diff options
| author | Christian Krinitsin <mail@krinitsin.com> | 2025-06-01 21:19:55 +0200 |
|---|---|---|
| committer | Christian Krinitsin <mail@krinitsin.com> | 2025-06-01 21:19:55 +0200 |
| commit | e5634e2806195bee44407853c4bf8776f7abfa4f (patch) | |
| tree | 6c6e2503f2cd56c0edf08a55d5c8fcb3832fabd0 /classification_output/05/graphic | |
| parent | 05ed9b00104b3da2e9cb0d541b5c5bbceb027fde (diff) | |
| download | qemu-analysis-e5634e2806195bee44407853c4bf8776f7abfa4f.tar.gz qemu-analysis-e5634e2806195bee44407853c4bf8776f7abfa4f.zip | |
add the outputs of the first five revisions of the classifier
Diffstat (limited to 'classification_output/05/graphic')
| -rw-r--r-- | classification_output/05/graphic/22219210 | 51 | ||||
| -rw-r--r-- | classification_output/05/graphic/30680944 | 603 | ||||
| -rw-r--r-- | classification_output/05/graphic/46572227 | 414 | ||||
| -rw-r--r-- | classification_output/05/graphic/53568181 | 86 | ||||
| -rw-r--r-- | classification_output/05/graphic/55961334 | 47 | ||||
| -rw-r--r-- | classification_output/05/graphic/73660729 | 39 |
6 files changed, 1240 insertions, 0 deletions
diff --git a/classification_output/05/graphic/22219210 b/classification_output/05/graphic/22219210 new file mode 100644 index 000000000..f05331491 --- /dev/null +++ b/classification_output/05/graphic/22219210 @@ -0,0 +1,51 @@ +graphic: 0.701 +device: 0.489 +mistranslation: 0.472 +semantic: 0.387 +other: 0.345 +network: 0.323 +instruction: 0.261 +socket: 0.244 +vnc: 0.204 +KVM: 0.099 +assembly: 0.078 +boot: 0.070 + +[BUG][CPU hot-plug]CPU hot-plugs cause the qemu process to coredump + +Hello,Recently, when I was developing CPU hot-plugs under the loongarch +architecture, +I found that there was a problem with qemu cpu hot-plugs under x86 +architecture, +which caused the qemu process coredump when repeatedly inserting and +unplugging +the CPU when the TCG was accelerated. + + +The specific operation process is as follows: + +1.Use the following command to start the virtual machine + +qemu-system-x86_64 \ +-machine q35 \ +-cpu Broadwell-IBRS \ +-smp 1,maxcpus=4,sockets=4,cores=1,threads=1 \ +-m 4G \ +-drive file=~/anolis-8.8.qcow2 \ +-serial stdio  \ +-monitor telnet:localhost:4498,server,nowait + + +2.Enter QEMU Monitor via telnet for repeated CPU insertion and unplugging + +telnet 127.0.0.1 4498 +(qemu) device_add +Broadwell-IBRS-x86_64-cpu,socket-id=1,core-id=0,thread-id=0,id=cpu1 +(qemu) device_del cpu1 +(qemu) device_add +Broadwell-IBRS-x86_64-cpu,socket-id=1,core-id=0,thread-id=0,id=cpu1 +3.You will notice that the QEMU process has a coredump + +# malloc(): unsorted double linked list corrupted +Aborted (core dumped) + diff --git a/classification_output/05/graphic/30680944 b/classification_output/05/graphic/30680944 new file mode 100644 index 000000000..f0d07b115 --- /dev/null +++ b/classification_output/05/graphic/30680944 @@ -0,0 +1,603 @@ +graphic: 0.965 +semantic: 0.953 +other: 0.944 +assembly: 0.940 +device: 0.935 +instruction: 0.919 +socket: 0.864 +boot: 0.840 +vnc: 0.815 +network: 0.813 +mistranslation: 0.799 +KVM: 0.701 + +[BUG]QEMU jump into interrupt when single-stepping on aarch64 + +Dear, folks, + +I try to debug Linux kernel with QEMU in single-stepping mode on aarch64 +platform, +the added breakpoint hits but after I type `step`, the gdb always jumps into +interrupt. + +My env: + + gdb-10.2 + qemu-6.2.0 + host kernel: 5.10.84 + VM kernel: 5.10.84 + +The steps to reproduce: + # host console: run a VM with only one core, the import arg: <qemu:arg +value='-s'/> + # details can be found here: +https://www.redhat.com/en/blog/debugging-kernel-qemulibvirt +virsh create dev_core0.xml + + # run gdb client + gdb ./vmlinux + + # gdb client on host console + (gdb) dir +./usr/src/debug/kernel-5.10.84/linux-5.10.84-004.alpha.ali5000.alios7.aarch64 + (gdb) target remote localhost:1234 + (gdb) info b + Num Type Disp Enb Address What + 1 breakpoint keep y <MULTIPLE> + 1.1 y 0xffff800010361444 +mm/memory-failure.c:1318 + 1.2 y 0xffff800010361450 in memory_failure + at mm/memory-failure.c:1488 + (gdb) c + Continuing. + + # console in VM, use madvise to inject a hwposion at virtual address +vaddr, + # which will hit the b inmemory_failur: madvise(vaddr, pagesize, +MADV_HWPOISON); + # and the VM pause + ./run_madvise.c + + # gdb client on host console + (gdb) + Continuing. + Breakpoint 1, 0xffff800010361444 in memory_failure () at +mm/memory-failure.c:1318 + 1318 res = -EHWPOISON; + (gdb) n + vectors () at arch/arm64/kernel/entry.S:552 + 552 kernel_ventry 1, irq // IRQ +EL1h + (gdb) n + (gdb) n + (gdb) n + (gdb) n + gic_handle_irq (regs=0xffff8000147c3b80) at +drivers/irqchip/irq-gic-v3.c:721 + # after several step, I got the irqnr + (gdb) p irqnr + $5 = 8262 + +Sometimes, the irqnr is 27ï¼ which is used for arch_timer. + +I was wondering do you have any comments on this? And feedback are welcomed. + +Thank you. + +Best Regards. +Shuai + +On 4/6/22 09:30, Shuai Xue wrote: +Dear, folks, + +I try to debug Linux kernel with QEMU in single-stepping mode on aarch64 +platform, +the added breakpoint hits but after I type `step`, the gdb always jumps into +interrupt. + +My env: + + gdb-10.2 + qemu-6.2.0 + host kernel: 5.10.84 + VM kernel: 5.10.84 + +The steps to reproduce: + # host console: run a VM with only one core, the import arg: <qemu:arg +value='-s'/> + # details can be found here: +https://www.redhat.com/en/blog/debugging-kernel-qemulibvirt +virsh create dev_core0.xml + + # run gdb client + gdb ./vmlinux + + # gdb client on host console + (gdb) dir +./usr/src/debug/kernel-5.10.84/linux-5.10.84-004.alpha.ali5000.alios7.aarch64 + (gdb) target remote localhost:1234 + (gdb) info b + Num Type Disp Enb Address What + 1 breakpoint keep y <MULTIPLE> + 1.1 y 0xffff800010361444 +mm/memory-failure.c:1318 + 1.2 y 0xffff800010361450 in memory_failure + at mm/memory-failure.c:1488 + (gdb) c + Continuing. + + # console in VM, use madvise to inject a hwposion at virtual address +vaddr, + # which will hit the b inmemory_failur: madvise(vaddr, pagesize, +MADV_HWPOISON); + # and the VM pause + ./run_madvise.c + + # gdb client on host console + (gdb) + Continuing. + Breakpoint 1, 0xffff800010361444 in memory_failure () at +mm/memory-failure.c:1318 + 1318 res = -EHWPOISON; + (gdb) n + vectors () at arch/arm64/kernel/entry.S:552 + 552 kernel_ventry 1, irq // IRQ +EL1h +The 'n' command is not a single-step: use stepi, which will suppress interrupts. +Anyway, not a bug. + +r~ + +å¨ 2022/4/7 AM12:57, Richard Henderson åé: +> +On 4/6/22 09:30, Shuai Xue wrote: +> +> Dear, folks, +> +> +> +> I try to debug Linux kernel with QEMU in single-stepping mode on aarch64 +> +> platform, +> +> the added breakpoint hits but after I type `step`, the gdb always jumps into +> +> interrupt. +> +> +> +> My env: +> +> +> +>     gdb-10.2 +> +>     qemu-6.2.0 +> +>     host kernel: 5.10.84 +> +>     VM kernel: 5.10.84 +> +> +> +> The steps to reproduce: +> +>     # host console: run a VM with only one core, the import arg: <qemu:arg +> +> value='-s'/> +> +>     # details can be found here: +> +> +https://www.redhat.com/en/blog/debugging-kernel-qemulibvirt +> +>     virsh create dev_core0.xml +> +>     +> +>     # run gdb client +> +>     gdb ./vmlinux +> +> +> +>     # gdb client on host console +> +>     (gdb) dir +> +> ./usr/src/debug/kernel-5.10.84/linux-5.10.84-004.alpha.ali5000.alios7.aarch64 +> +>     (gdb) target remote localhost:1234 +> +>     (gdb) info b +> +>     Num    Type          Disp Enb Address           What +> +>     1      breakpoint    keep y  <MULTIPLE> +> +>     1.1                        y  0xffff800010361444 +> +> mm/memory-failure.c:1318 +> +>     1.2                        y  0xffff800010361450 in memory_failure +> +>                                                    at +> +> mm/memory-failure.c:1488 +> +>     (gdb) c +> +>     Continuing. +> +> +> +>     # console in VM, use madvise to inject a hwposion at virtual address +> +> vaddr, +> +>     # which will hit the b inmemory_failur: madvise(vaddr, pagesize, +> +> MADV_HWPOISON); +> +>     # and the VM pause +> +>     ./run_madvise.c +> +> +> +>     # gdb client on host console +> +>     (gdb) +> +>     Continuing. +> +>     Breakpoint 1, 0xffff800010361444 in memory_failure () at +> +> mm/memory-failure.c:1318 +> +>     1318                   res = -EHWPOISON; +> +>     (gdb) n +> +>     vectors () at arch/arm64/kernel/entry.S:552 +> +>     552            kernel_ventry  1, irq                         // IRQ +> +> EL1h +> +> +The 'n' command is not a single-step: use stepi, which will suppress +> +interrupts. +> +Anyway, not a bug. +> +> +r~ +Hi, Richard, + +Thank you for your quick reply, I also try `stepi`, but it does NOT work either. + + (gdb) c + Continuing. + + Breakpoint 1, memory_failure (pfn=1273982, flags=1) at +mm/memory-failure.c:1488 + 1488 { + (gdb) stepi + vectors () at arch/arm64/kernel/entry.S:552 + 552 kernel_ventry 1, irq // IRQ +EL1h + +According to QEMU doc[1]: the default single stepping behavior is step with the +IRQs +and timer service routines off. I checked the MASK bits used to control the +single +stepping IE on my machine as bellow: + + # gdb client on host (x86 plafrom) + (gdb) maintenance packet qqemu.sstepbits + sending: "qqemu.sstepbits" + received: "ENABLE=1,NOIRQ=2,NOTIMER=4" + +The sstep MASK looks as expected, but does not work as expected. + +I also try the same kernel and qemu version on X86 platform: +> +> gdb-10.2 +> +> qemu-6.2.0 +> +> host kernel: 5.10.84 +> +> VM kernel: 5.10.84 +The command `n` jumps to the next instruction. + + # gdb client on host (x86 plafrom) + (gdb) b memory-failure.c:1488 + Breakpoint 1, memory_failure (pfn=1128931, flags=1) at +mm/memory-failure.c:1488 + 1488 { + (gdb) n + 1497 if (!sysctl_memory_failure_recovery) + (gdb) stepi + 0xffffffff812efdbc 1497 if +(!sysctl_memory_failure_recovery) + (gdb) stepi + 0xffffffff812efdbe 1497 if +(!sysctl_memory_failure_recovery) + (gdb) n + 1500 p = pfn_to_online_page(pfn); + (gdb) l + 1496 + 1497 if (!sysctl_memory_failure_recovery) + 1498 panic("Memory failure on page %lx", pfn); + 1499 + 1500 p = pfn_to_online_page(pfn); + 1501 if (!p) { + +Best Regrades, +Shuai + + +[1] +https://github.com/qemu/qemu/blob/master/docs/system/gdb.rst + +å¨ 2022/4/7 PM12:10, Shuai Xue åé: +> +å¨ 2022/4/7 AM12:57, Richard Henderson åé: +> +> On 4/6/22 09:30, Shuai Xue wrote: +> +>> Dear, folks, +> +>> +> +>> I try to debug Linux kernel with QEMU in single-stepping mode on aarch64 +> +>> platform, +> +>> the added breakpoint hits but after I type `step`, the gdb always jumps +> +>> into interrupt. +> +>> +> +>> My env: +> +>> +> +>>     gdb-10.2 +> +>>     qemu-6.2.0 +> +>>     host kernel: 5.10.84 +> +>>     VM kernel: 5.10.84 +> +>> +> +>> The steps to reproduce: +> +>>     # host console: run a VM with only one core, the import arg: <qemu:arg +> +>> value='-s'/> +> +>>     # details can be found here: +> +>> +https://www.redhat.com/en/blog/debugging-kernel-qemulibvirt +> +>>     virsh create dev_core0.xml +> +>>     +> +>>     # run gdb client +> +>>     gdb ./vmlinux +> +>> +> +>>     # gdb client on host console +> +>>     (gdb) dir +> +>> ./usr/src/debug/kernel-5.10.84/linux-5.10.84-004.alpha.ali5000.alios7.aarch64 +> +>>     (gdb) target remote localhost:1234 +> +>>     (gdb) info b +> +>>     Num    Type          Disp Enb Address           What +> +>>     1      breakpoint    keep y  <MULTIPLE> +> +>>     1.1                        y  0xffff800010361444 +> +>> mm/memory-failure.c:1318 +> +>>     1.2                        y  0xffff800010361450 in memory_failure +> +>>                                                    at +> +>> mm/memory-failure.c:1488 +> +>>     (gdb) c +> +>>     Continuing. +> +>> +> +>>     # console in VM, use madvise to inject a hwposion at virtual address +> +>> vaddr, +> +>>     # which will hit the b inmemory_failur: madvise(vaddr, pagesize, +> +>> MADV_HWPOISON); +> +>>     # and the VM pause +> +>>     ./run_madvise.c +> +>> +> +>>     # gdb client on host console +> +>>     (gdb) +> +>>     Continuing. +> +>>     Breakpoint 1, 0xffff800010361444 in memory_failure () at +> +>> mm/memory-failure.c:1318 +> +>>     1318                   res = -EHWPOISON; +> +>>     (gdb) n +> +>>     vectors () at arch/arm64/kernel/entry.S:552 +> +>>     552            kernel_ventry  1, irq                         // IRQ +> +>> EL1h +> +> +> +> The 'n' command is not a single-step: use stepi, which will suppress +> +> interrupts. +> +> Anyway, not a bug. +> +> +> +> r~ +> +> +Hi, Richard, +> +> +Thank you for your quick reply, I also try `stepi`, but it does NOT work +> +either. +> +> +(gdb) c +> +Continuing. +> +> +Breakpoint 1, memory_failure (pfn=1273982, flags=1) at +> +mm/memory-failure.c:1488 +> +1488 { +> +(gdb) stepi +> +vectors () at arch/arm64/kernel/entry.S:552 +> +552 kernel_ventry 1, irq // IRQ +> +EL1h +> +> +According to QEMU doc[1]: the default single stepping behavior is step with +> +the IRQs +> +and timer service routines off. I checked the MASK bits used to control the +> +single +> +stepping IE on my machine as bellow: +> +> +# gdb client on host (x86 plafrom) +> +(gdb) maintenance packet qqemu.sstepbits +> +sending: "qqemu.sstepbits" +> +received: "ENABLE=1,NOIRQ=2,NOTIMER=4" +> +> +The sstep MASK looks as expected, but does not work as expected. +> +> +I also try the same kernel and qemu version on X86 platform: +> +>> gdb-10.2 +> +>> qemu-6.2.0 +> +>> host kernel: 5.10.84 +> +>> VM kernel: 5.10.84 +> +> +> +The command `n` jumps to the next instruction. +> +> +# gdb client on host (x86 plafrom) +> +(gdb) b memory-failure.c:1488 +> +Breakpoint 1, memory_failure (pfn=1128931, flags=1) at +> +mm/memory-failure.c:1488 +> +1488 { +> +(gdb) n +> +1497 if (!sysctl_memory_failure_recovery) +> +(gdb) stepi +> +0xffffffff812efdbc 1497 if +> +(!sysctl_memory_failure_recovery) +> +(gdb) stepi +> +0xffffffff812efdbe 1497 if +> +(!sysctl_memory_failure_recovery) +> +(gdb) n +> +1500 p = pfn_to_online_page(pfn); +> +(gdb) l +> +1496 +> +1497 if (!sysctl_memory_failure_recovery) +> +1498 panic("Memory failure on page %lx", pfn); +> +1499 +> +1500 p = pfn_to_online_page(pfn); +> +1501 if (!p) { +> +> +Best Regrades, +> +Shuai +> +> +> +[1] +https://github.com/qemu/qemu/blob/master/docs/system/gdb.rst +Hi, Richard, + +I was wondering that do you have any comments to this? + +Best Regrades, +Shuai + diff --git a/classification_output/05/graphic/46572227 b/classification_output/05/graphic/46572227 new file mode 100644 index 000000000..ae72af541 --- /dev/null +++ b/classification_output/05/graphic/46572227 @@ -0,0 +1,414 @@ +semantic: 0.965 +graphic: 0.962 +mistranslation: 0.946 +assembly: 0.931 +other: 0.927 +instruction: 0.906 +vnc: 0.904 +device: 0.901 +boot: 0.900 +KVM: 0.857 +network: 0.841 +socket: 0.841 + +[Qemu-devel] [Bug?] Windows 7's time drift obviously while RTC rate switching frequently between high and low timer rate + +Hi, + +We tested with the latest QEMU, and found that time drift obviously (clock fast +in guest) +in Windows 7 64 bits guest in some cases. + +It is easily to reproduce, using the follow QEMU command line to start windows +7: + +# x86_64-softmmu/qemu-system-x86_64 -name win7_64_2U_raw -machine +pc-i440fx-2.6,accel=kvm,usb=off -cpu host -m 2048 -realtime mlock=off -smp +4,sockets=2,cores=2,threads=1 -rtc base=utc,clock=vm,driftfix=slew -no-hpet +-global kvm-pit.lost_tick_policy=discard -hda /mnt/nfs/win7_sp1_32_2U_raw -vnc +:11 -netdev tap,id=hn0,vhost=off -device rtl8139,id=net-pci0,netdev=hn0 -device +piix3-usb-uhci,id=usb -device usb-tablet,id=input0 -device usb-mouse,id=input1 +-device usb-kbd,id=input2 -monitor stdio + +Adjust the VM's time to host time, and run java application or run the follow +program +in windows 7: + +#pragma comment(lib, "winmm") +#include <stdio.h> +#include <windows.h> + +#define SWITCH_PEROID 13 + +int main() +{ + DWORD count = 0; + + while (1) + { + count++; + timeBeginPeriod(1); + DWORD start = timeGetTime(); + Sleep(40); + timeEndPeriod(1); + if ((count % SWITCH_PEROID) == 0) { + Sleep(1); + } + } + return 0; +} + +After few minutes, you will find that the time in windows 7 goes ahead of the +host time, drifts about several seconds. + +I have dug deeper in this problem. For windows systems that use the CMOS timer, +the base interrupt rate is usually 64Hz, but running some application in VM +will raise the timer rate to 1024Hz, running java application and or above +program will raise the timer rate. +Besides, Windows operating systems generally keep time by counting timer +interrupts (ticks). But QEMU seems not emulate the rate converting fine. + +We update the timer in function periodic_timer_update(): +static void periodic_timer_update(RTCState *s, int64_t current_time) +{ + + cur_clock = muldiv64(current_time, RTC_CLOCK_RATE, get_ticks_per_sec()); + next_irq_clock = (cur_clock & ~(period - 1)) + period; + ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +Here we calculate the next interrupt time by align the current clock with the +new period, I'm a little confused that why we care about the *history* time ? +If VM switches from high rate to low rate, the next interrupt time may come +earlier than it supposed to be. We have observed it in our test. we printed the +interval time of interrupts and the VM's current time (We got the time from VM). + +Here is part of the log: +... ... +period=512 irq inject 1534: 15625 us +Tue Mar 29 04:38:00 2016 +*irq_num_period_32=0, irq_num_period_512=64: [3]: Real time interval is 999696 +us +... ... +*irq_num_period_32=893, irq_num_period_512=9 [81]: Real time interval is 951086 +us +Convert 32 --- > 512: 703: 96578 us +period=512 irq inject 44391: 12702 us +Convert 512 --- > 32: 704: 12704 us11 +period=32 irq inject 44392: 979 us +... ... +32 --- > 512: 705: 24388 us +period=512 irq inject 44417: 6834 us +Convert 512 --- > 32: 706: 6830 us +period=32 irq inject 44418: 978 us +... ... +Convert 32 --- > 512: 707: 60525 us +period=512 irq inject 44480: 1945 us +Convert 512 --- > 32: 708: 1955 us +period=32 irq inject 44481: 977 us +... ... +Convert 32 --- > 512: 709: 36105 us +period=512 irq inject 44518: 10741 us +Convert 512 --- > 32: 710: 10736 us +period=32 irq inject 44519: 989 us +... ... +Convert 32 --- > 512: 711: 123998 us +period=512 irq inject 44646: 974 us +period=512 irq inject 44647: 15607 us +Convert 512 --- > 32: 712: 16560 us +period=32 irq inject 44648: 980 us +... ... +period=32 irq inject 44738: 974 us +Convert 32 --- > 512: 713: 88828 us +period=512 irq inject 44739: 4885 us +Convert 512 --- > 32: 714: 4882 us +period=32 irq inject 44740: 989 us +... ... +period=32 irq inject 44842: 974 us +Convert 32 --- > 512: 715: 100537 us +period=512 irq inject 44843: 8788 us +Convert 512 --- > 32: 716: 8789 us +period=32 irq inject 44844: 972 us +... ... +period=32 irq inject 44941: 979 us +Convert 32 --- > 512: 717: 95677 us +period=512 irq inject 44942: 13661 us +Convert 512 --- > 32: 718: 13657 us +period=32 irq inject 44943: 987 us +... ... +Convert 32 --- > 512: 719: 94690 us +period=512 irq inject 45040: 14643 us +Convert 512 --- > 32: 720: 14642 us +period=32 irq inject 45041: 974 us +... ... +Convert 32 --- > 512: 721: 88848 us +period=512 irq inject 45132: 4892 us +Convert 512 --- > 32: 722: 4931 us +period=32 irq inject 45133: 964 us +... ... +Tue Mar 29 04:39:19 2016 +*irq_num_period_32:835, irq_num_period_512:11 [82], Real time interval is +911520 us + +For windows 7, it has got 835 IRQs which injected during the period of 32, +and got 11 IRQs that injected during the period of 512. it updated the +wall-clock +time with one second, because it supposed it has counted +(835*976.5+11*15625)= 987252.5 us, but the real interval time is 911520 us. + +IMHO, we should calculate the next interrupt time based on the time of last +interrupt injected, and it seems to be more similar with hardware CMOS timer +in this way. +Maybe someone can tell me the reason why we calculated the interrupt timer +in that way, or is it a bug ? ;) + +Thanks, +Hailiang + +ping... + +It seems that we can eliminate the drift by the following patch. +(I tested it for two hours, and there is no drift, before, the timer +in Windows 7 drifts about 2 seconds per minute.) I'm not sure if it is +the right way to solve the problem. +Any comments are welcomed. Thanks. + +From bd6acd577cbbc9d92d6376c770219470f184f7de Mon Sep 17 00:00:00 2001 +From: zhanghailiang <address@hidden> +Date: Thu, 31 Mar 2016 16:36:15 -0400 +Subject: [PATCH] timer/mc146818rtc: fix timer drift in Windows OS while RTC + rate converting frequently + +Signed-off-by: zhanghailiang <address@hidden> +--- + hw/timer/mc146818rtc.c | 25 ++++++++++++++++++++++--- + 1 file changed, 22 insertions(+), 3 deletions(-) + +diff --git a/hw/timer/mc146818rtc.c b/hw/timer/mc146818rtc.c +index 2ac0fd3..e39d2da 100644 +--- a/hw/timer/mc146818rtc.c ++++ b/hw/timer/mc146818rtc.c +@@ -79,6 +79,7 @@ typedef struct RTCState { + /* periodic timer */ + QEMUTimer *periodic_timer; + int64_t next_periodic_time; ++ uint64_t last_periodic_time; + /* update-ended timer */ + QEMUTimer *update_timer; + uint64_t next_alarm_time; +@@ -152,7 +153,8 @@ static void rtc_coalesced_timer(void *opaque) + static void periodic_timer_update(RTCState *s, int64_t current_time) + { + int period_code, period; +- int64_t cur_clock, next_irq_clock; ++ int64_t cur_clock, next_irq_clock, pre_irq_clock; ++ bool change = false; + + period_code = s->cmos_data[RTC_REG_A] & 0x0f; + if (period_code != 0 +@@ -165,14 +167,28 @@ static void periodic_timer_update(RTCState *s, int64_t +current_time) + if (period != s->period) { + s->irq_coalesced = (s->irq_coalesced * s->period) / period; + DPRINTF_C("cmos: coalesced irqs scaled to %d\n", s->irq_coalesced); ++ if (s->period && period) { ++ change = true; ++ } + } + s->period = period; + #endif + /* compute 32 khz clock */ + cur_clock = + muldiv64(current_time, RTC_CLOCK_RATE, NANOSECONDS_PER_SECOND); ++ if (change) { ++ int offset = 0; + +- next_irq_clock = (cur_clock & ~(period - 1)) + period; ++ pre_irq_clock = muldiv64(s->last_periodic_time, RTC_CLOCK_RATE, ++ NANOSECONDS_PER_SECOND); ++ if ((cur_clock - pre_irq_clock) > period) { ++ offset = (cur_clock - pre_irq_clock) / period; ++ } ++ s->irq_coalesced += offset; ++ next_irq_clock = pre_irq_clock + (offset + 1) * period; ++ } else { ++ next_irq_clock = (cur_clock & ~(period - 1)) + period; ++ } + s->next_periodic_time = muldiv64(next_irq_clock, +NANOSECONDS_PER_SECOND, + RTC_CLOCK_RATE) + 1; + timer_mod(s->periodic_timer, s->next_periodic_time); +@@ -187,7 +203,9 @@ static void periodic_timer_update(RTCState *s, int64_t +current_time) + static void rtc_periodic_timer(void *opaque) + { + RTCState *s = opaque; +- ++ int64_t next_periodic_time; ++ ++ next_periodic_time = s->next_periodic_time; + periodic_timer_update(s, s->next_periodic_time); + s->cmos_data[RTC_REG_C] |= REG_C_PF; + if (s->cmos_data[RTC_REG_B] & REG_B_PIE) { +@@ -204,6 +222,7 @@ static void rtc_periodic_timer(void *opaque) + DPRINTF_C("cmos: coalesced irqs increased to %d\n", + s->irq_coalesced); + } ++ s->last_periodic_time = next_periodic_time; + } else + #endif + qemu_irq_raise(s->irq); +-- +1.8.3.1 + + +On 2016/3/29 19:58, Hailiang Zhang wrote: +Hi, + +We tested with the latest QEMU, and found that time drift obviously (clock fast +in guest) +in Windows 7 64 bits guest in some cases. + +It is easily to reproduce, using the follow QEMU command line to start windows +7: + +# x86_64-softmmu/qemu-system-x86_64 -name win7_64_2U_raw -machine +pc-i440fx-2.6,accel=kvm,usb=off -cpu host -m 2048 -realtime mlock=off -smp +4,sockets=2,cores=2,threads=1 -rtc base=utc,clock=vm,driftfix=slew -no-hpet +-global kvm-pit.lost_tick_policy=discard -hda /mnt/nfs/win7_sp1_32_2U_raw -vnc +:11 -netdev tap,id=hn0,vhost=off -device rtl8139,id=net-pci0,netdev=hn0 -device +piix3-usb-uhci,id=usb -device usb-tablet,id=input0 -device usb-mouse,id=input1 +-device usb-kbd,id=input2 -monitor stdio + +Adjust the VM's time to host time, and run java application or run the follow +program +in windows 7: + +#pragma comment(lib, "winmm") +#include <stdio.h> +#include <windows.h> + +#define SWITCH_PEROID 13 + +int main() +{ + DWORD count = 0; + + while (1) + { + count++; + timeBeginPeriod(1); + DWORD start = timeGetTime(); + Sleep(40); + timeEndPeriod(1); + if ((count % SWITCH_PEROID) == 0) { + Sleep(1); + } + } + return 0; +} + +After few minutes, you will find that the time in windows 7 goes ahead of the +host time, drifts about several seconds. + +I have dug deeper in this problem. For windows systems that use the CMOS timer, +the base interrupt rate is usually 64Hz, but running some application in VM +will raise the timer rate to 1024Hz, running java application and or above +program will raise the timer rate. +Besides, Windows operating systems generally keep time by counting timer +interrupts (ticks). But QEMU seems not emulate the rate converting fine. + +We update the timer in function periodic_timer_update(): +static void periodic_timer_update(RTCState *s, int64_t current_time) +{ + + cur_clock = muldiv64(current_time, RTC_CLOCK_RATE, +get_ticks_per_sec()); + next_irq_clock = (cur_clock & ~(period - 1)) + period; + ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ +Here we calculate the next interrupt time by align the current clock with the +new period, I'm a little confused that why we care about the *history* time ? +If VM switches from high rate to low rate, the next interrupt time may come +earlier than it supposed to be. We have observed it in our test. we printed the +interval time of interrupts and the VM's current time (We got the time from VM). + +Here is part of the log: +... ... +period=512 irq inject 1534: 15625 us +Tue Mar 29 04:38:00 2016 +*irq_num_period_32=0, irq_num_period_512=64: [3]: Real time interval is 999696 +us +... ... +*irq_num_period_32=893, irq_num_period_512=9 [81]: Real time interval is 951086 +us +Convert 32 --- > 512: 703: 96578 us +period=512 irq inject 44391: 12702 us +Convert 512 --- > 32: 704: 12704 us11 +period=32 irq inject 44392: 979 us +... ... +32 --- > 512: 705: 24388 us +period=512 irq inject 44417: 6834 us +Convert 512 --- > 32: 706: 6830 us +period=32 irq inject 44418: 978 us +... ... +Convert 32 --- > 512: 707: 60525 us +period=512 irq inject 44480: 1945 us +Convert 512 --- > 32: 708: 1955 us +period=32 irq inject 44481: 977 us +... ... +Convert 32 --- > 512: 709: 36105 us +period=512 irq inject 44518: 10741 us +Convert 512 --- > 32: 710: 10736 us +period=32 irq inject 44519: 989 us +... ... +Convert 32 --- > 512: 711: 123998 us +period=512 irq inject 44646: 974 us +period=512 irq inject 44647: 15607 us +Convert 512 --- > 32: 712: 16560 us +period=32 irq inject 44648: 980 us +... ... +period=32 irq inject 44738: 974 us +Convert 32 --- > 512: 713: 88828 us +period=512 irq inject 44739: 4885 us +Convert 512 --- > 32: 714: 4882 us +period=32 irq inject 44740: 989 us +... ... +period=32 irq inject 44842: 974 us +Convert 32 --- > 512: 715: 100537 us +period=512 irq inject 44843: 8788 us +Convert 512 --- > 32: 716: 8789 us +period=32 irq inject 44844: 972 us +... ... +period=32 irq inject 44941: 979 us +Convert 32 --- > 512: 717: 95677 us +period=512 irq inject 44942: 13661 us +Convert 512 --- > 32: 718: 13657 us +period=32 irq inject 44943: 987 us +... ... +Convert 32 --- > 512: 719: 94690 us +period=512 irq inject 45040: 14643 us +Convert 512 --- > 32: 720: 14642 us +period=32 irq inject 45041: 974 us +... ... +Convert 32 --- > 512: 721: 88848 us +period=512 irq inject 45132: 4892 us +Convert 512 --- > 32: 722: 4931 us +period=32 irq inject 45133: 964 us +... ... +Tue Mar 29 04:39:19 2016 +*irq_num_period_32:835, irq_num_period_512:11 [82], Real time interval is +911520 us + +For windows 7, it has got 835 IRQs which injected during the period of 32, +and got 11 IRQs that injected during the period of 512. it updated the +wall-clock +time with one second, because it supposed it has counted +(835*976.5+11*15625)= 987252.5 us, but the real interval time is 911520 us. + +IMHO, we should calculate the next interrupt time based on the time of last +interrupt injected, and it seems to be more similar with hardware CMOS timer +in this way. +Maybe someone can tell me the reason why we calculated the interrupt timer +in that way, or is it a bug ? ;) + +Thanks, +Hailiang + diff --git a/classification_output/05/graphic/53568181 b/classification_output/05/graphic/53568181 new file mode 100644 index 000000000..31dd76b6d --- /dev/null +++ b/classification_output/05/graphic/53568181 @@ -0,0 +1,86 @@ +semantic: 0.943 +graphic: 0.940 +assembly: 0.936 +device: 0.936 +vnc: 0.935 +instruction: 0.932 +network: 0.925 +other: 0.921 +KVM: 0.917 +boot: 0.876 +socket: 0.875 +mistranslation: 0.854 + +[BUG] x86/PAT handling severely crippled AMD-V SVM KVM performance + +Hi, I maintain an out-of-tree 3D APIs pass-through QEMU device models at +https://github.com/kjliew/qemu-3dfx +that provide 3D acceleration for legacy +32-bit Windows guests (Win98SE, WinME, Win2k and WinXP) with the focus on +playing old legacy games from 1996-2003. It currently supports the now-defunct +3Dfx propriety API called Glide and an alternative OpenGL pass-through based on +MESA implementation. + +The basic concept of both implementations create memory-mapped virtual +interfaces consist of host/guest shared memory with guest-push model instead of +a more common host-pull model for typical QEMU device model implementation. +Guest uses shared memory as FIFOs for drawing commands and data to bulk up the +operations until serialization event that flushes the FIFOs into host. This +achieves extremely good performance since virtual CPUs are fast with hardware +acceleration (Intel VT/AMD-V) and reduces the overhead of frequent VMEXITs to +service the device emulation. Both implementations work on Windows 10 with WHPX +and HAXM accelerators as well as KVM in Linux. + +On Windows 10, QEMU WHPX implementation does not sync MSR_IA32_PAT during +host/guest states sync. There is no visibility into the closed-source WHPX on +how things are managed behind the scene, but from measuring performance figures +I can conclude that it didn't handle the MSR_IA32_PAT correctly for both Intel +and AMD. Call this fair enough, if you will, it didn't flag any concerns, in +fact games such as Quake2 and Quake3 were still within playable frame rate of +40~60FPS on Win2k/XP guest. Until the same games were run on Win98/ME guest and +the frame rate blew off the roof (300~500FPS) on the same CPU and GPU. In fact, +the later seemed to be more inlined with runnng the games bare-metal with vsync +off. + +On Linux (at the time of writing kernel 5.6.7/Mesa 20.0), the difference +prevailed. Intel CPUs (and it so happened that I was on laptop with Intel GPU), +the VMX-based kvm_intel got it right while SVM-based kvm_amd did not. +To put this in simple exaggeration, an aging Core i3-4010U/HD Graphics 4400 +(Haswell GT2) exhibited an insane performance in Quake2/Quake3 timedemos that +totally crushed more recent AMD Ryzen 2500U APU/Vega 8 Graphics and AMD +FX8300/NVIDIA GT730 on desktop. Simply unbelievable! + +It turned out that there was something to do with AMD-V NPT. By loading kvm_amd +with npt=0, AMD Ryzen APU and FX8300 regained a huge performance leap. However, +AMD NPT issue with KVM was supposedly fixed in 2017 kernel commits. NPT=0 would +actually incur performance loss for VM due to intervention required by +hypervisors to maintain the shadow page tables. Finally, I was able to find the +pointer that pointed to MSR_IA32_PAT register. By updating the MSR_IA32_PAT to +0x0606xxxx0606xxxxULL, AMD CPUs now regain their rightful performance without +taking the hit of NPT=0 for Linux KVM. Taking the same solution into Windows, +both Intel and AMD CPUs no longer require Win98/ME guest to unleash the full +performance potentials and performance figures based on games measured on WHPX +were not very far behind Linux KVM. + +So I guess the problem lies in host/guest shared memory regions mapped as +uncacheable from virtual CPU perspective. As virtual CPUs now completely execute +in hardware context with x86 hardware virtualiztion extensions, the cacheability +of memory types would severely impact the performance on guests. WHPX didn't +handle it for both Intel EPT and AMD NPT, but KVM seems to do it right for Intel +EPT. I don't have the correct fix for QEMU. But what I can do for my 3D APIs +pass-through device models is to implement host-side hooks to reprogram and +restore MSR_IA32_PAT upon activation/deactivation of the 3D APIs. Perhaps there +is also a better solution of having the proper kernel drivers for virtual +interfaces to manage the memory types of host/guest shared memory in kernel +space, but to do that and the needs of Microsoft tools/DDKs, I will just forget +it. The guest stubs uses the same kernel drivers included in 3Dfx drivers for +memory mapping and the virtual interfaces remain driver-less from Windows OS +perspective. Considering the current state of halting progress for QEMU native +virgil3D to support Windows OS, I am just being pragmatic. I understand that +QEMU virgil3D will eventually bring 3D acceleration for Windows guests, but I do +not expect anything to support legacy 32-bit Windows OSes which have out-grown +their commercial usefulness. + +Regards, +KJ Liew + diff --git a/classification_output/05/graphic/55961334 b/classification_output/05/graphic/55961334 new file mode 100644 index 000000000..3bd906bbb --- /dev/null +++ b/classification_output/05/graphic/55961334 @@ -0,0 +1,47 @@ +graphic: 0.909 +KVM: 0.881 +instruction: 0.803 +semantic: 0.775 +device: 0.764 +mistranslation: 0.718 +other: 0.715 +vnc: 0.709 +network: 0.697 +socket: 0.636 +boot: 0.569 +assembly: 0.395 + +[Bug] "-ht" flag ignored under KVM - guest still reports HT + +Hi Community, +We have observed that the 'ht' feature bit cannot be disabled when QEMU runs +with KVM acceleration. +qemu-system-x86_64 \ + --enable-kvm \ + -machine q35 \ + -cpu host,-ht \ + -smp 4 \ + -m 4G \ + -drive file=rootfs.img,format=raw \ + -nographic \ + -append 'console=ttyS0 root=/dev/sda rw' +Because '-ht' is specified, the guest should expose no HT capability +(cpuid.1.edx[28] = 0), and /proc/cpuinfo shouldn't show HT feature, but we still +saw ht in linux guest when run 'cat /proc/cpuinfo'. +XiaoYao mentioned that: + +It has been the behavior of QEMU since + + commit 400281af34e5ee6aa9f5496b53d8f82c6fef9319 + Author: Andre Przywara <andre.przywara@amd.com> + Date: Wed Aug 19 15:42:42 2009 +0200 + + set CPUID bits to present cores and threads topology + +that we cannot remove HT CPUID bit from guest via "-cpu xxx,-ht" if the +VM has >= 2 vcpus. +I'd like to know whether there's a plan to address this issue, or if the current +behaviour is considered acceptable. +Best regards, +Ewan. + diff --git a/classification_output/05/graphic/73660729 b/classification_output/05/graphic/73660729 new file mode 100644 index 000000000..37e839f99 --- /dev/null +++ b/classification_output/05/graphic/73660729 @@ -0,0 +1,39 @@ +graphic: 0.858 +instruction: 0.753 +semantic: 0.698 +device: 0.697 +mistranslation: 0.633 +other: 0.620 +network: 0.598 +socket: 0.556 +vnc: 0.467 +assembly: 0.393 +boot: 0.367 +KVM: 0.272 + +[BUG]The latest qemu crashed when I tested cxl + +I test cxl with the patch:[v11,0/2] arm/virt: + CXL support via pxb_cxl. +https://patchwork.kernel.org/project/cxl/cover/20220616141950.23374-1-Jonathan.Cameron@huawei.com/ +But the qemu crashed,and showing an error: +qemu-system-aarch64: ../hw/arm/virt.c:1735: virt_get_high_memmap_enabled: + Assertion `ARRAY_SIZE(extended_memmap) - VIRT_LOWMEMMAP_LAST == ARRAY_SIZE(enabled_array)' failed. +Then I modify the patch to fix the bug: +diff --git a/hw/arm/virt.c b/hw/arm/virt.c +index ea2413a0ba..3d4cee3491 100644 +--- a/hw/arm/virt.c ++++ b/hw/arm/virt.c +@@ -1710,6 +1730,7 @@ static inline bool *virt_get_high_memmap_enabled(VirtMachineState + *vms, +&vms->highmem_redists, +&vms->highmem_ecam, +&vms->highmem_mmio, ++ &vms->cxl_devices_state.is_enabled, +}; +Now qemu works good. +Could you tell me when the patch( +arm/virt: + CXL support via pxb_cxl +) will be merged into upstream? + |