Stefan Hajnoczi: GDB Remote Debugging

Milestones:

• [b44] Tested and clean for mainline review.
• [r-o-m] Fix known ROM-o-matic bugs.

Git commits:

• [DMA] Add test for bounce buffers.
• [b44] Begin driver clean up work
• [b44] Auto-negotiate link

Posted DMA mapping patches to the mailing list. I am hoping to get feedback on whether this approach looks good.

Cleaning up b44 driver. I am trying to get rid of unused functionality from the driver. The aim is to simplify the code for humans and reduce code size. I noticed that loading a 6 MB initrd via HTTP hangs on my hardware, this is probably an issue in the b44 driver or DMA mapping code. Debugging this is my next task.

Fixed known ROM-o-matic bugs. Soon ROM-o-matic will be added to gPXE contrib/. This will allow anyone to run their own or contribute to mainline development. There were a few known bugs that I tracked down and fixed today.

E820 memory map splitting support. Yesterday's hang while loading an initrd turned out to be caused by me misunderstanding e820mangler.S. gPXE can mark memory regions as hidden. When gPXE or an image use the E280 BIOS memory map function, e820mangler.S will hide these memory regions. This prevents gPXE memory from being overwritten or used by another program.

The bug was my assumption that e820mangler.S would split memory regions around arbitrary hidden regions. I haven't deciphered all of the code yet, but printing the memory map after placing a hidden region inside a larger free region shows that arbitrary splits aren't performed. Here is the picture:

+----------------+                 +----------------+
|   free region  |                 | this region is |
|                |                 | not produced!  |
| +------------+ |  e820mangler.S  +----------------+
| | my hidden  | |   ==========>
| | region     | |
| +------------+ |                 +----------------+
|                |                 |   truncated    |
|                |                 |   region       |
+----------------+                 +----------------+

You can see only the bottom region is being produced. The top region is not produced by e820mangler.S. At least that's what I think is going on.

I had a look at the E820 interface and will start working on a solution tomorrow. It gets nasty when there are N hidden regions inside one free region, thereby splitting it up into many smaller regions.

Updated GDB debugging documentation now that debugging over UDP and watchpoints are in mainline. The documentation still needs work but I have at least updated the parts that are there so they are correct.

This week I haven't kept my journal up-to-date. I have not been writing much gPXE code. Instead I worked on ROM-o-matic, GDB documentation, bug reports on IRC, understanding gpxelinux.0 so we can merge changes into mainline, and understanding e820mangler.S to get DMA mapping working.

b44 progress has been stuck at E820 memory splitting. I will talk to mcb30 today about e820mangler.S. I am misunderstanding that code because its output does not match my expectations. I'd like to make progress on E820 mapping this weekend and it will open the door to getting DMA mapping and the b44 working and into mainline.

I am in dialogue with hpa, mcb30, and mdc so we can merge gpxelinux.0 into gPXE. SYSLINUX now builds gpxelinux.0, a PXE NBP that uses the UNDI driver with gPXE (aka undionly.kpxe) to run PXELINUX. A relatively small set of changes was needed to make this work on some hardware. The aim is to get this code merged into gPXE mainline.

On to Week 7.