + !!(vq == s->dvq)); + memory_region_unref(section.mr); + } } virtqueue_push(vq, elem, offset); @@ -374,6 +489,7 @@ static uint64_t virtio_balloon_get_features(VirtIODevice *vdev, uint64_t f, VirtIOBalloon *dev = VIRTIO_BALLOON(vdev); f |= dev->host_features; virtio_add_feature(&f, VIRTIO_BALLOON_F_STATS_VQ); + virtio_add_feature(&f, VIRTIO_BALLOON_F_PAGE_BITMAP); return f; } @@ -388,6 +504,7 @@ static void virtio_balloon_to_target(void *opaque, ram_addr_t target) { VirtIOBalloon *dev = VIRTIO_BALLOON(opaque); VirtIODevice *vdev = VIRTIO_DEVICE(dev); + ram_addr_t vm_ram_size = get_current_ram_size(); if (target > vm_ram_size) { diff --git a/include/standard-headers/linux/virtio_balloon.h b/include/standard-headers/linux/virtio_balloon.h index 9d06ccd..7c9686c 100644 --- a/include/standard-headers/linux/virtio_balloon.h +++ b/include/standard-headers/linux/virtio_balloon.h @@ -34,6 +34,7 @@ #define VIRTIO_BALLOON_F_MUST_TELL_HOST 0 /* Tell before reclaiming pages */ #define VIRTIO_BALLOON_F_STATS_VQ 1 /* Memory Stats virtqueue */ #define VIRTIO_BALLOON_F_DEFLATE_ON_OOM 2 /* Deflate balloon on OOM */ +#define VIRTIO_BALLOON_F_PAGE_BITMAP 3 /* Use page bitmap to send page info */ /* Size of a PFN in the balloon interface. */ #define VIRTIO_BALLOON_PFN_SHIFT 12
The implementation of the current virtio-balloon is not very efficient, Bellow is test result of time spends on inflating the balloon to 3GB of a 4GB idle guest: a. allocating pages (6.5%, 103ms) b. sending PFNs to host (68.3%, 787ms) c. address translation (6.1%, 96ms) d. madvise (19%, 300ms) It takes about 1577ms for the whole inflating process to complete. The test shows that the bottle neck is the stage b and stage d. If using a bitmap to send the page info instead of the PFNs, we can reduce the overhead spends on stage b quite a lot. Furthermore, it's possible to do the address translation and do the madvise with a bulk of pages, instead of the current page per page way, so the overhead of stage c and stage d can also be reduced a lot. This patch is the QEMU side implementation which is intended to speed up the inflating & deflating process by adding a new feature to the virtio-balloon device. And now, inflating the balloon to 3GB of a 4GB idle guest only takes 210ms, it's about 8 times as fast as before. TODO: optimize stage a by allocating/freeing a chunk of pages instead of a single page at a time. Signed-off-by: Liang Li <liang.z.li@intel.com> --- hw/virtio/virtio-balloon.c | 159 ++++++++++++++++++++---- include/standard-headers/linux/virtio_balloon.h | 1 + 2 files changed, 139 insertions(+), 21 deletions(-)