Hyeonggon Yoo
Posts
469Following
90Followers
94- Intern @ NVIDIA Korea (Security System Software) (2024.06 ~)
- Ex-Intern @ Panmneisa (CXL emulation stuff) (~2023.12)
- Undergraduate majoring CSE (estimated graduation: Feb. 2025)
- Working as reviewer at Linux Slab subsystem
- Born in August 6, 2000
Opinions are my own.
My interests are:
Memory Management,
Computer Architecture,
Circuit Design,
Virtualization
Hyeonggon Yoo
hyeyoo
Edited 1 year ago
https://coverage.kerneltesting.org/slab-experimental-6283c415/mm/
This is a merged LCOV output of my mm & slab test coverage from 30 different kernel configurations.
Looks mostly fine but there are a few small problems:
- Sometimes numbers are inconsistent
- Omission of (some) inline function coverage
This is a merged LCOV output of my mm & slab test coverage from 30 different kernel configurations.
Looks mostly fine but there are a few small problems:
- Sometimes numbers are inconsistent
- Omission of (some) inline function coverage
0
0
1
Hyeonggon Yoo
hyeyoo
Edited 1 year ago
I need to learn Ansible instead of writing non-portable scripts and memorizing all of them
0
0
1
Hyeonggon Yoo
repeated
repeated
Vlastimil Babka
vbabka
A LKML thread meme summary, pure shitposting because SCNR, not meant as an endorsement nor ridicule of any side
Show content
STOP DOING MSEAL()
SYSCALLS WERE NOT SUPPOSED TO BE GIVEN MM_SEAL_ FLAGS
YEARS OF MINMUTABLE() yet NO REAL-WORLD USE FOUND for using more than addr, len PARAMETERS
Wanted not to bother with that anyway to get a break? We had a tool for that: It was called "LD.SO"
"one place honors the sealing, random other places do not" - Statements dreamed up by the utterly Deranged.
LOOK at what Chrome has been demanding your Support for all this time, with all the mprotect(2) permission reduction & OPENBSD_MUTABLE we built for them.
(This is REAL Code, done by REAL Kernel developers).
ON_BEHALF_OF_KERNEL ?????
ON_BEHALF_OF_USERSPACE ???????
VM_SEAL_* -> MM_SEAL_* -> MM_ACTION_* ?????????????????
"Hello, I would like a MM_SEAL_MUNMAP but not MM_SEAL_MMAP VMA please."
They have played us for absolute fools
SYSCALLS WERE NOT SUPPOSED TO BE GIVEN MM_SEAL_ FLAGS
YEARS OF MINMUTABLE() yet NO REAL-WORLD USE FOUND for using more than addr, len PARAMETERS
Wanted not to bother with that anyway to get a break? We had a tool for that: It was called "LD.SO"
"one place honors the sealing, random other places do not" - Statements dreamed up by the utterly Deranged.
LOOK at what Chrome has been demanding your Support for all this time, with all the mprotect(2) permission reduction & OPENBSD_MUTABLE we built for them.
(This is REAL Code, done by REAL Kernel developers).
ON_BEHALF_OF_KERNEL ?????
ON_BEHALF_OF_USERSPACE ???????
VM_SEAL_* -> MM_SEAL_* -> MM_ACTION_* ?????????????????
"Hello, I would like a MM_SEAL_MUNMAP but not MM_SEAL_MMAP VMA please."
They have played us for absolute fools
3
8
21
Hyeonggon Yoo
hyeyoo
Edited 1 year ago
This is how a silently enabled kernel configuration can affect your system performance:
https://lore.kernel.org/stable/97397e8d-f447-4cf7-84a1-070989d0a7fd@amazon.com/T/#u
I fixed this at the beginning of this year, but did not backport to stable trees. Luiz Capitulino from Amazon reported that this resulted in performance regression in their 6.1 kernels.
DEBUG_PREEMPT has been enabled by default since pre-git era for PREEMPT kernels, but distros that were using PREEMPT_VOLUNTARY silently enabled PREEMPTION and DEBUG_PREEMPT after the introduction of PREEMPT_DYNAMIC. And DEBUG_PREEMPT is better disabled for production kernels regardless of the preemption model used.
https://lore.kernel.org/stable/97397e8d-f447-4cf7-84a1-070989d0a7fd@amazon.com/T/#u
I fixed this at the beginning of this year, but did not backport to stable trees. Luiz Capitulino from Amazon reported that this resulted in performance regression in their 6.1 kernels.
DEBUG_PREEMPT has been enabled by default since pre-git era for PREEMPT kernels, but distros that were using PREEMPT_VOLUNTARY silently enabled PREEMPTION and DEBUG_PREEMPT after the introduction of PREEMPT_DYNAMIC. And DEBUG_PREEMPT is better disabled for production kernels regardless of the preemption model used.
0
6
8
Hyeonggon Yoo
hyeyoo
Why are gcov/lcov/genhtml so sensitive to source code path :(
lcov/genhtml requires that there should be source code in the exactly same path of source code in the build machine.
lcov/genhtml requires that there should be source code in the exactly same path of source code in the build machine.
1
0
0
Hyeonggon Yoo
hyeyoo
Errr.... IDK why people contact me via Linkedin asking 'Is your company willing to by hardware from us?'
I DON'T KNOW I AM AN INTERN :(
I DON'T KNOW I AM AN INTERN :(
2
0
2
Hyeonggon Yoo
hyeyoo
Edited 1 year ago
https://it.chosun.com/news/articleView.html?idxno=2023092102422
...Linux for Desktop opens new world...
...New opportunity for Korean OS in goverment PCs...
https://www.hancom.com/product/productGooroomMain.do
...Goorm OS provides secure, convenient, unique experience to desktop users...
...Goorm OS is based on debian linux...
Seriously?
I'm tired of fancy new Korea's OS platforms (K-OS) which is either 1) just another Linux distribution or 2) not Linux based but being so unusable
...Linux for Desktop opens new world...
...New opportunity for Korean OS in goverment PCs...
https://www.hancom.com/product/productGooroomMain.do
...Goorm OS provides secure, convenient, unique experience to desktop users...
...Goorm OS is based on debian linux...
Seriously?
I'm tired of fancy new Korea's OS platforms (K-OS) which is either 1) just another Linux distribution or 2) not Linux based but being so unusable
0
0
1
Hyeonggon Yoo
repeated
repeated
nixCraft 🐧
nixCraft@mastodon.social
Edited 1 year ago
Windows: 0
Linux: 1
It is not Linux's job to stop you from shooting your foot. If you so choose to do so, then it is Linux's job to deliver Mr. Bullet to Mr. Foot in the most efficient way it knows.
2
3
1
Hyeonggon Yoo
hyeyoo
Hmm when running builds and testcases, my machine sometimes becomes very unstable and stuck for a while.
this causes some build/test failures because git clone or job submit timeouts are exceeded.
it's very annoying and I need to fix, but I couldn't find out why. I need to setup a monitoring system...
this causes some build/test failures because git clone or job submit timeouts are exceeded.
it's very annoying and I need to fix, but I couldn't find out why. I need to setup a monitoring system...
1
0
0
Hyeonggon Yoo
hyeyoo
Dealing with large git repositories in Jenkins
https://www.jenkins.io/files/2016/jenkins-world/large-git-repos.pdf
https://randyfay.com/content/reference-cache-repositories-speed-clones-git-clone-reference
https://www.jenkins.io/files/2016/jenkins-world/large-git-repos.pdf
https://randyfay.com/content/reference-cache-repositories-speed-clones-git-clone-reference
2
0
2
Hyeonggon Yoo
hyeyoo
Edited 1 year ago
now I'm slowly getting how CXL manages cache coherence of device memory (HDM-{H,D,DB}). my head aches.
BTW this is a good material to grasp some important concepts of of CXL.
https://arxiv.org/abs/2306.11227
Even if it's not the final version yet.
BTW this is a good material to grasp some important concepts of of CXL.
https://arxiv.org/abs/2306.11227
Even if it's not the final version yet.
0
0
2
Hyeonggon Yoo
hyeyoo
Edited 1 year ago
progress of this week:
#kerneltesting side project
Now my testing system can submit and run LAVA [1] test jobs automatically after jenkins [2] build!
[1] https://lava.kerneltesting.org/scheduler/alljobs
[2] https://jenkins.kerneltesting.org/job/slab/
currently it runs those:
- a subset of mm LTP testcases
- hackbench for a very short time
#kerneltesting side project
Now my testing system can submit and run LAVA [1] test jobs automatically after jenkins [2] build!
[1] https://lava.kerneltesting.org/scheduler/alljobs
[2] https://jenkins.kerneltesting.org/job/slab/
currently it runs those:
- a subset of mm LTP testcases
- hackbench for a very short time
0
1
6
Hyeonggon Yoo
hyeyoo
Edited 1 year ago
```
As a bit of history, in the early days of PCI the spec writers anticipated cases
where PCI would actually replace the processor bus.
```
from "PCI Express Technology 3.0", MindShare
And yeah, that seems to be revived in 2019 by Intel, in the name of IAL (Intel Accelerator Link), which was renamed to CXL (Compute Express Link) later. The main reason why Intel donated IAL was to avoid making the industry fragmented. Later CXL consortium was founded and over 250 companies now contribute to the specification. Including my company! (who hired me as an intern)
As a bit of history, in the early days of PCI the spec writers anticipated cases
where PCI would actually replace the processor bus.
```
from "PCI Express Technology 3.0", MindShare
And yeah, that seems to be revived in 2019 by Intel, in the name of IAL (Intel Accelerator Link), which was renamed to CXL (Compute Express Link) later. The main reason why Intel donated IAL was to avoid making the industry fragmented. Later CXL consortium was founded and over 250 companies now contribute to the specification. Including my company! (who hired me as an intern)
0
1
3
Hyeonggon Yoo
hyeyoo
Edited 1 year ago
it is interesting to learn how PCIe endpoints and switches implements their own protocols (similar to network protocols) for communication (PCIe transport/data link/physical layers)
but since I know very little about electronics some topics are hard to understand, like why it's difficult to increase speed in parallel bus architecture or why errors occur during transmission, or why DC components should be avoided.
but since I know very little about electronics some topics are hard to understand, like why it's difficult to increase speed in parallel bus architecture or why errors occur during transmission, or why DC components should be avoided.
3
0
0
Hyeonggon Yoo
repeated
repeated
LWN.net
LWN@fosstodon.org[$] Progress on no-GIL CPython https://lwn.net/Articles/947138/ #LWN
0
2
1
Hyeonggon Yoo
hyeyoo
Edited 1 year ago
Learning how to write a LAVA test definition, but the tricker thing is to decide which tests to run to verify a kernel works fine.
Candidates:
- LTP
- KUnit
- kselftests
btw it is funny that the entire LTP suite gets killed every time it runs oom testcases. and LTP takes quite long time for a lightweight testing.
hmm... maybe run only a smaller subset of them?
Candidates:
- LTP
- KUnit
- kselftests
btw it is funny that the entire LTP suite gets killed every time it runs oom testcases. and LTP takes quite long time for a lightweight testing.
hmm... maybe run only a smaller subset of them?
0
0
1
Hyeonggon Yoo
hyeyoo
Edited 1 year ago
Today I learned:
CXL memory can be mapped as 'System RAM' or 'Soft Reserved' by platform firmware. Or it can be dynamically provisioned by (since v6.3) CXL region driver.
And 'Soft Reserved' or dynamically provisioned CXL RAM region can be used in two ways:
1. Applications mmap() to /dev/daxX.Y files, just like traditional persistent memory devices.
2. Kernel use it as System RAM via dax_kmem driver.
And a weird fact is that when dax_kmem onlines CXL (and other performance-differentiated like pmem) memory, to ZONE_NORMAL, not ZONE_MOVABLE.
CXL memory can be mapped as 'System RAM' or 'Soft Reserved' by platform firmware. Or it can be dynamically provisioned by (since v6.3) CXL region driver.
And 'Soft Reserved' or dynamically provisioned CXL RAM region can be used in two ways:
1. Applications mmap() to /dev/daxX.Y files, just like traditional persistent memory devices.
2. Kernel use it as System RAM via dax_kmem driver.
And a weird fact is that when dax_kmem onlines CXL (and other performance-differentiated like pmem) memory, to ZONE_NORMAL, not ZONE_MOVABLE.
2
0
1