[Intel-gfx] [PATCH i-g-t] lib: add a function to lock memory into RAM

[Daniel Vetter]

On Mon, Dec 08, 2014 at 05:22:12PM +0000, Thomas Wood wrote:
> Add a function to lock memory into RAM and use it in the
> gem_tiled_swapping test to reduce the amount of allocated memory
> required to force swapping. This also reduces the amount of time
> required for the test to complete, since the data set is smaller.
> 
> Signed-off-by: Thomas Wood <thomas.wood at intel.com>

Some nitpicks below. Also please mention in the commit message how much
this speeds up the testcase.

btw as long as you still see lots of swapping going on with vmstat 1 then
the test still works.

Cheers, Daniel

> ---
>  lib/igt_aux.c              | 48 ++++++++++++++++++++++++++++++++++++++++++++++
>  lib/igt_aux.h              |  3 +++
>  tests/gem_tiled_swapping.c |  7 ++++++-
>  3 files changed, 57 insertions(+), 1 deletion(-)
> 
> diff --git a/lib/igt_aux.c b/lib/igt_aux.c
> index 3051d84..a98aa8f 100644
> --- a/lib/igt_aux.c
> +++ b/lib/igt_aux.c
> @@ -601,3 +601,51 @@ struct type_name connector_type_names[] = {
>  };
>  
>  type_name_fn(connector_type)
> +
> +
> +/**
> + * igt_lock_mem: @size: the amount of memory to lock into RAM, in MB
> + *
> + * Allocate @size MB of memory and lock it into RAM. This releases any
> + * previously locked memory.
> + *
> + * Use #igt_unlock_mem to release the currently locked memory.
> + */
> +static char *locked_mem;
> +static size_t locked_size;
> +
> +void igt_lock_mem(size_t size)
> +{
> +	long pagesize = sysconf(_SC_PAGESIZE);
> +	size_t i;
> +
> +	if (locked_mem)

Imo this should be an igt_warn_on since if we leak mlocked memory until
the next test that needs it there's a good chance we've locked a bit too
much.

> +		igt_unlock_mem();
> +
> +	locked_size = size * 1024 * 1024;
> +
> +	locked_mem = malloc(locked_size);
> +	igt_assert(locked_mem);

Imo this should be an igt_require: On 32bit platforms with lots of memory
there might not be enough virtual address space for mlocking.

> +
> +	/* write into each page to ensure it is allocated */
> +	for (i = 0; i < locked_size; i += pagesize)
> +		locked_mem[i] = i;
> +
> +	mlock(locked_mem, locked_size);

This should have an igt_assert to make sure we still do a useful test.

> +}
> +
> +/**
> + * igt_unlock_mem:
> + *
> + * Release and free the RAM used by #igt_lock_mem.
> + */
> +void igt_unlock_mem(void)
> +{
> +	if (!locked_mem)
> +		return;
> +
> +	munlock(locked_mem, locked_size);
> +
> +	free(locked_mem);
> +	locked_mem = NULL;
> +}
> diff --git a/lib/igt_aux.h b/lib/igt_aux.h
> index 6c83c53..b61555b 100644
> --- a/lib/igt_aux.h
> +++ b/lib/igt_aux.h
> @@ -90,4 +90,7 @@ void intel_require_memory(uint32_t count, uint32_t size, unsigned mode);
>  #define min(a, b) ((a) < (b) ? (a) : (b))
>  #define max(a, b) ((a) > (b) ? (a) : (b))
>  
> +void igt_lock_mem(size_t size);
> +void igt_unlock_mem(void);
> +
>  #endif /* IGT_AUX_H */
> diff --git a/tests/gem_tiled_swapping.c b/tests/gem_tiled_swapping.c
> index 3fac52f..0684ab2 100644
> --- a/tests/gem_tiled_swapping.c
> +++ b/tests/gem_tiled_swapping.c
> @@ -70,6 +70,7 @@ IGT_TEST_DESCRIPTION("Exercise swizzle code for swapping.");
>  static uint32_t current_tiling_mode;
>  
>  #define PAGE_SIZE 4096
> +#define AVAIL_RAM 512
>  
>  static uint32_t
>  create_bo_and_fill(int fd)
> @@ -156,8 +157,12 @@ igt_main
>  		intel_purge_vm_caches();
>  
>  		fd = drm_open_any();
> +
> +		/* lock RAM, leaving only 512MB available */
> +		igt_lock_mem(intel_get_total_ram_mb() - AVAIL_RAM);

We have machines with less than 512MB of memory (vpg has some underpowered
baytrails like that). So this needs a min(..., 0). And the calculation
below needs to substract that from the total ram.

> +
>  		/* need slightly more than available memory */
> -		count = intel_get_total_ram_mb() + intel_get_total_swap_mb() / 4;
> +		count = AVAIL_RAM * 1.25;
>  		bo_handles = calloc(count, sizeof(uint32_t));
>  		igt_assert(bo_handles);
>  
> -- 
> 2.1.0
> 
> _______________________________________________
> Intel-gfx mailing list
> Intel-gfx at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/intel-gfx

-- 
Daniel Vetter
Software Engineer, Intel Corporation
+41 (0) 79 365 57 48 - http://blog.ffwll.ch