Re[2]: [Xen-devel] PV DRM doesn't work without auto_translated_physmap feature in Dom0
Sure. I will make it a bit later, when I have the more stable result. Best regards, Alexander >Понедельник, 20 апреля 2020, 8:59 +03:00 от Oleksandr Andrushchenko >: > >Hi, > >On 4/19/20 14:26, Santucco wrote: >> Hello, >> I have found a source of the problem. >> In displ_be, BaseDump copies to the drm buffer with a size from >> i915 drm driver, but this size a bit more than a size of my frontend >> display buffer. I have made a quick and dirty fix, a copy a line of my >> display buffer to a middle of a line of the drm display buffer. Patch >> is attached. > >Thank you for the patch and your efforts to fix the issue. > >Could you please make a pull request to [1], so we can continue there? > >Thank you in advance, > >Oleksandr > >> Best regards, >> Alexander >> >> Четверг, 6 февраля 2020, 11:20 +03:00 от Oleksandr Andrushchenko >> < oleksandr_andrushche...@epam.com >: >> On 2/5/20 8:59 PM, Santucco wrote: >> > Hello, >> > Ok, I commented out the memcpy call and run the test. >> > displ_be hasn’t crached, I have seen FLIP events in the log. >> > But there hasn’t been the black screen, just a blink effect every >> > couple of seconds. >> > Logs are attached. >> Ok, so I believe that frontend - backend (displ_be) communication >> is ok >> and there is nothing to do there. >> >> Next, I would start debugging the following in Xen: >> (XEN) mm.c:2223:d2v0 Bad L1 flags 80 >> and have a look at [1]. Probably, someone on Xen x86 side can tell >> if this could be related to the flags at [2]. >> >> > Best regards, >> > Alexander >> > >> > Среда, 5 февраля 2020, 9:31 +03:00 от Oleksandr Andrushchenko >> > < oleksandr_andrushche...@epam.com >> >: >> > On 2/4/20 10:28 AM, Santucco wrote: >> > > Hello, >> > > displ_be was compiled without zero-copy support early. >> > > I have tried with the recompiled dom0 kernel, a result is the >> same. >> > > Logs and configs (+displ_be’s CMakeCache.txt ) are attached. >> > Ok, yet another test to localize the problem. >> > Could you please remove memcpy from >> > #1 0x55e5a1f28bec in Drm::DumbDrm::copy >> (this=0x7f9338000e00) at >> > >> /home/santucco/tmp/xen-troops/displ_be/src/displayBackend/drm/Dumb.cpp:149 >> > and just memset the destination with 0 or whatever. >> > >> > I expect that system won't crash, nothing will be shown (black >> > screen), but >> > displ_be will show page flip events in its logs. >> > > Best regards, >> > > Alexander >> > > >> > > Понедельник, 3 февраля 2020, 10:36 +03:00 от Oleksandr >> > > Andrushchenko < oleksandr_andrushche...@epam.com >> >> > >: >> > > >> > > >> > > On 2/1/20 4:39 PM, Santucco wrote: >> > > > Hello again, >> > > > I have not yet made to work my drm client, so I have tried >> to run >> > > > linux like a domU (to see how it should work), it doesn’t >> work too >> > > > — displ_be catches SIGSEGV: >> > > > >> > > > #0 0x7f4afed1c161 in ?? () from /lib64/libc.so.6 >> > > > #1 0x55723b9c5bec in Drm::DumbDrm::copy >> > > (this=0x7f4adc000e00) at >> > > > >> > > >> > >> /home/santucco/tmp/xen-troops/displ_be/src/displayBackend/drm/Dumb.cpp:149 >> > > > #2 0x55723b9a8f51 in BuffersStorage::getFrameBufferAndCopy >> > > > (this=0x7f4ae00010e0, fbCookie=18446612682295083264) at >> > > > >> > > >> > >> /home/santucco/tmp/xen-troops/displ_be/src/displayBackend/BuffersStorage.cpp:165 >> > > > It tries to copy to mBuffer with non-accessible address. >> > > > For the moment I see a strange offset for mmap call of >> > > /dev/drm/card0 >> > > > in the strace log — 0x1. Is that normal? >> > > > Any direction of which to dig will be very helpful. >> > > > Configuration details: >> > > > Xen 4.12.1 Dom0: Linux 4.20.17-gentoo #13 SMP Sat Dec 28 >> > > 11:12:24 MSK >> > > > 2019 x86_64 Intel(R) Celeron(R) CPU N3050 @ 1.60GHz GenuineIntel >> > > GNU/Linux >> > > > DomU: Linux 4.20.17-gentoo >> > > > last xen-troops/libxenbe and xen-troops/displ_be >> > > > Logs (dmesg, xl dmesg, displ_be, strace log of displ_be), a >> > &
Re: [Xen-devel] PV DRM doesn't work without auto_translated_physmap feature in Dom0
Ok, thank you very much. One more question - is that normal the /dev/drm/card0 device is mapped at offset more then 4Gb? Best regards, Alexander >Четверг, 6 февраля 2020, 11:20 +03:00 от Oleksandr Andrushchenko >: > >On 2/5/20 8:59 PM, Santucco wrote: >> Hello, >> Ok, I commented out the memcpy call and run the test. >> displ_be hasn’t crached, I have seen FLIP events in the log. >> But there hasn’t been the black screen, just a blink effect every >> couple of seconds. >> Logs are attached. >Ok, so I believe that frontend - backend (displ_be) communication is ok >and there is nothing to do there. > >Next, I would start debugging the following in Xen: >(XEN) mm.c:2223:d2v0 Bad L1 flags 80 >and have a look at [1]. Probably, someone on Xen x86 side can tell >if this could be related to the flags at [2]. > >> Best regards, >> Alexander >> >> Среда, 5 февраля 2020, 9:31 +03:00 от Oleksandr Andrushchenko >> < oleksandr_andrushche...@epam.com >: >> On 2/4/20 10:28 AM, Santucco wrote: >> > Hello, >> > displ_be was compiled without zero-copy support early. >> > I have tried with the recompiled dom0 kernel, a result is the same. >> > Logs and configs (+displ_be’s CMakeCache.txt ) are attached. >> Ok, yet another test to localize the problem. >> Could you please remove memcpy from >> #1 0x55e5a1f28bec in Drm::DumbDrm::copy (this=0x7f9338000e00) at >> /home/santucco/tmp/xen-troops/displ_be/src/displayBackend/drm/Dumb.cpp:149 >> and just memset the destination with 0 or whatever. >> >> I expect that system won't crash, nothing will be shown (black >> screen), but >> displ_be will show page flip events in its logs. >> > Best regards, >> > Alexander >> > >> > Понедельник, 3 февраля 2020, 10:36 +03:00 от Oleksandr >> > Andrushchenko < oleksandr_andrushche...@epam.com >> >: >> > >> > >> > On 2/1/20 4:39 PM, Santucco wrote: >> > > Hello again, >> > > I have not yet made to work my drm client, so I have tried to run >> > > linux like a domU (to see how it should work), it doesn’t work too >> > > — displ_be catches SIGSEGV: >> > > >> > > #0 0x7f4afed1c161 in ?? () from /lib64/libc.so.6 >> > > #1 0x55723b9c5bec in Drm::DumbDrm::copy >> > (this=0x7f4adc000e00) at >> > > >> > >> /home/santucco/tmp/xen-troops/displ_be/src/displayBackend/drm/Dumb.cpp:149 >> > > #2 0x55723b9a8f51 in BuffersStorage::getFrameBufferAndCopy >> > > (this=0x7f4ae00010e0, fbCookie=18446612682295083264) at >> > > >> > >> /home/santucco/tmp/xen-troops/displ_be/src/displayBackend/BuffersStorage.cpp:165 >> > > It tries to copy to mBuffer with non-accessible address. >> > > For the moment I see a strange offset for mmap call of >> > /dev/drm/card0 >> > > in the strace log — 0x1. Is that normal? >> > > Any direction of which to dig will be very helpful. >> > > Configuration details: >> > > Xen 4.12.1 Dom0: Linux 4.20.17-gentoo #13 SMP Sat Dec 28 >> > 11:12:24 MSK >> > > 2019 x86_64 Intel(R) Celeron(R) CPU N3050 @ 1.60GHz GenuineIntel >> > GNU/Linux >> > > DomU: Linux 4.20.17-gentoo >> > > last xen-troops/libxenbe and xen-troops/displ_be >> > > Logs (dmesg, xl dmesg, displ_be, strace log of displ_be), a >> > backtrace >> > > of gdb and kernel configs are attached. >> > > Thanks in advance. >> > Could you please try Dom0 kernel WITHOUT the options below: >> > CONFIG_XEN_GNTDEV_DMABUF=y >> > CONFIG_XEN_GRANT_DMA_ALLOC=y >> > >> > Then, just to make sure, did you build displ_be without zero-copy >> > support? >> > >> > > On 1/8/20 5:38 PM, Santucco wrote: >> > > > Thank you very much for all your answers. >> > > > >> > > > Среда, 8 января 2020, 10:54 +03:00 от Oleksandr Andrushchenko >> > > > < oleksandr_andrushche...@epam.com >> >> > > > >: >> > > > On 1/6/20 10:38 AM, Jürgen Groß wrote: >> > > > > On 06.01.20 08:56, Santucco wrote: >> > > > >> Hello, >> > > > >> >> > > > >> I’m trying to use vdispl interface from PV OS, it doesn’t >> work. >> > > > >> Configuration details: >> > > > >> Xen 4.12.1 >> > > > >> Dom0: Linux 4.20.17-gentoo #13 SMP Sat Dec 28 >> 11:1
Re: [Xen-devel] PV DRM doesn't work without auto_translated_physmap feature in Dom0
Hello, displ_be was compiled without zero-copy support early. I have tried with the recompiled dom0 kernel, a result is the same. Logs and configs (+displ_be’s CMakeCache.txt ) are attached. Best regards, Alexander >Понедельник, 3 февраля 2020, 10:36 +03:00 от Oleksandr Andrushchenko < >oleksandr_andrushche...@epam.com >: > > > >On 2/1/20 4:39 PM, Santucco wrote: >> Hello again, >> I have not yet made to work my drm client, so I have tried to run >> linux like a domU (to see how it should work), it doesn’t work too >> — displ_be catches SIGSEGV: >> >> #0 0x7f4afed1c161 in ?? () from /lib64/libc.so.6 >> #1 0x55723b9c5bec in Drm::DumbDrm::copy (this=0x7f4adc000e00) at >> /home/santucco/tmp/xen-troops/displ_be/src/displayBackend/drm/Dumb.cpp:149 >> #2 0x55723b9a8f51 in BuffersStorage::getFrameBufferAndCopy >> (this=0x7f4ae00010e0, fbCookie=18446612682295083264) at >> /home/santucco/tmp/xen-troops/displ_be/src/displayBackend/BuffersStorage.cpp:165 >> It tries to copy to mBuffer with non-accessible address. >> For the moment I see a strange offset for mmap call of /dev/drm/card0 >> in the strace log — 0x1. Is that normal? >> Any direction of which to dig will be very helpful. >> Configuration details: >> Xen 4.12.1 Dom0: Linux 4.20.17-gentoo #13 SMP Sat Dec 28 11:12:24 MSK >> 2019 x86_64 Intel(R) Celeron(R) CPU N3050 @ 1.60GHz GenuineIntel GNU/Linux >> DomU: Linux 4.20.17-gentoo >> last xen-troops/libxenbe and xen-troops/displ_be >> Logs (dmesg, xl dmesg, displ_be, strace log of displ_be), a backtrace >> of gdb and kernel configs are attached. >> Thanks in advance. >Could you please try Dom0 kernel WITHOUT the options below: >CONFIG_XEN_GNTDEV_DMABUF=y >CONFIG_XEN_GRANT_DMA_ALLOC=y > >Then, just to make sure, did you build displ_be without zero-copy support? > >> On 1/8/20 5:38 PM, Santucco wrote: >> > Thank you very much for all your answers. >> > >> > Среда, 8 января 2020, 10:54 +03:00 от Oleksandr Andrushchenko >> > < oleksandr_andrushche...@epam.com >> > < /compose?To=oleksandr_andrushche...@epam.com >>: >> > On 1/6/20 10:38 AM, Jürgen Groß wrote: >> > > On 06.01.20 08:56, Santucco wrote: >> > >> Hello, >> > >> >> > >> I’m trying to use vdispl interface from PV OS, it doesn’t work. >> > >> Configuration details: >> > >> Xen 4.12.1 >> > >> Dom0: Linux 4.20.17-gentoo #13 SMP Sat Dec 28 11:12:24 MSK >> > 2019 >> > >> x86_64 Intel(R) Celeron(R) CPU N3050 @ 1.60GHz GenuineIntel >> > GNU/Linux >> > >> DomU: x86 Plan9, PV >> > >> displ_be as a backend for vdispl and vkb >> > >> >> > >> when VM starts, displ_be reports about an error: >> > >> gnttab: error: ioctl DMABUF_EXP_FROM_REFS failed: Invalid argument >> > >> (displ_be.log:221) >> > >> >> > >> related Dom0 output is: >> > >> [ 191.579278] Cannot provide dma-buf: use_ptemode 1 >> > >> (dmesg.create.log:123) >> > > >> > > This seems to be a limitation of the xen dma-buf driver. It was >> > written >> > > for being used on ARM initially where PV is not available. >> > This is true and we never tried/targeted PV domains with this >> > implementation, >> > so if there is a need for that someone has to take a look on the >> > proper >> > implementation for PV… >> > >> > Have I got your right and there is no the proper implementation :-)? >> There is no >> > >> > > >> > > CC-ing Oleksandr Andrushchenko who is the author of that driver. He >> > > should be able to tell us what would be needed to enable PV dom0. >> > > >> > > Depending on your use case it might be possible to use PVH dom0, but >> > > support for this mode is "experimental" only and some features >> > are not >> > > yet working. >> > > >> > Well, one of the workarounds possible is to drop zero-copying use-case >> > (this is why display backend tries to create dmu-bufs from grants >> > passed >> > by the guest domain and fails because of "Cannot provide dma-buf: >> > use_ptemode 1") >> > So, in this case display backend will do memory copying for the >> > incoming >> > frames >> > and won't touch DMABUF_EXP_FROM_REFS ioctl. >> > To do so just disable zero-copying while buildi
Re: [Xen-devel] PV DRM doesn't work without auto_translated_physmap feature in Dom0
Hello again, I have not yet made to work my drm client, so I have tried to run linux like a domU (to see how it should work), it doesn’t work too — displ_be catches SIGSEGV: #0 0x7f4afed1c161 in ?? () from /lib64/libc.so.6 #1 0x55723b9c5bec in Drm::DumbDrm::copy (this=0x7f4adc000e00) at /home/santucco/tmp/xen-troops/displ_be/src/displayBackend/drm/Dumb.cpp:149 #2 0x55723b9a8f51 in BuffersStorage::getFrameBufferAndCopy (this=0x7f4ae00010e0, fbCookie=18446612682295083264) at /home/santucco/tmp/xen-troops/displ_be/src/displayBackend/BuffersStorage.cpp:165 It tries to copy to mBuffer with non-accessible address. For the moment I see a strange offset for mmap call of /dev/drm/card0 in the strace log — 0x1. Is that normal? Any direction of which to dig will be very helpful. Configuration details: Xen 4.12.1 Dom0: Linux 4.20.17-gentoo #13 SMP Sat Dec 28 11:12:24 MSK 2019 x86_64 Intel(R) Celeron(R) CPU N3050 @ 1.60GHz GenuineIntel GNU/Linux DomU: Linux 4.20.17-gentoo last xen-troops/libxenbe and xen-troops/displ_be Logs (dmesg, xl dmesg, displ_be, strace log of displ_be), a backtrace of gdb and kernel configs are attached. Thanks in advance. On 1/8/20 5:38 PM, Santucco wrote: > Thank you very much for all your answers. > > Среда, 8 января 2020, 10:54 +03:00 от Oleksandr Andrushchenko > < oleksandr_andrushche...@epam.com > < /compose?To=oleksandr_andrushche...@epam.com >>: > On 1/6/20 10:38 AM, Jürgen Groß wrote: > > On 06.01.20 08:56, Santucco wrote: > >> Hello, > >> > >> I’m trying to use vdispl interface from PV OS, it doesn’t work. > >> Configuration details: > >> Xen 4.12.1 > >> Dom0: Linux 4.20.17-gentoo #13 SMP Sat Dec 28 11:12:24 MSK > 2019 > >> x86_64 Intel(R) Celeron(R) CPU N3050 @ 1.60GHz GenuineIntel > GNU/Linux > >> DomU: x86 Plan9, PV > >> displ_be as a backend for vdispl and vkb > >> > >> when VM starts, displ_be reports about an error: > >> gnttab: error: ioctl DMABUF_EXP_FROM_REFS failed: Invalid argument > >> (displ_be.log:221) > >> > >> related Dom0 output is: > >> [ 191.579278] Cannot provide dma-buf: use_ptemode 1 > >> (dmesg.create.log:123) > > > > This seems to be a limitation of the xen dma-buf driver. It was > written > > for being used on ARM initially where PV is not available. > This is true and we never tried/targeted PV domains with this > implementation, > so if there is a need for that someone has to take a look on the > proper > implementation for PV… > > Have I got your right and there is no the proper implementation :-)? There is no > > > > > CC-ing Oleksandr Andrushchenko who is the author of that driver. He > > should be able to tell us what would be needed to enable PV dom0. > > > > Depending on your use case it might be possible to use PVH dom0, but > > support for this mode is "experimental" only and some features > are not > > yet working. > > > Well, one of the workarounds possible is to drop zero-copying use-case > (this is why display backend tries to create dmu-bufs from grants > passed > by the guest domain and fails because of "Cannot provide dma-buf: > use_ptemode 1") > So, in this case display backend will do memory copying for the > incoming > frames > and won't touch DMABUF_EXP_FROM_REFS ioctl. > To do so just disable zero-copying while building the backend [1] > > Thanks, I have just tried the workaround. The backend has failed > in an other place not corresponding with dma_buf. > Anyway it is enough to continue debugging my frontend implementation. > Do you know how big is performance penalty in comparison with > the zero-copy variant? Well, it solely depends on your setup, so I cannot tell what would the numbers be in your case. Comparing to what I have doesn't make any sense to me: one should compare apples to apples > Does it make a sense if I make a dedicated HVM domain with linux only > for the purpose of vdispl and vkbd backends? Is there a hope this > approach will work? You can try if this approach fits your design and requirements > > > > > Juergen > > > [1] > https://github.com/xen-troops/displ_be/blob/master/CMakeLists.txt#L12 > < > https://urldefense.com/v3/__https://github.com/xen-troops/displ_be/blob/master/CMakeLists.txt*L12__;Iw!!GF_29dbcQIUBPA!mz3gn1wQMX2DXeNuAV-1_dI7nxFYYZOgdPiJNSFMesCz9lAzOKlwVPlddbxbcLmUO44NOy0TFA$ > > > > Best regards, > Alexander Sychev logs.tgz Description: GNU Zip compressed data ___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] PV DRM doesn't work without auto_translated_physmap feature in Dom0
Thank you very much for all your answers. >Среда, 8 января 2020, 10:54 +03:00 от Oleksandr Andrushchenko < >oleksandr_andrushche...@epam.com >: > >On 1/6/20 10:38 AM, Jürgen Groß wrote: >> On 06.01.20 08:56, Santucco wrote: >>> Hello, >>> >>> I’m trying to use vdispl interface from PV OS, it doesn’t work. >>> Configuration details: >>> Xen 4.12.1 >>> Dom0: Linux 4.20.17-gentoo #13 SMP Sat Dec 28 11:12:24 MSK 2019 >>> x86_64 Intel(R) Celeron(R) CPU N3050 @ 1.60GHz GenuineIntel GNU/Linux >>> DomU: x86 Plan9, PV >>> displ_be as a backend for vdispl and vkb >>> >>> when VM starts, displ_be reports about an error: >>> gnttab: error: ioctl DMABUF_EXP_FROM_REFS failed: Invalid argument >>> (displ_be.log:221) >>> >>> related Dom0 output is: >>> [ 191.579278] Cannot provide dma-buf: use_ptemode 1 >>> (dmesg.create.log:123) >> >> This seems to be a limitation of the xen dma-buf driver. It was written >> for being used on ARM initially where PV is not available. >This is true and we never tried/targeted PV domains with this >implementation, >so if there is a need for that someone has to take a look on the proper >implementation for PV… Have I got your right and there is no the proper implementation :-)? >> >> CC-ing Oleksandr Andrushchenko who is the author of that driver. He >> should be able to tell us what would be needed to enable PV dom0. >> >> Depending on your use case it might be possible to use PVH dom0, but >> support for this mode is "experimental" only and some features are not >> yet working. >> >Well, one of the workarounds possible is to drop zero-copying use-case >(this is why display backend tries to create dmu-bufs from grants passed >by the guest domain and fails because of "Cannot provide dma-buf: >use_ptemode 1") >So, in this case display backend will do memory copying for the incoming >frames >and won't touch DMABUF_EXP_FROM_REFS ioctl. >To do so just disable zero-copying while building the backend [1] Thanks, I have just tried the workaround. The backend has failed in an other place not corresponding with dma_buf. Anyway it is enough to continue debugging my frontend implementation. Do you know how big is performance penalty in comparison with the zero-copy variant? Does it make a sense if I make a dedicated HVM domain with linux only for the purpose of vdispl and vkbd backends? Is there a hope this approach will work? >> >> Juergen >> >[1] https://github.com/xen-troops/displ_be/blob/master/CMakeLists.txt#L12 Best regards, Alexander Sychev___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel
[Xen-devel] PV DRM doesn't work without auto_translated_physmap feature in Dom0
Hello, I’m trying to use vdispl interface from PV OS, it doesn’t work. Configuration details: Xen 4.12.1 Dom0: Linux 4.20.17-gentoo #13 SMP Sat Dec 28 11:12:24 MSK 2019 x86_64 Intel(R) Celeron(R) CPU N3050 @ 1.60GHz GenuineIntel GNU/Linux DomU: x86 Plan9, PV displ_be as a backend for vdispl and vkb when VM starts, displ_be reports about an error: gnttab: error: ioctl DMABUF_EXP_FROM_REFS failed: Invalid argument (displ_be.log:221) related Dom0 output is: [ 191.579278] Cannot provide dma-buf: use_ptemode 1 (dmesg.create.log:123) According to the linux kernel sources use_ptemod is: gntdev.c:1213: use_ptemod = !xen_feature(XENFEAT_auto_translated_physmap); But xen-features.pl script doesn't show this feature is set for Dom0: Xen features: 28f0 enabled: pae_pgdir_above_4gb enabled: mmu_pt_update_preserve_ad enabled: dom0 enabled: memory_op_vnode_supported According to the Xen sources, auto_translated_physmap is set up in a case of : xen/common/kernel.c:467:if ( paging_mode_translate(d) ) xen/common/kernel.c-468-fi.submap |= xen/common/kernel.c-469-(1U << XENFEAT_writable_page_tables) | xen/common/kernel.c-470-(1U << XENFEAT_auto_translated_physmap); where xen/include/asm-x86/paging.h:73:#define paging_mode_translate(_d) (!!((_d)->arch.paging.mode & PG_translate)) But PG_translate is applied only for HVM, at least I have not fount any other points: xen/arch/x86/hvm/hvm.c:606:rc = paging_enable(d, PG_refcounts|PG_translate|PG_external); What configuration should be for vdispl is able to work? Or maybe I can to something to make Xen set auto_translated_physmap feature up to continue using of vdispl? And should I make any special support for the feature in PV OS? All logs and configs are attached. Thanks in advance. Best regards, Alexander Sychev logs.tar.gz Description: GNU Zip compressed data ___ Xen-devel mailing list Xen-devel@lists.xenproject.org https://lists.xenproject.org/mailman/listinfo/xen-devel