Re: [PATCH] hv_balloon: Add module parameter to configure balloon floor value
From: Vitaly Kuznetsov <vkuznets@redhat.com>
Date: 2023-10-17 14:42:09
Also in:
lkml
Angelina Vu [off-list ref] writes:
quoted hunk ↗ jump to hunk
Currently, the balloon floor value is automatically computed, but may be too small depending on app usage of memory. This patch adds a balloon_floor value as a module parameter that can be used to manually configure the balloon floor value. Signed-off-by: Angelina Vu <redacted> --- drivers/hv/hv_balloon.c | 7 +++++++ 1 file changed, 7 insertions(+)diff --git a/drivers/hv/hv_balloon.c b/drivers/hv/hv_balloon.c index 64ac5bdee3a6..87b312f99b2e 100644 --- a/drivers/hv/hv_balloon.c +++ b/drivers/hv/hv_balloon.c@@ -1101,6 +1101,10 @@ static void process_info(struct hv_dynmem_device *dm, struct dm_info_msg *msg) } } +unsigned long balloon_floor; +module_param(balloon_floor, ulong, 0644); +MODULE_PARM_DESC(balloon_floor, "Memory level (in megabytes) that ballooning will not remove"); + static unsigned long compute_balloon_floor(void) { unsigned long min_pages;@@ -1117,6 +1121,9 @@ static unsigned long compute_balloon_floor(void) * 8192 744 (1/16) * 32768 1512 (1/32) */ + if (balloon_floor) + return MB2PAGES(balloon_floor); + if (nr_pages < MB2PAGES(128)) min_pages = MB2PAGES(8) + (nr_pages >> 1); else if (nr_pages < MB2PAGES(512))
A module parameter is probably useful for debugging but it can hardly be applied in production environments as it must be 'one size fits all' and e.g. for different VM sizes it can be different (that's the purpose of compute_balloon_floor() heuristics). In fact, does it has to be statically set? Can we have a sysfs entity so this can be a policy (userspace decision)? We can keep the current compute_balloon_floor() as the default but users will be able to adjust accordingly. -- Vitaly