sj
Posts
241Following
41Followers
118
sj
sj
Today was my last day at Meta. It was great to be connected and work with the awesome and nice people. I believe nothing is really being ended though, since we will keep being connected and work together on upstream open source communities.
1
0
6
@gregkh Inspired by the results, I bought a ~$320 mini PC from Aamzon and ran kcbench.
```
Processor: AMD Ryzen 7 6800H with Radeon Graphics [16 threads]
Cpufreq; Memory: powersave [amd-pstate-epp]; 27841 MiB
Linux running: 6.12.48+deb13-amd64 [x86_64]
Compiler: gcc (Debian 14.2.0-19) 14.2.0
Linux compiled: 6.17.0 [.../.cache/kcbench/linux-6.17]
Config; Environment: defconfig; CCACHE_DISABLE="1"
Build command: make vmlinux
Filling caches: This might take a while... Done
Run 1 (-j 16): 161.38 seconds / 22.31 kernels/hour [P:1440%, 134 maj. pagefaults]
Run 2 (-j 16): 162.53 seconds / 22.15 kernels/hour [P:1441%, 140 maj. pagefaults]
Run 3 (-j 19): 172.87 seconds / 20.82 kernels/hour [P:1366%, 266 maj. pagefaults]
Run 4 (-j 19): 164.76 seconds / 21.85 kernels/hour [P:1446%, 258 maj. pagefaults]
Run 5 (-j 8): 190.83 seconds / 18.86 kernels/hour [P:742%, 49 maj. pagefaults]
Run 6 (-j 8): 190.21 seconds / 18.93 kernels/hour [P:743%, 55 maj. pagefaults]
Run 7 (-j 11): 178.62 seconds / 20.15 kernels/hour [P:1011%, 96 maj. pagefaults]
Run 8 (-j 11): 185.62 seconds / 19.39 kernels/hour [P:975%, 126 maj. pagefaults]
```
Seems not a bad option for a hobbyist kernel hacker! Thanks @kernellogger for making kcbench!
```
Processor: AMD Ryzen 7 6800H with Radeon Graphics [16 threads]
Cpufreq; Memory: powersave [amd-pstate-epp]; 27841 MiB
Linux running: 6.12.48+deb13-amd64 [x86_64]
Compiler: gcc (Debian 14.2.0-19) 14.2.0
Linux compiled: 6.17.0 [.../.cache/kcbench/linux-6.17]
Config; Environment: defconfig; CCACHE_DISABLE="1"
Build command: make vmlinux
Filling caches: This might take a while... Done
Run 1 (-j 16): 161.38 seconds / 22.31 kernels/hour [P:1440%, 134 maj. pagefaults]
Run 2 (-j 16): 162.53 seconds / 22.15 kernels/hour [P:1441%, 140 maj. pagefaults]
Run 3 (-j 19): 172.87 seconds / 20.82 kernels/hour [P:1366%, 266 maj. pagefaults]
Run 4 (-j 19): 164.76 seconds / 21.85 kernels/hour [P:1446%, 258 maj. pagefaults]
Run 5 (-j 8): 190.83 seconds / 18.86 kernels/hour [P:742%, 49 maj. pagefaults]
Run 6 (-j 8): 190.21 seconds / 18.93 kernels/hour [P:743%, 55 maj. pagefaults]
Run 7 (-j 11): 178.62 seconds / 20.15 kernels/hour [P:1011%, 96 maj. pagefaults]
Run 8 (-j 11): 185.62 seconds / 19.39 kernels/hour [P:975%, 126 maj. pagefaults]
```
Seems not a bad option for a hobbyist kernel hacker! Thanks @kernellogger for making kcbench!
0
0
2
sj
sj
According to LWN, I was [1] 14-th most active developer for Linux v6.17, in terms of the number of changes. Numbers are no more no less just numbers, but being listed there twice [2] means something to me :)
[1] https://lwn.net/Articles/1038358/
[2] https://social.kernel.org/notice/AsT2MVtG6Lv7PONehE
#linux #kernel #damon #lwn
[1] https://lwn.net/Articles/1038358/
[2] https://social.kernel.org/notice/AsT2MVtG6Lv7PONehE
#linux #kernel #damon #lwn
0
0
3
sj
sj
Edited 29 days ago
I got my first DAMON t-shirt.
You might need to zoom-in to see why I'm saying it is a DAMON t-shirt, like you may need to zoom in/out DAMON-generated access pattern heatmaps, e.g., 'damo report heatmap --interactive_edit'[1].
[1] https://github.com/damonitor/damo/blob/next/USAGE.md#heatmap-scoping
#linux #kernel #damon #t-shirt
You might need to zoom-in to see why I'm saying it is a DAMON t-shirt, like you may need to zoom in/out DAMON-generated access pattern heatmaps, e.g., 'damo report heatmap --interactive_edit'[1].
[1] https://github.com/damonitor/damo/blob/next/USAGE.md#heatmap-scoping
#linux #kernel #damon #t-shirt
1
0
2
sj
sj
Kangrejos was awesome. I'm now in Paris for presenting [1] DAMON at Kernel Recipes, Tuesday.
[1] https://kernel-recipes.org/en/2025/schedule/overcoming-observer-effects-in-memory-management-with-damon/
#kangrejos #kernelrecipes
RE: https://social.kernel.org/objects/7f1cced2-39e6-4a15-8b0a-ba3cf3b3955b
[1] https://kernel-recipes.org/en/2025/schedule/overcoming-observer-effects-in-memory-management-with-damon/
#kangrejos #kernelrecipes
RE: https://social.kernel.org/objects/7f1cced2-39e6-4a15-8b0a-ba3cf3b3955b
0
0
3
sj
sj
Edited 1 month ago
In 2023, I walked [1] on the France route of Camino de Santiago from Burgos to Santiago de Compostela (~500 km) to prepare (walk on days, work on slides and code on nights) for DAMON talk [2] at OSSummit Europe. This year, I walked it again from Saint Jean Pied de Port to Burgos to complete the route, and further from Burgos to Carrion de los condes (~322 km) to prepare for Kangjeros [3] and Kernel Recipes [4]. Now I’m ready to attend talks and made a first draft [5] of my Kernel Recipes talk slides. Switching from a pilgrim to a conference attendee and heading to Kangrejos now. Looking forward to meeting people on Kangrejos and Kernel Recipes!
[1] https://social.kernel.org/notice/AZrKdcNdDyaDwkShou
[2] https://static.sched.com/hosted_files/osseu2023/92/damo_ossummit_eu_2023.pdf
[3] https://kangrejos.com/
[4] https://kernel-recipes.org/en/2025/schedule/overcoming-observer-effects-in-memory-management-with-damon/
[5] https://github.com/damonitor/talks/blob/master/2025/kernel_recipes/damon_kernel_recipes2025.pdf
#kangjeros #kernelrecipes #caminodesantiago
[1] https://social.kernel.org/notice/AZrKdcNdDyaDwkShou
[2] https://static.sched.com/hosted_files/osseu2023/92/damo_ossummit_eu_2023.pdf
[3] https://kangrejos.com/
[4] https://kernel-recipes.org/en/2025/schedule/overcoming-observer-effects-in-memory-management-with-damon/
[5] https://github.com/damonitor/talks/blob/master/2025/kernel_recipes/damon_kernel_recipes2025.pdf
#kangjeros #kernelrecipes #caminodesantiago
0
2
6
sj
sj
Edited 2 months ago
DAMON results are arguably hard to understand. And no single visualization fits all. Hence DAMON user-space tool provides a feature[1] that lets users program their own visualization in Python code.
A visualization recently suggested was showing size of memory for different hotness, and sort the size values by the hotness. It can intuitively show cold memory tail. We implemented it as another custom visualization program[2].
The script outputs the visualization in text format, that can easily be plotted as a chart, like attaching.
[1] https://github.com/damonitor/damo/blob/next/USAGE.md#damo-report-access-programming-visualization
[2] https://github.com/damonitor/damo/blob/next/report_access_exec_scripts/idle_time_mem_sz.py
#linux #kernel #damon #damo
A visualization recently suggested was showing size of memory for different hotness, and sort the size values by the hotness. It can intuitively show cold memory tail. We implemented it as another custom visualization program[2].
The script outputs the visualization in text format, that can easily be plotted as a chart, like attaching.
[1] https://github.com/damonitor/damo/blob/next/USAGE.md#damo-report-access-programming-visualization
[2] https://github.com/damonitor/damo/blob/next/report_access_exec_scripts/idle_time_mem_sz.py
#linux #kernel #damon #damo
0
0
0
sj
sj
Exploring mails on deep threads with hkml is not that easy. It is easy to miss on what mail currently the cursor is on, and which mail is replying to which mail.
To help this, below three new key bindings are implemented and pushed to master branch of hkml[1].
'w' and 'b' key move the cursor to next and previous words, respectively, similar to those on 'vim'.
'g' key highlights and unhighlights (toggles) the row and the column of the current cursor.
The word-based cursor movement is useful for moving it to beginning of the mail.
The highlighted lines can be useful for seeing on what mail the cursor is on, and what mails are replying to what mails.
[1] https://github.com/sjp38/hackermail
#hkml
To help this, below three new key bindings are implemented and pushed to master branch of hkml[1].
'w' and 'b' key move the cursor to next and previous words, respectively, similar to those on 'vim'.
'g' key highlights and unhighlights (toggles) the row and the column of the current cursor.
The word-based cursor movement is useful for moving it to beginning of the mail.
The highlighted lines can be useful for seeing on what mail the cursor is on, and what mails are replying to what mails.
[1] https://github.com/sjp38/hackermail
#hkml
0
0
1
sj
sj
Access pattern heatmap is powerful visualization, but exploring the time/space is not that easy[1]. To help that, we added an interactive zoom and scroll mode of the heatmap. You can try that with `damo report heatmap --interacrive_edit`. Attaching a demo gif.
[1] https://damonitor.github.io/posts/why_the_heatmap_is_not_showing_the_expected_access_patterns/
#linux #kernel #dmon #damo
[1] https://damonitor.github.io/posts/why_the_heatmap_is_not_showing_the_expected_access_patterns/
#linux #kernel #dmon #damo
0
0
2
sj
sj
Edited 4 months ago
A new DAMON user-space tool feature of today is drawing heatmap after sorting regions on the space by their access temperature.
The feature can be used with 'damo report heatmap --sort_temperature'. The access temperature calculation can be customized using '--temperature_weights'.
Attaching heatmps of a DAMON-snapshot and a DAMON-record, for unsorted and sorted heatmaps.
#linux #kernel #damon #damo
The feature can be used with 'damo report heatmap --sort_temperature'. The access temperature calculation can be customized using '--temperature_weights'.
Attaching heatmps of a DAMON-snapshot and a DAMON-record, for unsorted and sorted heatmaps.
#linux #kernel #damon #damo
0
0
2
sj
sj
Just pushed a new hkml feature for finding patch mails touching files for a reviewer based on MAINTAINERS file. More stabilizations and improvements would be needed, but seems working in a way? Kudos to @ljs for sharing the idea!
To use the feature, open mails list using 'hkml list' on linux tree, press 'm' key to open menu, and select 'search mails' item.
[1] https://github.com/sjp38/hackermail/blob/master/release_note
#hkml
To use the feature, open mails list using 'hkml list' on linux tree, press 'm' key to open menu, and select 'search mails' item.
[1] https://github.com/sjp38/hackermail/blob/master/release_note
#hkml
0
0
3
sj
sj
Edited 5 months ago
Based on a feedback from my colleague, we implemented an hkml [1] feature for finding patch mails that need more or no more attentions based on 'Reviewed-by:' tags.
Open mails list using 'hkml list', press 'm' key to open menu, and select 'search mails' item.
[1] https://github.com/sjp38/hackermail
#hkml
Open mails list using 'hkml list', press 'm' key to open menu, and select 'search mails' item.
[1] https://github.com/sjp38/hackermail
#hkml
0
0
2
sj
sj
Just released damo v2.6.1. Using this version and kernels that built with an RFC patch series, you can try page level properties based access monitoring. For example, you can know how much of anonymous and non-Young pages are in regions of different access pattern, like attaching image. Please refer to a blog post[1] for more details.
[1] https://damonitor.github.io/posts/damon_sz_filter_passed/
#linux #kernel #damon #damo #sz_ops_filter_passed
[1] https://damonitor.github.io/posts/damon_sz_filter_passed/
#linux #kernel #damon #damo #sz_ops_filter_passed
0
1
2
sj
sj
Just implemented heatmap visualization for each DAMON's access pattern snapshot. It is inpsired by Lorenzo's page layout idea[1].
Someone might be able to hack DAMON for generating page types snapshot instead of access pattern snapshot, and reuse some of the heatmap visualization code, if they are getting bored and get plenty spare time ;)
[1] https://social.kernel.org/notice/Ao0id4tH8kLlNztK0u
#linux #kernel #damon #damo
Someone might be able to hack DAMON for generating page types snapshot instead of access pattern snapshot, and reuse some of the heatmap visualization code, if they are getting bored and get plenty spare time ;)
[1] https://social.kernel.org/notice/Ao0id4tH8kLlNztK0u
#linux #kernel #damon #damo
0
1
2
sj
sj
Just released damo v2.5.4[1]. The major new feature in this release is recency/temperature histogram visualization[2] of DAMON's access pattern snapshots.
[1] https://github.com/damonitor/damo/blob/v2.5.4/release_note#L4
[2] https://github.com/damonitor/damo/blob/v2.5.4/USAGE.md#access-report-styles
#linux #kernel #damon #damo
[1] https://github.com/damonitor/damo/blob/v2.5.4/release_note#L4
[2] https://github.com/damonitor/damo/blob/v2.5.4/USAGE.md#access-report-styles
#linux #kernel #damon #damo
0
1
0
sj
sj
Just realized the new member looks similar to the penguin of FB kernel team logo hat I found from https://opensource.fb.com/linux/ (file name: kernel-team-logo.svg)
RE: https://social.kernel.org/objects/11adc48b-1b76-4d76-8e86-c6c89d87d895
RE: https://social.kernel.org/objects/11adc48b-1b76-4d76-8e86-c6c89d87d895
0
0
3
sj
sj
The bubble is not popped yet. The 30-days PyPI downloads of 'damo' has doubled (8,000 -> 16,000) again, about ten days after the last doubling.
#linux #kernel #damon #damo #pypistats
RE: https://social.kernel.org/objects/d7273398-9f50-43bc-97c8-14673a13d783
#linux #kernel #damon #damo #pypistats
RE: https://social.kernel.org/objects/d7273398-9f50-43bc-97c8-14673a13d783
0
1
1
sj
sj
After about ten days from the last posting[1] of damo [2] monthly downloads record, the count has doubled (4,000 -> 9,000) again. The time for doubling the count has been reduced from 18 months to 7 months, then to 10 days this time.
Because daily stat graphs on PyPI are not very convenient for monthly download stats, I plotted the 'rolling 30-day downloads' count. It shows this might be a temporal bubble that will pop soon. But, doubling is doubling. I'll seize tonight :D
[1] https://social.kernel.org/notice/AmYHJtlc5aI4ObXnKS
[2] https://github.com/damonitor/damo
#linux #kernel #damon #damo #pypistats
Because daily stat graphs on PyPI are not very convenient for monthly download stats, I plotted the 'rolling 30-day downloads' count. It shows this might be a temporal bubble that will pop soon. But, doubling is doubling. I'll seize tonight :D
[1] https://social.kernel.org/notice/AmYHJtlc5aI4ObXnKS
[2] https://github.com/damonitor/damo
#linux #kernel #damon #damo #pypistats
0
1
1