qemu-monitor screendump very slow

Bug #1618122 reported by vans163
6
This bug affects 1 person
Affects Status Importance Assigned to Milestone
QEMU
Expired
Undecided
Unassigned

Bug Description

qemu-monitor screendump often using 10-20% cpu usage of one core to take a small capture.

Most of the CPU usage seems to come from libpixman. There were many reports of libpixman becoming 8 times slower in newer releases.

https://github.com/qemu/qemu/blob/0c56c6ab68902281094c7aac6305e2321c34c187/ui/console.c#L285

Simple Valgrind Ir report.

--------------------------------------------------------------------------------
           Ir
--------------------------------------------------------------------------------
9,994,313,959 PROGRAM TOTALS

--------------------------------------------------------------------------------
           Ir file:function
--------------------------------------------------------------------------------
4,883,603,360 /usr/src/debug/pixman-0.34.0/pixman/pixman-access.c:store_scanline_b8g8r8 [/usr/lib64/libpixman-1.so.0.34.0]
  282,312,800 /usr/src/debug/pixman-0.34.0/pixman/pixman-implementation.c:_pixman_implementation_iter_init [/usr/lib64/libpixman-1.so.0.34.0]
  267,394,160 /usr/src/debug/pixman-0.34.0/pixman/pixman-sse2.c:sse2_fetch_x8r8g8b8 [/usr/lib64/libpixman-1.so.0.34.0]
  256,761,600 /usr/src/debug/pixman-0.34.0/pixman/pixman-private.h:store_scanline_b8g8r8
  254,676,199 ???:0x0000000000011f40 [/usr/lib64/libjemalloc.so.2]
  199,990,526 ???:0x0000000000011a10 [/usr/lib64/libjemalloc.so.2]

Calls to there start from.
qemu_pixman_linebuf_fill
pixman_image_composite
????
pixman_image_composite32
general_composite_rect

I tried taking a look on how to fix this, but it seems pixman is deeply enrooted inside the monitor. I wanted to try to simply take whats on the display and memcpy it into .ppm format manually creating the file header, but the image is all broken with colors being off or its smeared.

For example this is DisplaySurface:

struct DisplaySurface {
    pixman_format_code_t format;
    pixman_image_t *image;
    uint8_t flags;
#ifdef CONFIG_OPENGL
    GLenum glformat;
    GLenum gltype;
    GLuint texture;
#endif
};

It appears graphic_hw_update(con) renders the vram to a pixman_image. Then ppm_save(filename, surface, errp) goes over it yet again, mutates the bytes in an expensive way to another pixman_image, and saves that row by row to a ppm.

Using sdl graphics (the default) the drawing starts in https://github.com/qemu/qemu/blob/7fa124b273acd22a808e742ead78c065ccd9b4c4/hw/display/vga.c#L1466

vans163 (vans-163)
description: updated
description: updated
Revision history for this message
vans163 (vans-163) wrote :

Most of the CPU is coming from ppm_save(filename, surface, errp);

graphic_hw_update(con) takes an insignificant amount.

vans163 (vans-163)
description: updated
vans163 (vans-163)
description: updated
vans163 (vans-163)
description: updated
Revision history for this message
Thomas Huth (th-huth) wrote :

The QEMU project is currently considering to move its bug tracking to
another system. For this we need to know which bugs are still valid
and which could be closed already. Thus we are setting older bugs to
"Incomplete" now.

If you still think this bug report here is valid, then please switch
the state back to "New" within the next 60 days, otherwise this report
will be marked as "Expired". Or please mark it as "Fix Released" if
the problem has been solved with a newer version of QEMU already.

Thank you and sorry for the inconvenience.

Changed in qemu:
status: New → Incomplete
Revision history for this message
Launchpad Janitor (janitor) wrote :

[Expired for QEMU because there has been no activity for 60 days.]

Changed in qemu:
status: Incomplete → Expired
To post a comment you must log in.
This report contains Public information  
Everyone can see this information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.