by default translation of virtqueue descriptors is done with
caching enabled, but caching will add only extra cost
in cases of trashing workload where majority descriptors
are translated to different memory regions.
So add an option to allow exclude cache miss cost for such cases.
Performance with cashing enabled for sequential workload
doesn't seem to be affected much vs version without static key switch,
i.e. still the same 0.2% of total time with key(NOPs) consuming
5ms on 5min workload.
Signed-off-by: Igor Mammedov <imammedo@xxxxxxxxxx>
---
I don't have a test case for trashing workload though, but jmp
instruction adds up ~6ms(55M instructions) minus excluded caching
around 24ms on 5min workload.
---
drivers/vhost/vhost.c | 20 ++++++++++++++++----
1 file changed, 16 insertions(+), 4 deletions(-)
diff --git a/drivers/vhost/vhost.c b/drivers/vhost/vhost.c
index 5bcb323..78290b7 100644
--- a/drivers/vhost/vhost.c
+++ b/drivers/vhost/vhost.c
@@ -29,6 +29,13 @@
#include "vhost.h"
+struct static_key translation_cache_key = STATIC_KEY_INIT_TRUE;
+static bool translation_cache = true;
+module_param(translation_cache, bool, 0444);
+MODULE_PARM_DESC(translation_cache,
+ "Enables/disables virtqueue descriptor translation caching,"
+ " Set to 0 to disable. (default: 1)");
+
enum {
VHOST_MEMORY_MAX_NREGIONS = 64,
VHOST_MEMORY_F_LOG = 0x1,
@@ -944,10 +951,12 @@ static const struct vhost_memory_region *find_region(struct vhost_memory *mem,
const struct vhost_memory_region *reg;
int start = 0, end = mem->nregions;
- reg = mem->regions + *cached_reg;
- if (likely(addr >= reg->guest_phys_addr &&
- reg->guest_phys_addr + reg->memory_size > addr))
- return reg;
+ if (static_key_true(&translation_cache_key)) {
+ reg = mem->regions + *cached_reg;
+ if (likely(addr >= reg->guest_phys_addr &&
+ reg->guest_phys_addr + reg->memory_size > addr))
+ return reg;
+ }
while (start < end) {
int slot = start + (end - start) / 2;
@@ -1612,6 +1621,9 @@ EXPORT_SYMBOL_GPL(vhost_disable_notify);
static int __init vhost_init(void)
{
+ if (!translation_cache)
+ static_key_slow_dec(&translation_cache_key);
+
return 0;
}
--
1.8.3.1
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
