base: Add ATA/IDE userspace disk driver for DragonWare - #16
Draft
tseli0s wants to merge 33 commits into
Draft
Conversation
tseli0s
added
enhancement
tseli0s
marked this pull request as draft
August 22, 2026 19:51
tseli0s
force-pushed
the
disk-ata-support
branch
2 times, most recently
from
August 24, 2026 11:48
26e688c to
78e6e9d
Compare
tseli0s
force-pushed
the
disk-ata-support
branch
3 times, most recently
from
August 30, 2026 06:53
41785d9 to
13f1464
Compare
tseli0s
force-pushed
the
disk-ata-support
branch
2 times, most recently
from
August 31, 2026 18:14
1969810 to
2ebc46c
Compare
tseli0s
force-pushed
the
disk-ata-support
branch
from
September 7, 2026 23:00
2ebc46c to
d86bf3a
Compare
tseli0s
force-pushed
the
disk-ata-support
branch
3 times, most recently
from
September 9, 2026 08:55
8d67407 to
19bf30f
Compare
Originally, when DragonWare was first being written (Not even called DragonWare) the idea was that there would be two sets of system calls: One for DragonWare applications, and one for POSIX applications. At the time, it was not a microkernel, so the idea was that there would be system calls like open(), read(), sysctl(), et al. However, as DragonWare evolved, the stub code to eventually support POSIX system calls became bitrot that would never happen. It was only an idea to make porting software easier and address the fact that I was only familiar with UNIX like operating systems. In an effort to make the microkernel as pure as possible, remove all POSIX-related system call stub code inside kernel space. It will never be used because DragonWare won't implement POSIX support in kernel space. In the future, a POSIX server is planned instead - It will handle POSIX system calls entirely in userspace, and the kernel will remain tiny. But the kernel should not even know what POSIX is. Technically this is also a performance improvement, as some checks can now be removed entirely. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
About time we start tackling this, though this is just a very barebones thing and not anything useful (performs basic probing). Almost all of the code is taken verbatim from the diff of 2348fb0 and so far it appears to work well, though there's definitely some room for improvement. There's a long road ahead actually to finish what's being started in this commit, possibly talking about months here. But who the fuck cares? YOLO Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
It's still a work in progress (I mean, I'm still coding the driver :P) but this is the first step forward. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Instead of having everything in the main thread (and therefore being only able to do one reading at a time at one drive at a time, which is horribly inefficient), create listener threads instead. You think the code is ugly, hackish and makes zero sense right? So do I. Don't worry, this is not making it to the master branch, I just want to add some other changes I made which need this file to be committed. Again, sorry for the horrors you will see in this diff. PS. Yes, I'm dead serious, I should really implement malloc in libc and I should do so fast. Holy fuck, I want to gauge my eyes out. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Not needed, because there are separate ports for each bus, so each message will be directly dispatched to the correct listener. Or maybe not, I don't know, I'll test it out later. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
I have no idea why I did it like this last night, I was probably too tired. There's no reason to create those listener and IRQ ports on the main thread, they should be in the listener threads. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Writing the bytes one by one will be a pain in the ass and will look ugly as shit (Let's not even get on the technical aspect of this with all the ABI breakages that could occur and so on). Better to have structs instead that we copy directly on the payload. In the future, I will add some macro wizardry to do this automatically for any amount of (sufficiently fitting) structs instead of calling memcpy() on each and every one of them, and manually calculating the offsets of each, but for now, I want to get something working. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Part 1/2 (The actual disk read isn't implemented yet, but this is the skeleton of it). I already tested it with shared memory and everything, appears to work well, which means we're in good direction. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Will be needed for a tiny part in the ATA/IDE driver to make our lifes easier in PIO mode, so time to implement them now. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Why limit ourselves to 28 bits when we can immediately lay the future ahead with 48 bit support? (Or whatever the saying is). We can just force LBA28 for the time being, of course. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Needed to implement waiting on DRQs, checking ERR/DF and the rest. Maybe in the future they will go in another header, but for now, they just need to be available to the driver. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Fixes BSY never being cleared when attempting a READ command (the driver was testing the wrong bit). In preparation to push an early prototype of a readonly ATA/IDE driver. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
An attempt to fix a bug I am running into (And I am unable to even pinpoint right now). These functions may also come in handy in the future, and having them in a separate header is because I want to add more control function helpers in the future. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
* Define the macros ATA_ALTSTATUS_PRIMARY and ATA_ALTSTATUS_SECONDARY to be able to read without affecting interrupts the status of the current operation. * Fix the definitions of ATA_STATUS_* macros, otherwise checking or DRQ or RDY is actually incorrect when using these definitions to mask out the bits (In a test function that is not pushed here yet, which is where I discovered the bug). Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Will be shared by many parts of the driver because of how ATA/IDE works and the specification is structured, so time to share it. identify.c already has ATASoftResetDelay() (Which I have no idea why I named as such). Documentation is a bit generalizations and blah blah blah, don't worry, I'll fix it. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
IdentifyDrive() will disable ATA interrupts (nIEN bit) during the identify call, since there's no handler for this interrupt and the INTRQ line remains high (or something like that). Preparation to convert the driver into an interrupt-driven model. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Fixes some compiler warnings because the return value of inb() is being discarded (This is intentional, those are dummy reads to delay the CPU as the function name says, we don't care about the returned value) Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
* Replace calls to ATASoftResetDelay with the now exported Wait400ns() function instead to avoid code duplication. * Flush data returned by the IDENTIFY command (Read it but don't store it anywhere) to fix a soft lock on drive requests in QEMU (Better detailed in the comment left there) * Small other code cleanups. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Just following along with the DragonWare coding style where magic numbers are forbidden, while I am writing a basic read only ATA driver :P Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Found out I need it while developing a read function, because a driver bug may appear as hardware failure by accident without it to the caller programs. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
About time. The code here took about a week of debugging and ironing out, and yes, it's not perfect. I've tested it with a small utility function (Yes, from another process) and it does work however, so this is a great milestone to hit. Write support isn't implemented yet, and there are some optimizations I would like to do, but at the very least, there's a disk driver in DragonWare now. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Fixes a bug where listener threads were created for buses without any drives, causing programs to be able to see drives that weren't connected in the system at all. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
… attached idedrv is fundamentally an interrupt-driven driver, so we NEED interrupts to be able to issue further requests without polling. IdentifyDrive() disables them, so we need to reenable them manually. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
…`0x1000` Technically has no effect here (and never will) except trapping into the kernel to be 100% sure the number is correct. On x86, all page sizes will be 4KBs for the foreseeable future, but if this driver is to ever become portable to other architectures, we will need to know what page size we have, to allocate the correct stack size. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Will be needed to implement write support in the idedrv driver. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Both commands will be needed now that I am working on a prototype write implementation for the ATA driver. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Not implemented as of this commit, just a simple declaration while I am working on implementing it. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
After a long week of battling it out, I finally did it. It appears to work when running it through a basic test. To be honest, it's actually not that perfect, but as far as "get something written on a disk image" goes, we're finally here. Unfortunately, this write support is purely polling-based. I can't find how to replace it with interrupts instead. I also haven't implemented any error checking and I am not even sure this code is 100% standards compliant. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
With commit e1d590f now implementing some basic write support, it's time to handle this protocol request as well. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
There is a lot of ugly code because there's no malloc() in DragonWare as it stands in this file. About time we start cleaning it up, now that dwuser also has some nice wrappers to create a thread, for example. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Should speed up I/O performance since we don't have to wait for the disk to fetch the data physically, it's all in memory and we just copy it back into buf. Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
Fixes a bug in real hardware where the write call would fail as we didn't wait for the drive to "spin up". Signed-off-by: Aggelos Tselios <aggelostselios777@gmail.com>
tseli0s
force-pushed
the
disk-ata-support
branch
from
September 10, 2026 08:16
19bf30f to
f60fce1
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
About time we start tackling this, though this is just a very barebones thing and not anything useful (performs basic probing). Almost all of the code is taken verbatim from the diff of 2348fb0 and so far it appears to work well, though there's definitely some room for improvement.
There's a long road ahead actually to finish what's being started in this commit, possibly talking about months here.
Signed-off-by: Aggelos Tselios aggelostselios777@gmail.com