[PATCH] powerpc: Fix hypervisor TLB batching
Benjamin Herrenschmidt
benh at kernel.crashing.org
Mon Oct 12 19:44:33 EST 2009
On Mon, 2009-10-12 at 18:47 +1100, Anton Blanchard wrote:
> Profiling of a page fault scalability microbenchmark shows flush_hash_range
> is not calling the batch hpte invalidate hcall (H_BULK_REMOVE).
>
> It turns out we have a duplicate firmware feature for hcall-bulk and the
> current setup code stops after finding the first match. This meant we never
> batch and always do individual invalidates.
>
> The patch below removes the duplicate and shifts FW_FEATURE_CMO to close
> the gap. With the patch applied the single threaded page fault rate improves
> from 217169 to 238755 per second on a POWER5 test box, a 10% improvement.
>
> Signed-off-by: Anton Blanchard <anton at samba.org>
> ---
Good catch !
Ben.
> Index: linux.trees.git/arch/powerpc/include/asm/firmware.h
> ===================================================================
> --- linux.trees.git.orig/arch/powerpc/include/asm/firmware.h 2009-10-12 18:11:56.000000000 +1100
> +++ linux.trees.git/arch/powerpc/include/asm/firmware.h 2009-10-12 18:21:20.000000000 +1100
> @@ -37,7 +37,7 @@
> #define FW_FEATURE_VIO ASM_CONST(0x0000000000004000)
> #define FW_FEATURE_RDMA ASM_CONST(0x0000000000008000)
> #define FW_FEATURE_LLAN ASM_CONST(0x0000000000010000)
> -#define FW_FEATURE_BULK ASM_CONST(0x0000000000020000)
> +#define FW_FEATURE_BULK_REMOVE ASM_CONST(0x0000000000020000)
> #define FW_FEATURE_XDABR ASM_CONST(0x0000000000040000)
> #define FW_FEATURE_MULTITCE ASM_CONST(0x0000000000080000)
> #define FW_FEATURE_SPLPAR ASM_CONST(0x0000000000100000)
> @@ -45,8 +45,7 @@
> #define FW_FEATURE_LPAR ASM_CONST(0x0000000000400000)
> #define FW_FEATURE_PS3_LV1 ASM_CONST(0x0000000000800000)
> #define FW_FEATURE_BEAT ASM_CONST(0x0000000001000000)
> -#define FW_FEATURE_BULK_REMOVE ASM_CONST(0x0000000002000000)
> -#define FW_FEATURE_CMO ASM_CONST(0x0000000004000000)
> +#define FW_FEATURE_CMO ASM_CONST(0x0000000002000000)
>
> #ifndef __ASSEMBLY__
>
> @@ -58,8 +57,9 @@ enum {
> FW_FEATURE_PERF | FW_FEATURE_DUMP | FW_FEATURE_INTERRUPT |
> FW_FEATURE_MIGRATE | FW_FEATURE_PERFMON | FW_FEATURE_CRQ |
> FW_FEATURE_VIO | FW_FEATURE_RDMA | FW_FEATURE_LLAN |
> - FW_FEATURE_BULK | FW_FEATURE_XDABR | FW_FEATURE_MULTITCE |
> - FW_FEATURE_SPLPAR | FW_FEATURE_LPAR | FW_FEATURE_CMO,
> + FW_FEATURE_BULK_REMOVE | FW_FEATURE_XDABR |
> + FW_FEATURE_MULTITCE | FW_FEATURE_SPLPAR | FW_FEATURE_LPAR |
> + FW_FEATURE_CMO,
> FW_FEATURE_PSERIES_ALWAYS = 0,
> FW_FEATURE_ISERIES_POSSIBLE = FW_FEATURE_ISERIES | FW_FEATURE_LPAR,
> FW_FEATURE_ISERIES_ALWAYS = FW_FEATURE_ISERIES | FW_FEATURE_LPAR,
> Index: linux.trees.git/arch/powerpc/platforms/pseries/firmware.c
> ===================================================================
> --- linux.trees.git.orig/arch/powerpc/platforms/pseries/firmware.c 2009-10-12 18:08:37.000000000 +1100
> +++ linux.trees.git/arch/powerpc/platforms/pseries/firmware.c 2009-10-12 18:12:54.000000000 +1100
> @@ -51,11 +51,10 @@ firmware_features_table[FIRMWARE_MAX_FEA
> {FW_FEATURE_VIO, "hcall-vio"},
> {FW_FEATURE_RDMA, "hcall-rdma"},
> {FW_FEATURE_LLAN, "hcall-lLAN"},
> - {FW_FEATURE_BULK, "hcall-bulk"},
> + {FW_FEATURE_BULK_REMOVE, "hcall-bulk"},
> {FW_FEATURE_XDABR, "hcall-xdabr"},
> {FW_FEATURE_MULTITCE, "hcall-multi-tce"},
> {FW_FEATURE_SPLPAR, "hcall-splpar"},
> - {FW_FEATURE_BULK_REMOVE, "hcall-bulk"},
> };
>
> /* Build up the firmware features bitmask using the contents of
More information about the Linuxppc-dev
mailing list