Monday, February 16, 2004

PNIC, how I hate thee Struggled most of the day trying to get the PNIC to work under darwin-x86. This code really had not been touched for about a year. When I last left it, I'm pretty sure it worked at 10/half. The machine I was testing it on then (PowerMac G4 533) was a nice little box. The machine I have now, a dual G5, doesn't have the right PCI slot to take the 5volt card so I can't test it there anymore. It ain't working over in x86-land. By all appearances it is working, but nothing hits the wire. I can successfully download the setup frame, and the MII logic detects link up and down. It also successfully receives broadcast and multicast traffic. However, it won't transmit anything. Since it successfully sees the broadcasts, I know the packet byte flipping is working correctly (I use an IOMbufNaturalMemoryCursor which picks the natural byte-order of the host architecture... works great for all the other cards). TX failure... I'm assuming this has something to do with the GPIO register. I was testing a 10/100 switch and it was writing 0x20 to CSR12. I switched back to using the 10/half hub and it went back to writing 0x34. My notes from a year ago say that is correct. I need to overhaul that logic to see if I can get this thing going. Right now in the log it prints that it is set for 10/half even when it was plugged in at 100/full. Argh. Very frustrating. I'm stuck on my QT Component project and this one isn't faring much better. Can I get a brain transplant, cuz I'm just not smart enough to figure this shit out. 10:37:19 PM    comment []

Primary Interrupt filter wrappers Changed the static method for managing the primary interrupt filter so that I could override it in a subclass. This is the same technique I used for the IOInterruptEventSource and IOTimerEventSource methods, so I'm pretty comfortable with it. It serves two functions. One, it gets rid of a ton of gcc warnings during compilation (I like things clean). Two, it allows me to override the primary interrupt filter in a subclass. I'm doing that right now to track down an ugly problem with the PNIC on x86. It gets stuck in an interrupt loop, constantly calling the driver to handle a secondary interrupt. This must mean that the interrupt condition isn't being cleared by the primary interrupt filter, but it works just fine for the DEC2114x and ADMtek classes. Odd. 3:28:13 PM    comment []

Sharking the driver I had a ton of trouble profiling the driver a couple of weeks ago. Shark would generate some information for me, but double-clicking on the offending routine name would cause it to jump to the wrong source code in the file. This frustrated me no end. I checked the search paths, I did clean builds, I reinstalled Shark... nothing cured the problem. Today I was rereading an <a href="http://www.opendarwin.org/en/faq.php#chud2">entry I wrote for the OpenDarwin FAQ on performance and tuning</a>.  In a flash it occurred to me what I was doing wrong. Here's how I was testing. I had a known, good copy of the driver sitting in /System/Library/Extensions. I would kextunload it. Next, I would copy my debug or test version to /tmp and kextload it from there. Then I'd run my performance test. The idea was that if the machine crashed or needed a reboot, /tmp gets blown out so the "bad" driver is erased, and the good driver will get loaded since it still lives at /S/L/E. Here's the problem. Shark was loading the symbol file from the driver sitting in /S/L/E. It didn't know about the one sitting in /tmp. The one in /tmp was, at times, pretty radically modified from the stock version so it makes total sense that the symbols and source code line numbers no longer matched up in Shark. Two ways to fix this. One, move the good version of the driver out of /S/L/E and do all testing there. This is a bad solution since a buggy, crashy driver could force me to boot to another partition to remove it (e.g. it crashes during boot). Second, edit the Shark preference file to point to the driver sitting in /tmp. I like Option 2. 11:42:05 AM    comment []

Need a break from this project... Oh, my head is throbbing. I've been banging it against the QT wall for the past two days. I'll be taking a break from this project today to work on my other project. Here's an excerpt from a thread on the quicktime-api list. I'm having trouble getting the component to match by filename extension. This is necessary since FLAC files won't contain any regular QT data as they get shuffled around from platform to platform. The only way to tell it is FLAC would be if it is called "some_audio_file.flac". But QT isn't being very cooperative and it's pissing me off. --- start message --- From: cremes@mac.com Subject: Re: QT Movie Import, filename matching Date: February 16, 2004 11:02:47 AM CST To: quicktime-api@lists.apple.com On Feb 15, 2004, at 9:42 PM, chuck remes wrote: >On Feb 15, 2004, at 5:50 PM, Steve Nicolai wrote: >>QuickTime uses the thng and thga resources to find the importers. The component type must be 'eat '. The component subtype is >> a finder file type if the component flag movieImportSubTypeIsFileExtension is not set, or an uppercase version of the file >> extension if movieImportSubTypeIsFileExtension is set. >> >>If you have both a file type and an extension that you want to support, you need two components, generally a thng and a thga. >> >>Speculating on what QuickTime does internally, if it is given a file type, it will first search for a component: >> >>type 'eat ' >>subtype >>manufacturer 0 >>flags 0 >>flagsMask movieImportSubTypeIsFileExtension >> >>If it doesn't find a component or a file type is not given, it then looks for an extension on the file. If it finds one, it then searches >> for: >> >>type 'eat ' >>subtype >>manufacturer 0 >>flags movieImportSubTypeIsFileExtension >>flagsMask movieImportSubTypeIsFileExtension >> >>The mcfg resource isn't used by QuickTime to find components. I think it's used to register with Windows that QuickTime >> supports these file types. >> > I've double-checked all of this and it appears just as you outlined. I'm now concerned that the reason it isn't working is due to > some incorrect setting within the XCode project itself. I created the project from scratch and added files to it rather than cloning > the ElectricImage project and replacing the files. > >I'd like to put a breakpoint in the code where QT interfaces to this component. Is there a recommended way of doing this? I've read > all the archived messages on this list and coreaudio-api for using gdb with QT components, but none of them spelled out how to > trap entry into the component. In other words, I'm not 100% certain this component is executing when the open file dialog comes > up. The experiment continues. I "cloned" the ElectricImage sample and added my files to it. After building, still no joy on matching by filename. At least I eliminated the possibility that it is related to my XCode project settings. My next thought was to start changing things in the original ElectricImage sample until I broke the filename match. Imagine my surprise when it broke after I modified the EI_GraphicsImport.r resource file. This resource file contains a 'thng' and 'thga' resource of type 'grip' just like the EI_MovieImport.r resource file contains one of type 'eat '. However, the 'grip' resource seems to be the one controlling whether or not the OpenFile dialog box matches on the filename. If I remove this file from the project and compile the project with the MovieImporter providing the only 'thng' and 'thga' resources of type 'eat ', it will NOT match on filename. Adding it back in gets the match to work again. I tried flipping the resource IDs so the MovieImport 'eat ' resource had a lower ID, but it didn't help. So what I've learned from this is that I can't match by filename without a 'thng' and 'thga' resource that registers itself as type 'grip'. Having a properly configured 'thng' and 'thga' of type 'eat ' is insufficient for mapping by filename. I am well and truly stuck. I'm trying to create an audio component, so the concept of a 'grip' component doesn't fit at all. Help? cr _______________________________________________ quicktime-api mailing list | quicktime-api@lists.apple.com Help/Unsubscribe/Archives: http://www.lists.apple.com/mailman/listinfo/quicktime-api Do not post admin requests to the list. They will be ignored. --- end message ---- 11:31:12 AM    comment []