Sign In
Create Account
25
On a hackintosh, the sandy bridge graphics driver exhibits a constant event ring stall until the stall is cleared via either hotplugging the display connector (and leaving it removed for ~5 seconds to clear the stall), or by using more than 1 display connectors simultaneously. This stalling can be seen as very slow graphics display updates (~5 second delays between mouse clicks and responses) and or by looking at the kernel log, /var/log/kernel, which shows errors such as:
May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: WaitForStamp: Overflowed waiting for stamp 0x2f3 on Main ring: called from May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: timestamp = 0x02cc May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: **** Debug info for apparent hang in Main graphics engine **** May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: ring head = 0x00600eb4, wrap count = 0x 3 May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: ring tail = 0x00000110 ring control = 0x00003801 enabled, auto report disabled, waiting, semaphore not waiting, length = 0x004 4KB pages May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: timestamps = 0x02cc May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: Semaphore register values: May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: VRSYNC: (0x12044) = 0x2cc May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: BRSYNC: (0x22040) = 0x0 May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: RVSYNC: (0x 2040) = 0x0 May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: BVSYNC: (0x22044) = 0x0 May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: RBSYNC: (0x 2044) = 0x0 May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: VBSYNC: (0x12040) = 0x0 May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: Looks like Main ring is stuck waiting on an event May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: After attempt to clear wait condition = 0x00003001 no longer waiting May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: kIPEHR: 0x7a000003 May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: kINSTDONE: 0xfffb May 25 23:20:45 xs-MacBook-Pro-2 kernel[0]: kINSTDONE_1: 0x230003fFor the driver fix, some prerequisites:
- osx 10.7 and intel sandy bridge graphics hardware (intel h67/z68 chipset)
- Installation of OSX Developer tools (if you need to lookup a PlatformInformationList that has not already been done by someone else). The OSX developer tools are found within Apple's xcode distribution, here: http://developer.apple.com/xcode/, also most commonly available via apple's app store.
- Understanding of hex editing
- Ability to troubleshoot by looking at ioregistry
Older code (10.6 including the 2011 MBP sandy bridge update) seems to strictly rely upon the os-info ioregistry entry for this information. I have not, and probably won't, look into editing os-info to change connector information.
With a 10.7 based version of AppleIntelSNBGraphicsFB.kext, the code uses the board-id as an index into the PlatformInformationList[] table. This table is pretty handy, and looks like:
/* 12 byte connectorinfo */
typedef struct {
char byte0;
char byte1; /* i2cid? */
char unused[2];
uint32 connectortype; /* The connector type, see below */
char byte[4];
} connectorinfo_t;
/* 60 byte total */
typedef struct {
char byte0; /* Presence? */
char byte1; /* For display pipe max connector index? */
char byte2; /* Usable Connector count */
char byte3; /* Appears unused */
char byte[8];
connectorinfo_t connectors[4];
} PlatformInformationList_type;
PlatformInformationList_type PlatformInformationList[8];
where integers are represented in intel little-endian byte order, and bit 0 is the LSB.
In English, there are 8 tables in the list, each 60 bytes long. Each 60 byte table entry starts with a 12 byte header, followed by 4 connector information blocks. The connector information blocks are each 12 bytes long.The connectorytype field appears to use the same codes as the ATI driver, where the only used values used by the driver thus far are:
#define CONNECTORTYPE_LVDS 0x00000002 /* Ie internal Low Voltage display, such as laptop */
#define CONNECTORTYPE_DP 0x00000400 /* Displayport */
#define CONNECTORTYPE_HDMI 0x00000800
If you don't set your board-id to a recognized value, a default connector table is instead taken from the address PlatformInformationDefault. THe default table only has 1 usable connector so this is likely to give you reduced functionality.
Setting SMboardproduct to various sandy bridge board-ids defines the index into PlatformInformationList[]. Interestingly only indexes 0 thru 5 are currently allowed (even though there are two more table entries...) In any case, for the purpose of this note, we'll use index 0 by setting the board-id to the MacBookPro8,1 value. In smbios.plist:
<key>SMboardproduct</key> <string>Mac-94245B3640C91C81</string>
Unfortunately all 8 table entries use an LVDS connector for the first value, and the driver seems to malfunction if you have nothing plugged into this port. In my case I would see the kernel log fill up with errors about a stuck ring:
kernel[0]: WaitForStamp: Overflowed waiting for stamp 0x2f3 on Main ring: called from ... kernel[0]: Looks like Main ring is stuck waiting on an event kernel[0]: After attempt to clear wait condition = 0x00003001 no longer waitingAnd it would get re-stuck every 5 seconds or so. More importantly this would make the graphics unusably slow unless I unplugged the display for > 5 seconds first. At that point the stuck ring problem would "heal"... Using a dual-head setup also seems to avoid this problem.
However, rather than hot-plugging the graphics display upon every boot, one can instead modify the above table to avoid this problem.
First we find the PlatformInformationList table. In the example below I'm using DP4:
% cd /System/Library/Extensions/AppleIntelSNBGraphicsFB.kext/Contents/MacOS % nm -arch x86_64 AppleIntelSNBGraphicsFB | grep 'D _PlatformInformationList' 00000000000295f0 D _PlatformInformationList % lipo -detailed_info AppleIntelSNBGraphicsFB Fat header in: AppleIntelSNBGraphicsFB fat_magic 0xcafebabe nfat_arch 2 architecture x86_64 cputype CPU_TYPE_X86_64 cpusubtype CPU_SUBTYPE_X86_64_ALL offset 4096 size 283008 align 2^12 (4096) architecture i386 cputype CPU_TYPE_I386 cpusubtype CPU_SUBTYPE_I386_ALL offset 290816 size 293092 align 2^12 (4096) %So now we know the disk address for this table is:
0x295f0+4096=173552
Since there are 8 tables of 60 bytes, we could dump 8*60=480 bytes to see all 8 tables. Since I've chosen to work on the first entry, I'll dump just 60 bytes.
ALso since each connector is 12 bytes, I dump 12 bytes per line for our viewing pleasure.
% dd if=AppleIntelSNBGraphicsFB of=/tmp/table bs=1 skip=173552 count=60 60+0 records in 60+0 records out 60 bytes transferred in 0.000142 secs (422955 bytes/sec) % cd /tmp % xxd -c 12 < table > hex % cat hex 0000000: 0102 0400 1007 0000 1007 0000 ............ 000000c: 0503 0000 0200 0000 3000 0000 ........0... 0000018: 0205 0000 0004 0000 0700 0000 ............ 0000024: 0304 0000 0004 0000 0900 0000 ............ 0000030: 0406 0000 0004 0000 0900 0000 ............ %In the above table you can see that 4 connectors are marked usable, 1 LVDS and 3 DP. On my h67 motherboard, I have 4 connectors: VGA, DVI, DP and HDMI. The DVI, DP and HDMI connectors all work with the DP connectortype shown in this table (the intel driver doesn't seem to use the connectortype for much).
I believe I can tell which of the above entries match my hardware by checking which entry has av-signal-type set, (or when AAPL,DisplayPipe != 0xffff?). I certainly can by exhausting testing of the combinations...
Now, to remove the problematic LVDS entry, I edit the hex file, producing:
0000000: 0102 0300 1007 0000 1007 0000 ............ 000000c: 0304 0000 0004 0000 0900 0000 ............ 0000018: 0205 0000 0004 0000 0700 0000 ............ 0000024: 0406 0000 0004 0000 0900 0000 ............ 0000030: 0000 0000 0000 0000 0000 0000 ............In the above I moved the hex from the 3rd connector (row 4) to the first (row 2), zeroed the hex in row 5, and subtracted 1 from the connector count in row 0.
Now I write back the result:
xxd -c 12 -r hex >! table % dd if=table of=/System/Library/Extensions/AppleIntelSNBGraphicsFB.kext/Contents/MacOS/AppleIntelSNBGraphicsFB bs=1 seek=173552 conv=notrunc % touch /System/Library/ExtensionsViola, 3 working connectors and no driver hangs, instead of 1 broken connector & 3 working connectors with driver hangs.
