On 15/09/2023 04:20, “William Roche wrote:
> From: William Roche <william.ro...@oracle.com>
> 
> A memory page poisoned from the hypervisor level is no longer readable.
> Thus, it is now treated as a zero-page for the ram saving migration phase.
> 
> The migration of a VM will crash Qemu when it tries to read the
> memory address space and stumbles on the poisoned page with a similar
> stack trace:
> 
> Program terminated with signal SIGBUS, Bus error.
> #0  _mm256_loadu_si256
> #1  buffer_zero_avx2
> #2  select_accel_fn
> #3  buffer_is_zero
> #4  save_zero_page_to_file
> #5  save_zero_page
> #6  ram_save_target_page_legacy
> #7  ram_save_host_page
> #8  ram_find_and_save_block
> #9  ram_save_iterate
> #10 qemu_savevm_state_iterate
> #11 migration_iteration_run
> #12 migration_thread
> #13 qemu_thread_start
> 
> Fix it by considering poisoned pages as if they were zero-pages for
> the migration copy. This fix also works with underlying large pages,
> taking into account the RAMBlock segment "page-size".
> 
> Standard migration and compressed transfers are handled by this code.
> RDMA transfer isn't touched.
> 


I'm okay with "RDMA isn't touched".
BTW, could you share your reproducing program/hacking to poison the page, so 
that
i am able to take a look the RDMA part later when i'm free.

Not sure it's suitable to acknowledge a not touched part. Anyway
Acked-by: Li Zhijian <lizhij...@fujitsu.com> # RDMA


> Signed-off-by: William Roche <william.ro...@oracle.com>
> ---
>   accel/kvm/kvm-all.c      | 14 ++++++++++++++
>   accel/stubs/kvm-stub.c   |  5 +++++
>   include/sysemu/kvm.h     | 10 ++++++++++
>   migration/ram-compress.c |  3 ++-
>   migration/ram.c          | 23 +++++++++++++++++++++--
>   migration/ram.h          |  2 ++
>   6 files changed, 54 insertions(+), 3 deletions(-)
> 
> diff --git a/accel/kvm/kvm-all.c b/accel/kvm/kvm-all.c
> index ff1578bb32..7fb13c8a56 100644
> --- a/accel/kvm/kvm-all.c
> +++ b/accel/kvm/kvm-all.c
> @@ -1152,6 +1152,20 @@ static void kvm_unpoison_all(void *param)
>       }
>   }
>   
> +bool kvm_hwpoisoned_page(RAMBlock *block, void *offset)
> +{
> +    HWPoisonPage *pg;
> +    ram_addr_t ram_addr = (ram_addr_t) offset;
> +
> +    QLIST_FOREACH(pg, &hwpoison_page_list, list) {
> +        if ((ram_addr >= pg->ram_addr) &&
> +            (ram_addr - pg->ram_addr < block->page_size)) {
> +            return true;
> +        }
> +    }
> +    return false;
> +}
> +
>   void kvm_hwpoison_page_add(ram_addr_t ram_addr)
>   {
>       HWPoisonPage *page;
> diff --git a/accel/stubs/kvm-stub.c b/accel/stubs/kvm-stub.c
> index 235dc661bc..c0a31611df 100644
> --- a/accel/stubs/kvm-stub.c
> +++ b/accel/stubs/kvm-stub.c
> @@ -133,3 +133,8 @@ uint32_t kvm_dirty_ring_size(void)
>   {
>       return 0;
>   }
> +
> +bool kvm_hwpoisoned_page(RAMBlock *block, void *ram_addr)
> +{
> +    return false;
> +}
> diff --git a/include/sysemu/kvm.h b/include/sysemu/kvm.h
> index ee9025f8e9..858688227a 100644
> --- a/include/sysemu/kvm.h
> +++ b/include/sysemu/kvm.h
> @@ -570,4 +570,14 @@ bool kvm_arch_cpu_check_are_resettable(void);
>   bool kvm_dirty_ring_enabled(void);
>   
>   uint32_t kvm_dirty_ring_size(void);
> +
> +/**
> + * kvm_hwpoisoned_page - indicate if the given page is poisoned
> + * @block: memory block of the given page
> + * @ram_addr: offset of the page
> + *
> + * Returns: true: page is poisoned
> + *          false: page not yet poisoned
> + */
> +bool kvm_hwpoisoned_page(RAMBlock *block, void *ram_addr);
>   #endif
> diff --git a/migration/ram-compress.c b/migration/ram-compress.c
> index 06254d8c69..1916ce709d 100644
> --- a/migration/ram-compress.c
> +++ b/migration/ram-compress.c
> @@ -34,6 +34,7 @@
>   #include "qemu/error-report.h"
>   #include "migration.h"
>   #include "options.h"
> +#include "ram.h"
>   #include "io/channel-null.h"
>   #include "exec/target_page.h"
>   #include "exec/ramblock.h"
> @@ -198,7 +199,7 @@ static CompressResult do_compress_ram_page(QEMUFile *f, 
> z_stream *stream,
>   
>       assert(qemu_file_buffer_empty(f));
>   
> -    if (buffer_is_zero(p, page_size)) {
> +    if (migration_buffer_is_zero(block, offset, page_size)) {
>           return RES_ZEROPAGE;
>       }
>   
> diff --git a/migration/ram.c b/migration/ram.c
> index 9040d66e61..fd337f7e65 100644
> --- a/migration/ram.c
> +++ b/migration/ram.c
> @@ -1129,6 +1129,26 @@ void ram_release_page(const char *rbname, uint64_t 
> offset)
>       ram_discard_range(rbname, offset, TARGET_PAGE_SIZE);
>   }
>   
> +/**
> + * migration_buffer_is_zero: indicate if the page at the given
> + * location is entirely filled with zero, or is a poisoned page.
> + *
> + * @block: block that contains the page
> + * @offset: offset inside the block for the page
> + * @len: size to consider
> + */
> +bool migration_buffer_is_zero(RAMBlock *block, ram_addr_t offset,
> +                                     size_t len)
> +{
> +    uint8_t *p = block->host + offset;
> +
> +    if (kvm_enabled() && kvm_hwpoisoned_page(block, (void *)offset)) {
> +        return true;
> +    }
> +
> +    return buffer_is_zero(p, len);
> +}
> +
>   /**
>    * save_zero_page_to_file: send the zero page to the file
>    *
> @@ -1142,10 +1162,9 @@ void ram_release_page(const char *rbname, uint64_t 
> offset)
>   static int save_zero_page_to_file(PageSearchStatus *pss, QEMUFile *file,
>                                     RAMBlock *block, ram_addr_t offset)
>   {
> -    uint8_t *p = block->host + offset;
>       int len = 0;
>   
> -    if (buffer_is_zero(p, TARGET_PAGE_SIZE)) {
> +    if (migration_buffer_is_zero(block, offset, TARGET_PAGE_SIZE)) {
>           len += save_page_header(pss, file, block, offset | 
> RAM_SAVE_FLAG_ZERO);
>           qemu_put_byte(file, 0);
>           len += 1;
> diff --git a/migration/ram.h b/migration/ram.h
> index 145c915ca7..805ea2a211 100644
> --- a/migration/ram.h
> +++ b/migration/ram.h
> @@ -65,6 +65,8 @@ void ram_handle_compressed(void *host, uint8_t ch, uint64_t 
> size);
>   void ram_transferred_add(uint64_t bytes);
>   void ram_release_page(const char *rbname, uint64_t offset);
>   
> +bool migration_buffer_is_zero(RAMBlock *block, ram_addr_t offset, size_t 
> len);
> +
>   int ramblock_recv_bitmap_test(RAMBlock *rb, void *host_addr);
>   bool ramblock_recv_bitmap_test_byte_offset(RAMBlock *rb, uint64_t 
> byte_offset);
>   void ramblock_recv_bitmap_set(RAMBlock *rb, void *host_addr);

Reply via email to