7:48 p.m.
Hi,
I was following the instructions for running a KVM image and found out the
graphic environment does not work as documented.
I followed the instructions located at https://clearlinux.org/
documentation/clear-linux/get-started/virtual-machine-install/kvm.html
and there the configuration is changed to use qxl driver, however the X log
file complains about the driver not present.
[ 170.356] (II) LoadModule: "qxl"
[ 170.356] (WW) Warning, couldn't open module qxl
[ 170.356] (II) UnloadModule: "qxl"
[ 170.356] (II) Unloading qxl
[ 170.356] (EE) Failed to load module "qxl" (module does not exist, 0)
in the X drivers directlry there is no qxl.so module, the only things
present are:
# ls /usr/lib64/xorg/modules/drivers/
amdgpu_drv.so fbdev_drv.so nouveau_drv.so vesa_drv.so
ati_drv.so modesetting_drv.so radeon_drv.so
I tried searching for the file using "swupd search qxl" but there is no
package (bundle?) containing the required .so file.
Is there something else needed to be done?
Thanks!
10:04 p.m.
On 11/5/2017 11:48 AM, Ignacio Hernandez wrote:
Hi,
I was following the instructions for running a KVM image and found out the
graphic environment does not work as documented.
I followed the instructions located at https://clearlinux.org/
documentation/clear-linux/get-started/virtual-machine-install/kvm.html
and there the configuration is changed to use qxl driver, however the X log
file complains about the driver not present.
[ 170.356] (II) LoadModule: "qxl"
[ 170.356] (WW) Warning, couldn't open module qxl
[ 170.356] (II) UnloadModule: "qxl"
[ 170.356] (II) Unloading qxl
[ 170.356] (EE) Failed to load module "qxl" (module does not exist, 0)
in the X drivers directlry there is no qxl.so module, the only things
present are:
# ls /usr/lib64/xorg/modules/drivers/
amdgpu_drv.so fbdev_drv.so nouveau_drv.so vesa_drv.so
ati_drv.so modesetting_drv.so radeon_drv.so
the modesetting driver is supposed to be used for this....
(but also I'm surprised by qxl; I also was expecting to see vesa)
10:19 p.m.
Looking a bit longer, eventually the server crashes, and segfaults because
of an illegal instruction. That happens after giving up on qxl and trying
vesa and fbdev.
I ran the server under gdb and got to this instruction sequence.
0x00007f5f63c9a5af: test %eax,%eax
0x00007f5f63c9a5b1: je 0x7f5f63c9a5cd
=> 0x00007f5f63c9a5b3: vpxor %xmm0,%xmm0,%xmm0
0x00007f5f63c9a5b7: movq $0x0,0x20(%rax)
0x00007f5f63c9a5bf: add $0x30,%rax
0x00007f5f63c9a5c3: vmovaps %xmm0,-0x30(%rax)
0x00007f5f63c9a5c8: vmovaps %xmm0,-0x20(%rax)
0x00007f5f63c9a5cd: mov %rax,0xd96c94(%rip) # 0x7f5f64a31268
Program received signal SIGILL, Illegal instruction.
0x00007f5f63c9a5b3 in ?? () from /usr/lib64/dri/kms_swrast_dri.so
(gdb) disassemble
No function contains program counter for selected frame.
(gdb) bt
#0 0x00007f5f63c9a5b3 in ?? () from /usr/lib64/dri/kms_swrast_dri.so
#1 0x00007f5f63c9a6ea in ?? () from /usr/lib64/dri/kms_swrast_dri.so
#2 0x00007f5f63c9a726 in ?? () from /usr/lib64/dri/kms_swrast_dri.so
#3 0x00007f5f63c9d3aa in ?? () from /usr/lib64/dri/kms_swrast_dri.so
#4 0x00007f5f6682e86a in call_init (l=<optimized out>, argc=argc@entry=1,
argv=argv@entry=0x7fff4f8e77d8, env=env@entry=0x7fff4f8e77e8)
The faulting instruction is an AVX one, so either, the documentation
listing see4_1 as requirement is wrong or the build system is producing
binaries for newer targets.
Thanks for looking into this.
On Sun, Nov 5, 2017 at 2:04 PM, Arjan van de Ven <arjan(a)linux.intel.com>
wrote:
On 11/5/2017 11:48 AM, Ignacio Hernandez wrote:
> Hi,
>
> I was following the instructions for running a KVM image and found out the
> graphic environment does not work as documented.
>
> I followed the instructions located at https://clearlinux.org/
> documentation/clear-linux/get-started/virtual-machine-install/kvm.html
>
> and there the configuration is changed to use qxl driver, however the X
> log
> file complains about the driver not present.
>
> [ 170.356] (II) LoadModule: "qxl"
> [ 170.356] (WW) Warning, couldn't open module qxl
> [ 170.356] (II) UnloadModule: "qxl"
> [ 170.356] (II) Unloading qxl
> [ 170.356] (EE) Failed to load module "qxl" (module does not exist, 0)
>
> in the X drivers directlry there is no qxl.so module, the only things
> present are:
>
> # ls /usr/lib64/xorg/modules/drivers/
> amdgpu_drv.so fbdev_drv.so nouveau_drv.so vesa_drv.so
> ati_drv.so modesetting_drv.so radeon_drv.so
>
>
the modesetting driver is supposed to be used for this....
(but also I'm surprised by qxl; I also was expecting to see vesa)
10:33 p.m.
On 11/5/2017 2:19 PM, Ignacio Hernandez wrote:
Looking a bit longer, eventually the server crashes, and segfaults
because of an illegal instruction. That happens after giving up on qxl and trying vesa and
fbdev.
I ran the server under gdb and got to this instruction sequence.
0x00007f5f63c9a5af: test %eax,%eax
0x00007f5f63c9a5b1: je 0x7f5f63c9a5cd
=> 0x00007f5f63c9a5b3: vpxor %xmm0,%xmm0,%xmm0
0x00007f5f63c9a5b7: movq $0x0,0x20(%rax)
0x00007f5f63c9a5bf: add $0x30,%rax
0x00007f5f63c9a5c3: vmovaps %xmm0,-0x30(%rax)
0x00007f5f63c9a5c8: vmovaps %xmm0,-0x20(%rax)
0x00007f5f63c9a5cd: mov %rax,0xd96c94(%rip) # 0x7f5f64a31268
Program received signal SIGILL, Illegal instruction.
0x00007f5f63c9a5b3 in ?? () from /usr/lib64/dri/kms_swrast_dri.so
(gdb) disassemble
No function contains program counter for selected frame.
(gdb) bt
#0 0x00007f5f63c9a5b3 in ?? () from /usr/lib64/dri/kms_swrast_dri.so
#1 0x00007f5f63c9a6ea in ?? () from /usr/lib64/dri/kms_swrast_dri.so
#2 0x00007f5f63c9a726 in ?? () from /usr/lib64/dri/kms_swrast_dri.so
#3 0x00007f5f63c9d3aa in ?? () from /usr/lib64/dri/kms_swrast_dri.so
#4 0x00007f5f6682e86a in call_init (l=<optimized out>, argc=argc@entry=1,
argv=argv@entry=0x7fff4f8e77d8, env=env@entry=0x7fff4f8e77e8)
The faulting instruction is an AVX one, so either, the documentation listing see4_1 as
requirement is wrong or the build system is producing binaries for newer targets.
I am 100% sure this is pointing at a miscompile; the avx(2) version of swrast lives in a
different file in the filesystem.
4:54 p.m.
On 11/5/2017 2:19 PM, Ignacio Hernandez wrote:
Looking a bit longer, eventually the server crashes, and segfaults
because of an illegal instruction. That happens after giving up on qxl and trying vesa and
fbdev.
I ran the server under gdb and got to this instruction sequence.
0x00007f5f63c9a5af: test %eax,%eax
0x00007f5f63c9a5b1: je 0x7f5f63c9a5cd
=> 0x00007f5f63c9a5b3: vpxor %xmm0,%xmm0,%xmm0
0x00007f5f63c9a5b7: movq $0x0,0x20(%rax)
0x00007f5f63c9a5bf: add $0x30,%rax
0x00007f5f63c9a5c3: vmovaps %xmm0,-0x30(%rax)
0x00007f5f63c9a5c8: vmovaps %xmm0,-0x20(%rax)
0x00007f5f63c9a5cd: mov %rax,0xd96c94(%rip) # 0x7f5f64a31268
I found the cause of this and just fixed it
thanks for the report!
1739
days inactive
1745
days old
4 comments
2 participants
participants (2)
-
Arjan van de Ven -
Ignacio Hernandez