09 Jul 2019
For a long period of time, I’m cultivating the desire of having a habit of writing monthly status updates. Someway, Drew DeVault’s Blog posts and Martin Peres’s advice leverage me towards this direction. So, here I am! I have decided to embrace the challenge of composing a report per month. I hope this new habit helps me to improve my writing and communication skills but most importantly, help me to keep track of my work. I want to start this update by describing my work conditions and then focus on the technical stuff.
In the last two months, I’ve been facing an infrastructure problem to work. I’m dealing with obstacles such as restricted Internet access and long hours in public transportation from my home to my workplace. Unfortunately, I can’t work in my house due to the lack of space, and the best place to work is a public library at the University of Brasilia (UnB). Going to UnB every day makes me waste around 3h per day in a bus. The library has a great environment, but it also has thousands of internet restrictions. The fact that I can’t access websites with ‘.me’ domain and connect to my IRC bouncer is an example of that. In summary: It’s been hard to work these days. So let’s stop talking about non-technical stuff and get into the heart of the matter.
I really like working on VKMS. I know this is not news to anybody, and in June, most of my efforts were dedicated to VKMS. One of my paramount endeavors it was found and fixed a bug in vkms that makes kms_cursor_crc, and kms_pipe_crc_basic fails. I was chasing this bug for a long time as can be seen here [1]. After many hours debugging it, I sent a patch for handling this issue [2], however, after Daniel’s review, I realized that my patch didn’t fix correctly the problem. So Daniel decided to dig into this issue to find the root of the problem and later sent a final fix. If you want to see the solution, take a look at [3]. One day, I want to write a post about this fix since it is an interesting subject to discuss.
Daniel also noticed some concurrency problems in the CRC code and sent a patchset composed of 10 patches that tackle the issue. These patches focused on creating better framebuffers manipulation and avoiding race conditions. It took me around 4 days to take a look and test this series. During my review, I asked many things related to concurrency and other clarification about DRM. Daniel always replied with a very nice and detailed explanation. If you want to learn a little bit more about locks, I recommend you to take a look at [4]. Seriously, it is really nice!
I also worked for adding the writeback support in vkms; since XDC2018 I could not stop to think about the idea of adding writeback connector in vkms due to the benefits it could bring, such as new test and assist developers with visual output. As a result, I started some clumsy attempts to implement it in January; but I really dove in this issue in the middle of April, and in June I was focused on making it work. It was tough for me to implement these features due to the following reasons:
- There is not i-g-t test for writeback in the main repository, I had to use a WIP patchset made by Brian and Liviu.
- I was not familiar with framebuffer, connectors, and fancy manipulation.
As a result of the above limitations, I had to invest many hours reading the documentation and the DRM/IGT code. In the end, I think that add writeback connectors paid well for me since I feel much more comfortable with many things related to DRM these days. The writeback support was not landed yet, however, at this moment the patch is under review (V3) and changed a lot since the first version; for details about this series take a look at [5]. I’ll write a post about this feature after it gets merged.
After having the writeback connectors working in vkms, I felt so grateful for
Brian, Liviu, and Daniel for all the assistance they provided to me. In
particular, I was thrilled that Brian and Liviu made kms_writeback test which
worked as an implementation guideline for me. As a result, I updated their
patchsets for making it work in the latest version of IGT and made some tiny
fixes. My goal was helping them to upstream kms_writeback. I submitted the
series with the hope to see it landed in the IGT [9].
Parallel to my work with ‘writeback’ I was trying to figure out how I could expose vkms configurations to the userspace via configfs. After many efforts, I submitted the first version of configfs support; in this patchset I exposed the virtual and writeback connectors. Take a look at [6] for more information about this feature, and definitely, I’ll write a post about this feature after it gets landed.
Finally, I’m still trying to upstream a patch that makes
drm_wait_vblank_ioctl return EOPNOTSUPP instead of EINVAL if the driver
does not support vblank get landed. Since this change is in the DRM core and
also change the userspace, it is not easy to make this patch get landed. For
the details about this patch, you can take a look here [7]. I also implemented some changes in the kms_flip to validate
the changes that I made in the function drm_wait_vblank_ioctl and it got
landed [8].
July Aims
In June, I was totally dedicated to vkms, now I want to slow my road a little bit and study more about userspace. I want to take a step back and make some tiny programs using libdrm with the goal of understanding the interaction among userspace and kernel space. I also want to take a look at the theoretical part related to computer graphics.
I want to put some effort to improve a tool named kw that help me during my work with Linux Kernel. I also want to take a look at real overlay planes support in vkms. I noted that I have to find a “contribution protocol” (review/write code) that works for me in my current work conditions; otherwise, work will become painful for my relatives and me. Finally, and most importantly, I want to take some days off to enjoy my family.
Info: If you find any problem with this text, please let me know. I will be glad to fix it.
References
[1] “First discussion in the Shayenne’s patch about the CRC problem”. URL: https://lkml.org/lkml/2019/3/10/197. ⤴
[2] “Patch fix for the CRC issue”. URL: https://patchwork.freedesktop.org/patch/308617/. ⤴
[3] “Daniel final fix for CRC”. URL: https://patchwork.freedesktop.org/patch/308881/?series=61703&rev=1. ⤴
[4] “Rework crc worker”. URL: https://patchwork.freedesktop.org/series/61737/. ⤴
[5] “Introduces writeback support”. URL: https://patchwork.freedesktop.org/series/61738/. ⤴
[6] “Introduce basic support for configfs”. URL: https://patchwork.freedesktop.org/series/63010/. ⤴
[7] “Change EINVAL by EOPNOTSUPP when vblank is not supported”. URL: https://patchwork.freedesktop.org/patch/314399/?series=50697&rev=7. ⤴
[8] “Skip VBlank tests in modules without VBlank”. URL: https://gitlab.freedesktop.org/drm/igt-gpu-tools/commit/2d244aed69165753f3adbbd6468db073dc1acf9a. ⤴
[9] “Add support for testing writeback connectors”. URL: https://patchwork.freedesktop.org/series/39229/. ⤴
