Re: [Qemu-devel] [PATCH 4/6] memory: find dirty range

[Avi Kivity <avi@redhat.com>]

On 12/10/2011 06:45 PM, Blue Swirl wrote:
> Instead of each target knowing or guessing the guest page size,
> iterate through the dirty ranges.
>
> Signed-off-by: Blue Swirl <blauwirbel@gmail.com>
> ---
>  cpu-all.h |   30 ++++++++++++++++++++++++++++++
>  hw/tcx.c  |   54 ++++++++++++++++++++++--------------------------------
>  hw/vga.c  |   16 +++++++---------
>  memory.c  |   16 ++++++++++++++++
>  memory.h  |   24 ++++++++++++++++++++++++
>  5 files changed, 99 insertions(+), 41 deletions(-)
>
> diff --git a/cpu-all.h b/cpu-all.h
> index 0cb62ca..a5c6670 100644
> --- a/cpu-all.h
> +++ b/cpu-all.h
> @@ -574,6 +574,36 @@ int cpu_physical_memory_set_dirty_tracking(int enable);
>
>  int cpu_physical_memory_get_dirty_tracking(void);
>
> +static inline void cpu_physical_memory_range_find_dirty(ram_addr_t start,
> +                                                        ram_addr_t end,
> +                                                        ram_addr_t *pstart,
> +                                                        ram_addr_t *pend,
> +                                                        int flags)
> +{
> +    ram_addr_t idx;
> +
> +    start >>= TARGET_PAGE_BITS;
> +    end += TARGET_PAGE_SIZE - 1;
> +    end >>= TARGET_PAGE_BITS;
> +
> +    for (idx = start; idx < end; idx++) {
> +        if (ram_list.phys_dirty[idx] & flags) {
> +            *pstart = idx << TARGET_PAGE_BITS;
> +            for (; idx < end; idx++) {
> +                if (!(ram_list.phys_dirty[idx] & flags)) {
> +                    *pend = (idx << TARGET_PAGE_BITS) - 1;
> +                    return;
> +                }
> +            }
> +            *pend = (end << TARGET_PAGE_BITS) - 1;

This uses a convention of fully inclusive ranges, not semi inclusive
which is what we usually use.

> +            return;
> +        }
> +    }
> +    /* everything pristine */
> +    *pstart = (end << TARGET_PAGE_BITS) - 1;
> +    *pend = (end << TARGET_PAGE_BITS) - 1;
> +}
> +

I prefer this to be implemented using memory_region primitives, less
work for me to covert afterwards.

Also, no need to inline this.

>  int cpu_physical_sync_dirty_bitmap(target_phys_addr_t start_addr,
>                                     target_phys_addr_t end_addr);
>
> diff --git a/hw/tcx.c b/hw/tcx.c
> index fd45ce8..d031cbd 100644
> --- a/hw/tcx.c
> +++ b/hw/tcx.c
> @@ -212,7 +212,7 @@ static inline void reset_dirty(TCXState *ts,
> ram_addr_t page_min,
>  static void tcx_update_display(void *opaque)
>  {
>      TCXState *ts = opaque;
> -    ram_addr_t page, page_min, page_max;
> +    ram_addr_t page, page_min, page_max, p;
>      int y, y_start, dd, ds;
>      uint8_t *d, *s;
>      void (*f)(TCXState *s1, uint8_t *dst, const uint8_t *src, int width);
> @@ -244,37 +244,28 @@ static void tcx_update_display(void *opaque)
>          return;
>      }
>
> -    for(y = 0; y < ts->height; y += 4, page += TARGET_PAGE_SIZE) {
> -        if (memory_region_get_dirty(&ts->vram_mem, page, DIRTY_MEMORY_VGA)) {
> -            if (y_start < 0)
> -                y_start = y;
> -            if (page < page_min)
> -                page_min = page;
> -            if (page > page_max)
> -                page_max = page;
> -            f(ts, d, s, ts->width);
> -            d += dd;
> -            s += ds;
> -            f(ts, d, s, ts->width);
> -            d += dd;
> -            s += ds;
> -            f(ts, d, s, ts->width);
> -            d += dd;
> -            s += ds;
> -            f(ts, d, s, ts->width);
> -            d += dd;
> -            s += ds;
> -        } else {
> -            if (y_start >= 0) {
> -                /* flush to display */
> -                dpy_update(ts->ds, 0, y_start,
> -                           ts->width, y - y_start);
> -                y_start = -1;
> -            }
> -            d += dd * 4;
> -            s += ds * 4;
> +    assert(MAXX == 1024);
> +    y = 0;
> +    for (p = 0; p < MAXX * MAXY;)  {
> +        target_phys_addr_t dirty_start, dirty_end;
> +
> +        memory_region_find_dirty(&ts->vram_mem, p, MAXX * MAXY, &dirty_start,
> +                                 &dirty_end, DIRTY_MEMORY_VGA);
> +        if (dirty_start == MAXX * MAXY - 1) {
> +            break;
> +        }

Gives no way to indicate that just that one byte is dirty (possible if
MAXX * MAXY == 1 (mod TARGET_PAGE_SIZE))

> +        page = dirty_start;
> +        f(ts, d + (page >> 10) * dd, s + page, dirty_end - dirty_start);
> +        if (y_start < 0) {
> +            page_min = dirty_start;
> +            /* divide by MAXX */
> +            y_start = page_min >> 10;

page_min / MAXX?

>          }
> +        page_max = dirty_end;
> +        y = page_max >> 10;
> +        p = dirty_end + 1;
>      }
> +
>      if (y_start >= 0) {
>          /* flush to display */
>          dpy_update(ts->ds, 0, y_start,
> @@ -282,8 +273,7 @@ static void tcx_update_display(void *opaque)
>      }
>      /* reset modified pages */
>      if (page_max >= page_min) {
> -        memory_region_reset_dirty(&ts->vram_mem,
> -                                  page_min, page_max + TARGET_PAGE_SIZE,
> +        memory_region_reset_dirty(&ts->vram_mem, page_min, page_max,
>                                    DIRTY_MEMORY_VGA);

fully-inclusive/semi-inclusive mixup

>  /**
> + * memory_region_find_dirty: Find dirty ranges in a memory range for a
> + *                           specified client.
> + *
> + * Returns lowest contiguous dirty memory range within the specified
> memory range.
> + * Dirty logging must be enabled.
> + *
> + * @mr: the memory region being queried.
> + * @start: the start address (relative to the start of the region) of the
> + * region being searched.
> + * @end: the end address (relative to the start of the region) of the
> + * region being searched.
> + * *@pstart: the returned start address (relative to the start of the region)
> + * of the lowest contiguous dirty range found, or @end if not found.
> + * *@pend: the returned end address (relative to the start of the region)
> + * of the lowest contiguous dirty range found.
> + * @client: the user of the logging information; %DIRTY_MEMORY_MIGRATION or
> + *          %DIRTY_MEMORY_VGA.
> + */
> +void memory_region_find_dirty(MemoryRegion *mr, target_phys_addr_t start,
> +                              target_phys_addr_t end,
> +                              target_phys_addr_t *pstart,
> +                              target_phys_addr_t *pend,
> +                              unsigned client);
>

Might wrap in a MEMORY_REGION_FOR_EACH_DIRTY_RANGE() macro.

-- 
error compiling committee.c: too many arguments to function