From a6ca7519f89c52e9fab205cded0f2bf32d914cd6 Mon Sep 17 00:00:00 2001 From: Justin Hibbits Date: Fri, 30 Apr 2021 19:58:11 -0500 Subject: powerpc64: Optimize radix trap handling a little more Summary: Since PCPU can live in a GPR for a while longer, let it, rather than re-getting it in yet another register. MFSPR is an expensive operation, 12 clock latency on POWER9, so the fewer operations we need, the better. Since the check is tightly coupled to the fetch, by reducing the number of fetch+check, we reduce the stalls, and improve the performance marginally. Buildworld was measured at a ~5-7% improvement on a single run. Reviewed By: nwhitehorn Differential Revision: https://reviews.freebsd.org/D30003 --- sys/powerpc/aim/trap_subr64.S | 22 +++++++++++----------- 1 file changed, 11 insertions(+), 11 deletions(-) diff --git a/sys/powerpc/aim/trap_subr64.S b/sys/powerpc/aim/trap_subr64.S index 8e0f43ed0eeb..8ab2c57be7cb 100644 --- a/sys/powerpc/aim/trap_subr64.S +++ b/sys/powerpc/aim/trap_subr64.S @@ -56,10 +56,10 @@ /* * User SRs are loaded through a pointer to the current pmap. + * PCPU already in %r3 */ restore_usersrs: - GET_CPUINFO(%r28) - ld %r28,PC_USERSLB(%r28) + ld %r28,PC_USERSLB(%r3) cmpdi %r28, 0 /* If user SLB pointer NULL, exit */ beqlr @@ -84,13 +84,13 @@ restore_usersrs: /* * Kernel SRs are loaded directly from the PCPU fields + * PCPU in %r1 */ restore_kernsrs: - GET_CPUINFO(%r28) - lwz %r29, PC_FLAGS(%r28) + lwz %r29, PC_FLAGS(%r1) mtcr %r29 btlr 0 - addi %r28,%r28,PC_KERNSLB + addi %r28,%r1,PC_KERNSLB ld %r29,16(%r28) /* One past USER_SLB_SLOT */ cmpdi %r29,0 beqlr /* If first kernel entry is invalid, @@ -269,21 +269,21 @@ restore_kernsrs: /* Restore user SRs */ \ GET_CPUINFO(%r3); \ std %r27,(savearea+CPUSAVE_R27)(%r3); \ + lwz %r27,PC_FLAGS(%r3); \ + mtcr %r27; \ + bt 0, 0f; /* Check to skip restoring SRs. */ \ std %r28,(savearea+CPUSAVE_R28)(%r3); \ std %r29,(savearea+CPUSAVE_R29)(%r3); \ std %r30,(savearea+CPUSAVE_R30)(%r3); \ std %r31,(savearea+CPUSAVE_R31)(%r3); \ - lwz %r28,PC_FLAGS(%r3); \ - mtcr %r28; \ - bt 0, 0f; /* Check to skip restoring SRs. */ \ mflr %r27; /* preserve LR */ \ bl restore_usersrs; /* uses r28-r31 */ \ mtlr %r27; \ -0: \ ld %r31,(savearea+CPUSAVE_R31)(%r3); \ ld %r30,(savearea+CPUSAVE_R30)(%r3); \ ld %r29,(savearea+CPUSAVE_R29)(%r3); \ ld %r28,(savearea+CPUSAVE_R28)(%r3); \ +0: \ ld %r27,(savearea+CPUSAVE_R27)(%r3); \ 1: mfsprg2 %r3; /* restore cr */ \ mtcr %r3; \ @@ -778,12 +778,12 @@ realtrap: overwritten) */ bf 17,k_trap /* branch if PSL_PR is false */ GET_CPUINFO(%r1) - ld %r1,PC_CURPCB(%r1) mr %r27,%r28 /* Save LR, r29 */ mtsprg2 %r29 bl restore_kernsrs /* enable kernel mapping */ mfsprg2 %r29 mr %r28,%r27 + ld %r1,PC_CURPCB(%r1) b s_trap /* @@ -839,12 +839,12 @@ s_trap: bf 17,k_trap /* branch if PSL_PR is false */ GET_CPUINFO(%r1) u_trap: - ld %r1,PC_CURPCB(%r1) mr %r27,%r28 /* Save LR, r29 */ mtsprg2 %r29 bl restore_kernsrs /* enable kernel mapping */ mfsprg2 %r29 mr %r28,%r27 + ld %r1,PC_CURPCB(%r1) /* * Now the common trap catching code. -- cgit v1.2.3