diff mbox series

mips64el-softmmu: Enable MTTCG

Message ID 20240511-mips_mttcg-v1-1-1b71d9b85234@flygoat.com
State New
Headers show
Series mips64el-softmmu: Enable MTTCG | expand

Commit Message

Jiaxun Yang May 11, 2024, 8:26 p.m. UTC 
MTTCG was disabled in a092a9554771 ("configure: disable MTTCG
for MIPS guests") due to test case instability.

I was able to reproduce this issue with in latest QEMU and look
into reason behind that.

What actually happend is kernel's CP0 timer synchronisation
mechanism assumed a consistent latency in memory access between
cores, which TCG can't guarantee. Thus there is a huge drift in
count value between cores, and in early kernel versions CP0 timer
is always used as sched_clock.

sched_clock drift back on some cores triggered RCU watchdog in
some extreme cases.

This can be resolved by setting clocksource to MIPS, which allows
clocksource to drift together with sched_clock. However this will
leed to other problems after boot.

Another option would beupdating kernel to later version, which
will use GIC as sched_clock.

In non-MTTCG build, the execution is slow enough so kernel won't
observe back drifts.

Test results:

With clocksource=MIPS
```
 ~/tmp/retry/retry.py -n 100 -c -- ./qemu-system-mips64el \
    -display none -vga none -serial mon:stdio \
    -machine malta -kernel ./vmlinux-4.7.0-rc1.I6400 \
    -cpu I6400 -smp 8 -vga std \
    -append "printk.time=0 clocksource=MIPS console=tty0 console=ttyS0 panic=-1" \
    --no-reboot

100, 0, PASS, 5.258126, 100, 100, -
Results summary:
0: 100 times (100.00%), avg time 6.508 (55.53 varience/7.45 deviation)
Ran command 100 times, 100 passes
```

With linux-next:
```
 ~/tmp/retry/retry.py -n 100 -c -- ./qemu-system-mips64el \
    -display none -vga none -serial mon:stdio \
    -machine malta -kernel ~/linux-next/vmlinux \
    -cpu I6400 -smp 8 -vga std \
    -append "printk.time=0 console=tty0 console=ttyS0 panic=-1" \
    --no-reboot

100, 0, PASS, 4.507921, 100, 100, -
Results summary:
0: 100 times (100.00%), avg time 4.233 (0.04 varience/0.21 deviation)
Ran command 100 times, 100 passes
```

Signed-off-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
---
I'll leave the test case alone as it's already marked as QEMU_TEST_FLAKY_TESTS
---
 configs/targets/mips64el-softmmu.mak | 1 +
 1 file changed, 1 insertion(+)


---
base-commit: 248f6f62df073a3b4158fd0093863ab885feabb5
change-id: 20240511-mips_mttcg-47a6b19074b3

Best regards,
diff mbox series

Patch

diff --git a/configs/targets/mips64el-softmmu.mak b/configs/targets/mips64el-softmmu.mak
index 8d9ab3ddc4b1..199b1d909a7d 100644
--- a/configs/targets/mips64el-softmmu.mak
+++ b/configs/targets/mips64el-softmmu.mak
@@ -1,3 +1,4 @@ 
 TARGET_ARCH=mips64
 TARGET_BASE_ARCH=mips
+TARGET_SUPPORTS_MTTCG=y
 TARGET_NEED_FDT=y