tjh.dev
Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
KVM: PPC: Book3S: PR: Optimize entry path
By shuffling a few instructions around we can execute more memory
loads in parallel, giving us a small performance boost.
With this patch and a simple priviledged SPR access loop guest, I get
a speed bump from 2013052 to 2035607 exits per second.
Signed-off-by: Alexander Graf <agraf@suse.de>
+7
-6
+7
-6
arch/powerpc/kvm/book3s_segment.S
+7
-6
arch/powerpc/kvm/book3s_segment.S
···
128
128
/* First clear RI in our current MSR value */
129
129
li r0, MSR_RI
130
130
andc r6, r6, r0
131
-
MTMSR_EERI(r6)
132
-
mtsrr0 r9
133
-
mtsrr1 r4
134
131
135
132
PPC_LL r0, SVCPU_R0(r3)
136
133
PPC_LL r1, SVCPU_R1(r3)
137
134
PPC_LL r2, SVCPU_R2(r3)
138
-
PPC_LL r4, SVCPU_R4(r3)
139
135
PPC_LL r5, SVCPU_R5(r3)
140
-
PPC_LL r6, SVCPU_R6(r3)
141
136
PPC_LL r7, SVCPU_R7(r3)
142
137
PPC_LL r8, SVCPU_R8(r3)
143
-
PPC_LL r9, SVCPU_R9(r3)
144
138
PPC_LL r10, SVCPU_R10(r3)
145
139
PPC_LL r11, SVCPU_R11(r3)
146
140
PPC_LL r12, SVCPU_R12(r3)
147
141
PPC_LL r13, SVCPU_R13(r3)
148
142
143
+
MTMSR_EERI(r6)
144
+
mtsrr0 r9
145
+
mtsrr1 r4
146
+
147
+
PPC_LL r4, SVCPU_R4(r3)
148
+
PPC_LL r6, SVCPU_R6(r3)
149
+
PPC_LL r9, SVCPU_R9(r3)
149
150
PPC_LL r3, (SVCPU_R3)(r3)
150
151
151
152
RFI