December 2001 - Darwin-kernel

Re(2): Heathrow interrupt weirdness
by Benjamin Herrenschmidt 11 Dec '01

11 Dec '01

On Mon, Dec 10, 2001, Josh de Cesare <decesare(a)apple.com> wrote: > We have know about the typos for some time. The are all >basically benign. We plan to fix them when there is time for the >changes to be properly tested. I plan to get rid of the pending >interrupt variable. They were a product of paranoia. When the >handler fires, if the level sensitive interrupts are active then they >will be dispatched. When I fix the driver I will go over the level >masks. > As far as lost interrupts I have not seen any. There is of >course the well know behavior of the interrupt controllers of this >type. If an interrupt becomes active (level or edge) while it is >masked, it will not generate an interrupt when it is unmasked. Yes, that's well known and we have a workaround as well in linux, by testing the "level" register when unmasking. I'm not too sure if we might still lose edge interrupts there as their level may have been back to 0 already, but there's nothing much we can do if this is the case. I think my problem with the PMU _might_ be related to a buglet in my code where I write the "clear" register (we call it ack) twice, I've cleaned up my code a bit and am now waiting for more user reports. (Basically, I stopped receiving CB_1 interrupts after a while, can be pretty long, about several hours. Just forcing the driver to send a new interrupt ack command "fixed" it). BTW. I also noticed that on core99, I get spurrious GPIO1 interrupts. Basically, things acts like if after every "normal" GPIO1 interrupt, I get a spurrious one (that is the PMU doesn't return any data back from the interrupt ack command). It might be normal behaviour, but I suspect it may be also a subtle issue with the GPIO used for the interrupt. I don't know much about how GPIOs work on KeyLargo outside of setting the output enable & data bits, but I did find a few bit definitions in the AppleTexasAudio driver, which seem to explain that interrupt GPIOs can be programmed to trigger on dual-edge. That may explain why any 0-1-0 transition of ther PMU interrupt would cause 2 interrupts to be actually triggered. I'm not sure nor I know if Darwin has the same issue as the PMU driver isn't open source. It's mostly harmless except for the additional interrupt/CPU load caused which may slightly reduce battery life ;-) Regards, Ben.

1 0

Re: Heathrow interrupt weirdness
by David A. Gatwood 10 Dec '01

10 Dec '01

On Mon, 10 Dec 2001, Benjamin Herrenschmidt wrote: > While looking at the Heathrow.cpp code to compare with what I'm doing > in Linux (chasing a lost interrupt problem with the PMU that occurs > occasionally), I found something weird in the Darwin implementation. Hey, Ben. At least it only occurs occasionally for you. In MkLinux, it's relatively easy to trigger. :-| > That gives on a wallstreet the NMI & cascade (second heathrow) beeing > edge (while the NMI is specifically level triggered on openpic machines > and that makes more sense to have it level triggered) Not sure on the NMI. > , while all others > beeing edge (especially IDE disk interrupts which are usually known to > be level interrupts as the disk assert them until the status register > is read) My gut is telling me that continuing to trigger interrupts until the status register is read could generate spurious interrupts that would cause performance problems. After all, clearing the interrupt bit happens in the interrupt context, but reading the status bit probably happens in a kernel thread. Just my gut. > There's also a typo in the code for handling lost interrupts which > uses the first level register in the code for checking the second > register set (I just posted a bug report about this). Cough. Guess Darwin doesn't lose many interrupts or someone would have noticed this. :-) > Finally, the code og getVectorType() is confusing. It should probably > test the vector is in the range of the first register set before > doing (1 << vectorNumber) for completeness. The current code works > as long as (1 << X) with X >= 32 returns 0 but I did see compilers > optimizing this badly, so I feel this is dangerous coding. I don't seem to have the Heathrow driver handy, but looking at the GC driver (which should be very similar, I hope), it's being &ed with kTypeLevelMask, which is a 32 bit quantity. How could that be non-zero? Later, David --------------------------------------------------------------------- Check out my weekly web comic: http://www.techmagazine.org

1 0

Heathrow interrupt weirdness
by Benjamin Herrenschmidt 10 Dec '01

10 Dec '01

Hi ! While looking at the Heathrow.cpp code to compare with what I'm doing in Linux (chasing a lost interrupt problem with the PMU that occurs occasionally), I found something weird in the Darwin implementation. It defines interrupts matching the mask 0x1ff00000 of the first register set (the one at macio_base + 0x2n) as beeing level interrupts, all others (including the second register set at + 0x1n) as beeing edge interrupts. The result is that level interrupts end up beeing interrupt number 20 to 28, all other beeing edge triggered. Are you sure this is right ? That gives on a wallstreet the NMI & cascade (second heathrow) beeing edge (while the NMI is specifically level triggered on openpic machines and that makes more sense to have it level triggered), while all others beeing edge (especially IDE disk interrupts which are usually known to be level interrupts as the disk assert them until the status register is read) There's also a typo in the code for handling lost interrupts which uses the first level register in the code for checking the second register set (I just posted a bug report about this). Finally, the code og getVectorType() is confusing. It should probably test the vector is in the range of the first register set before doing (1 << vectorNumber) for completeness. The current code works as long as (1 << X) with X >= 32 returns 0 but I did see compilers optimizing this badly, so I feel this is dangerous coding. Regards, Ben.

1 0

Re: How to disable power down of PCI adapter?
by Shawn Erickson 07 Dec '01

07 Dec '01

I now have time to revisit this issue and have a few questions. First off the card in question is a dual ported adapter which has a PCI Bridge on board. So some issue may exist in the PCI Bridge code when you have two bridges back-to-back. I was reading up on power management (in I/O Kit Fundamentals) and it looks like in theory I would only need to implement a power-controller in my driver object (by overriding a few IOService methods) if I wanted to support power management. In the long run I do... but if I don't should the system not allow power down of my adapter? Seems dangerous to me (as you implied it shouldn't and we may be seeing a bug). Anyway, my driver is currently a sub-class of IOSCSIParallelController. IOSCSIParallelController doesn't appear to call PMinit in its start method and its super class IOService (naturally) doesn't call it. So if I understand things correctly my driver doesn't have a power policy-maker. Is this correct? or does it fall back to some higher/different object in this case My device is a PCI device and attaches to a PCIDevice Nub, does that connect it with a power policy-maker somehow? To enable power management in my driver do I need to implement both the power policy-maker and power-controller functionality? Or just power-controller functionality and some other object will manage policy for me (most likely dumbly because lack of particular knowledge my adapter). In the short term can I do something to block the system from powering off my adapter out from under me? This would make some early users very happy. I guess I am a little lost on this... any help would be greatly appreciated. -Shawn On Monday, October 1, 2001, at 07:03 PM, Shawn Erickson wrote: It is a QLogic 2202 64Bit/33Mhz PCI to Fibre Channel adapter. I see this problem on my test system with is a Dual 450 G4... the first of the G4 dual models. -Shawn On Monday, October 1, 2001, at 05:16 PM, Dave Suurballe wrote: You wrote: Started to test my PCI adapter driver and noticed that the system could be put to sleep and my adapters power would be shut off. On system wake the power comes back on but my driver is off in the weeds because it didn't know that the power was going to be shut off. Anyway... In the short term how can I ALWAYS veto the power down of my adapter? In the long term any pointers on how to get plugged into the power management system so I can put my driver in a safe state to allow power down of the adapter. Also how to a get up an going again at system wake? I looked over the PM related method in IOService but I am still lost at the moment... help :( -Shawn p.s. My driver is a sub-class of IOSCSIParallelController and drives a PCI based Fibre Channel adapter. I am testing Mac OS X build 5G64. 1. I don't know why the system slept; perhaps the card claims it's well-behaved with respect to power management. Still, having a non-power-managed driver should have prevented sleep. 2. There is no way to veto the power down of your adapter. However, the system should realize that there's a problem and not power down the PCI bus. I don't know why that didn't happen. 3. It isn't trivial to add power-management to your driver, but I can make it sound simple. The driver makes a couple calls to something, maybe itself. The driver implements an array which describes the power states of the device, minimally an on state and an off state. The driver implements a method that some object will call when it is time for your device to change power state. It will be told to turn off at sleep time, and it can save state, etc and turn off. At wake time it will be told to turn back on, and it can restore state, etc. The complexities are what to do if you get i/o requests while the device is off or is in the process of being turned off, what to do with i/o requests that are in progress when you are told to turn off, etc. There is a sample driver on Darwin, I think in a project called IOKitExamples or something like that. It's the networking driver. What is the PCI card? What make/model, etc? Dave Suurballe

1 0

Re: IOLocks deprecated ???
by Shawn Erickson 04 Dec '01

04 Dec '01

On Tuesday, December 4, 2001, at 04:58 AM, Francis Bouchard wrote: In IOLocks.h I can see this: #ifndef IOKIT_DEPRECATED #define IOKIT_DEPRECATED 1 #endif Does that means I shouldn't use it ? Yes... but it applies only to the stuff wrapped with IOKIT_DEPRECATED. Look around in the IOLocks.h header. Only the following are deprecated, everything else you should be able to use (IOSyncer and others heavily utilize the stuff defined in IOLocks). #if IOKIT_DEPRECATED /* The following API is deprecated */ typedef enum { kIOLockStateUnlocked = 0, kIOLockStateLocked = 1, } IOLockState; void IOLockInitWithState( IOLock * lock, IOLockState state); #define IOLockInit( l ) IOLockInitWithState( l, kIOLockStateUnlocked); static __inline__ void IOTakeLock( IOLock * lock) { IOLockLock(lock); } static __inline__ boolean_t IOTryLock( IOLock * lock) { return(IOLockTryLock(lock)); } static __inline__ void IOUnlock( IOLock * lock) { IOLockUnlock(lock); } #endif /* IOKIT_DEPRECATED */ #if IOKIT_DEPRECATED /* The following API is deprecated */ static __inline__ void IOReadLock( IORWLock * lock) { IORWLockRead(lock); } static __inline__ void IOWriteLock( IORWLock * lock) { IORWLockWrite(lock); } static __inline__ void IORWUnlock( IORWLock * lock) { IORWLockUnlock(lock); } #endif /* IOKIT_DEPRECATED */

1 0

Re: Multiple OS's
by Eric Hall 04 Dec '01

04 Dec '01

On Mon, Dec 03, 2001 at 12:08:40AM -0500, Rob Latham wrote: > Eduardo I. Jim?nez (eijimenez(a)mac.com) said: > > > Hello again. > > > > I have another question. Is it possible to have Classic, OS X, Darwin > > and Linux on the same machine?. I would really like to make some > > comparisons among OS's to try to bring the best of them to Darwin. > > yep it's possible. you can add netbsd to the mix too. i've done it. > and i'm just about to post the results :> > [snip] And OpenBSD too. I'm interested in the results as well, should be interesting to see how the various OSes compare in various areas and how that relates to their development lifetime on the platform. -eric

1 0

Re: [SOLVED] Placement new operator
by Francis Bouchard 04 Dec '01

04 Dec '01

I thought this new was provided by the language... Sorry I created the missing new operator. Francis Francis Bouchard wrote something like : > Hi, I'm trying to mage a generic class and I would like to use the placement > new operator. Is this allowed ? This would allow me to make a little > template that would take a class that woud do the allocation (using > different IOMalloc method). > > Now creating an object would look like this: > void* myPreAllocatedMemory = myAllocator.Allocate(sizeof(CMyCustomClass)); > CMyCustomClass* pCustomClass; > pCustomClass = new(myPreAllocatedMemory) CMyCustomClass(initData); > > This gives me the error: > too many arguments to function `void * operator new(long unsigned int)'

1 0

Placement new operator
by Francis Bouchard 04 Dec '01

04 Dec '01

Hi, I'm trying to mage a generic class and I would like to use the placement new operator. Is this allowed ? This would allow me to make a little template that would take a class that woud do the allocation (using different IOMalloc method). Now creating an object would look like this: void* myPreAllocatedMemory = myAllocator.Allocate(sizeof(CMyCustomClass)); CMyCustomClass* pCustomClass; pCustomClass = new(myPreAllocatedMemory) CMyCustomClass(initData); This gives me the error: too many arguments to function `void * operator new(long unsigned int)' Francis

1 0

IOLocks deprecated ???
by Francis Bouchard 04 Dec '01

04 Dec '01

In IOLocks.h I can see this: #ifndef IOKIT_DEPRECATED #define IOKIT_DEPRECATED 1 #endif Does that means I shouldn't use it ? If so, is there any API that let me protect a small peace of code (let's say the cay critical section works or the posix sandard). Oh please, telling me to use the commandGate won't help me at all. I NEED this funcionnality. I don't want to enter in the details, but There must be something that does what I need. Francis

1 0

Re: lmbench numbers ( for real )
by Rob Latham 03 Dec '01

03 Dec '01

Here are the real numbers: -- Rob Latham Allentown, PA USA EAE8 DE90 85BB 526F 3181 1FCF 51C4 B6CB 08CC 0897 cd results && make summary percent 2>/dev/null | more make[1]: Entering directory `/home/rob/src/LMbench/results' L M B E N C H 2 . 0 S U M M A R Y ------------------------------------ Basic system parameters ---------------------------------------------------- Host OS Description Mhz --------- ------------- ----------------------- ---- localhost Darwin 1.3.7 powerpc-apple-darwin1.3 397 localhost Darwin 1.3.7 powerpc-apple-darwin1.3 397 localhost Darwin 1.3.7 powerpc-apple-darwin1.3 397 dyn168.pl Darwin 1.4 powerpc-apple-darwin1.4 399 dyn168.pl Darwin 1.4 powerpc-apple-darwin1.4 399 dyn168.pl Darwin 1.4 powerpc-apple-darwin1.4 399 localhost Darwin 1.4 powerpc-apple-darwin1.4 400 localhost Darwin 5.1 powerpc-apple-darwin5.1 399 localhost Darwin 5.1 powerpc-apple-darwin5.1 399 localhost Darwin 5.1 powerpc-apple-darwin5.1 399 aragorn NetBSD 1.5Y powerpc-apple-netbsd1.5 400 aragorn NetBSD 1.5Y powerpc-apple-netbsd1.5 400 aragorn NetBSD 1.5Y powerpc-apple-netbsd1.5 400 aragorn Linux 2.4.9-b powerpc-linux-gnu 399 aragorn Linux 2.4.9-b powerpc-linux-gnu 399 aragorn Linux 2.4.13- powerpc-linux-gnu 399 aragorn Linux 2.4.13- powerpc-linux-gnu 399 Processor, Processes - times in microseconds - smaller is better ---------------------------------------------------------------- Host OS Mhz null null open selct sig sig fork exec sh call I/O stat clos TCP inst hndl proc proc proc --------- ------------- ---- ---- ---- ---- ---- ----- ---- ---- ---- ---- ---- localhost Darwin 1.3.7 397 3.38 5.56 35.2 49.6 37.7 13.8 19.3 2901 11.K 28.K localhost Darwin 1.3.7 397 3.48 5.71 35.7 50.0 13.9 19.3 3038 12.K 28.K localhost Darwin 1.3.7 397 3.40 5.58 35.2 49.6 13.8 19.1 3034 12.K 28.K dyn168.pl Darwin 1.4 399 3.13 4.73 22.4 28.6 56.2 5.48 15.5 2268 6394 18.K dyn168.pl Darwin 1.4 399 3.14 4.74 22.2 28.6 56.5 5.63 15.4 2294 6376 19.K dyn168.pl Darwin 1.4 399 3.13 4.75 22.2 28.6 56.7 5.51 15.4 2253 6367 19.K localhost Darwin 1.4 400 3.13 4.79 22.5 28.8 56.5 5.55 15.4 2226 6252 18.K localhost Darwin 5.1 399 3.16 4.78 22.6 27.7 56.4 5.52 15.8 2199 6141 18.K localhost Darwin 5.1 399 3.16 4.77 23.0 28.4 56.6 5.59 15.7 2162 6209 18.K localhost Darwin 5.1 399 3.17 4.75 22.9 28.1 56.6 5.51 15.5 2157 6125 18.K aragorn NetBSD 1.5Y 400 0.99 3.32 12.6 15.3 54.3 2.20 5.25 1801 6475 14.K aragorn NetBSD 1.5Y 400 0.99 3.32 12.6 15.4 54.3 2.20 5.24 1803 6409 14.K aragorn NetBSD 1.5Y 400 0.99 3.31 12.6 15.3 54.3 2.21 5.22 1798 6408 14.K aragorn Linux 2.4.9-b 399 0.52 0.89 3.79 5.23 43.7 1.86 5.08 654. 2965 16.K aragorn Linux 2.4.9-b 399 0.52 0.89 3.79 5.23 43.8 1.86 5.03 651. 2889 16.K aragorn Linux 2.4.13- 399 0.55 0.91 3.96 5.47 43.5 1.88 5.08 634. 2801 13.K aragorn Linux 2.4.13- 399 0.55 0.91 3.99 5.49 43.5 1.88 5.08 643. 2880 13.K Context switching - times in microseconds - smaller is better ------------------------------------------------------------- Host OS 2p/0K 2p/16K 2p/64K 8p/16K 8p/64K 16p/16K 16p/64K ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw ctxsw --------- ------------- ----- ------ ------ ------ ------ ------- ------- localhost Darwin 1.3.7 12.0 33.7 105.8 47.4 198.4 55.8 293.3 localhost Darwin 1.3.7 12.3 32.5 111.4 46.6 186.8 53.9 267.5 localhost Darwin 1.3.7 12.0 33.2 106.7 49.5 169.3 62.2 283.5 dyn168.pl Darwin 1.4 10.7 30.1 104.8 42.4 203.4 54.8 289.8 dyn168.pl Darwin 1.4 10.9 30.2 105.7 39.7 179.2 53.9 282.6 dyn168.pl Darwin 1.4 10.8 30.6 108.4 41.5 175.2 47.5 272.2 localhost Darwin 1.4 10.9 30.9 106.8 39.7 189.3 45.8 273.0 localhost Darwin 5.1 9.700 28.1 101.5 46.4 199.3 47.8 286.6 localhost Darwin 5.1 9.720 29.2 102.9 40.7 178.1 48.8 274.8 localhost Darwin 5.1 9.790 29.1 104.9 39.0 182.4 51.1 285.6 aragorn NetBSD 1.5Y 2.500 12.3 88.5 26.8 98.9 27.8 221.0 aragorn NetBSD 1.5Y 2.560 13.6 88.4 29.4 122.3 33.2 224.1 aragorn NetBSD 1.5Y 2.590 14.9 88.6 30.7 130.8 31.2 219.2 aragorn Linux 2.4.9-b 1.120 7.5500 85.3 24.1 143.8 26.3 219.6 aragorn Linux 2.4.9-b 1.100 7.4900 85.0 25.1 153.3 32.7 232.7 aragorn Linux 2.4.13- 1.130 8.1300 85.0 23.8 134.7 24.9 211.2 aragorn Linux 2.4.13- 1.140 7.9900 96.7 23.8 139.7 27.7 223.7 *Local* Communication latencies in microseconds - smaller is better ------------------------------------------------------------------- Host OS 2p/0K Pipe AF UDP RPC/ TCP RPC/ TCP ctxsw UNIX UDP TCP conn --------- ------------- ----- ----- ---- ----- ----- ----- ----- ---- localhost Darwin 1.3.7 12.0 59.0 59.6 131.0 192.8 154.6 231.9 369. localhost Darwin 1.3.7 12.3 58.9 59.3 130.9 192.7 153.5 233.4 357. localhost Darwin 1.3.7 12.0 59.4 59.0 130.7 191.0 155.4 232.2 358. dyn168.pl Darwin 1.4 10.7 54.9 54.0 121.3 140.0 372. dyn168.pl Darwin 1.4 10.9 55.3 53.0 119.8 142.5 355. dyn168.pl Darwin 1.4 10.8 54.9 53.6 121.8 142.1 357. localhost Darwin 1.4 10.9 53.4 54.1 121.3 140.9 362. localhost Darwin 5.1 9.700 53.1 50.2 119.2 140.8 366. localhost Darwin 5.1 9.720 52.7 49.9 117.8 140.3 345. localhost Darwin 5.1 9.790 51.0 50.2 118.6 141.9 357. aragorn NetBSD 1.5Y 2.500 20.8 20.4 47.2 59.7 286. aragorn NetBSD 1.5Y 2.560 20.4 20.4 46.8 59.4 285. aragorn NetBSD 1.5Y 2.590 20.4 20.4 47.0 59.3 284. aragorn Linux 2.4.9-b 1.120 7.954 13.9 39.0 168. aragorn Linux 2.4.9-b 1.100 7.998 13.7 38.6 168. aragorn Linux 2.4.13- 1.130 8.097 14.0 23.7 71.9 37.7 105.7 176. aragorn Linux 2.4.13- 1.140 8.154 13.9 23.7 71.0 37.7 107.1 173. File & VM system latencies in microseconds - smaller is better -------------------------------------------------------------- Host OS 0K File 10K File Mmap Prot Page Create Delete Create Delete Latency Fault Fault --------- ------------- ------ ------ ------ ------ ------- ----- ----- localhost Darwin 1.3.7 407.8 451.9 2352.9 561.8 5202.0 76.6 12.3K localhost Darwin 1.3.7 374.5 452.3 1938.0 513.6 5155.0 77.0 14.1K localhost Darwin 1.3.7 442.7 496.5 2012.1 599.9 5257.0 76.1 7270.0 dyn168.pl Darwin 1.4 520.0 572.4 2222.2 666.7 3270.0 63.3 15.4K dyn168.pl Darwin 1.4 425.2 619.2 2331.0 664.0 3281.0 56.1 17.0K dyn168.pl Darwin 1.4 427.2 596.7 2551.0 712.8 3251.0 56.1 16.0K localhost Darwin 1.4 363.8 434.6 3610.1 1025.6 3466.0 57.7 16.0K localhost Darwin 5.1 451.7 506.3 2739.7 673.4 3337.0 59.0 15.7K localhost Darwin 5.1 439.2 529.9 2747.3 643.5 3254.0 55.5 17.4K localhost Darwin 5.1 438.2 544.7 2314.8 619.6 3249.0 55.5 17.0K aragorn NetBSD 1.5Y 809.7 778.2 1757.5 1533.7 0.956 1076.0 aragorn NetBSD 1.5Y 1838.2 741.3 1776.2 1560.1 0.938 673.0 aragorn NetBSD 1.5Y 1841.6 755.3 1792.1 5464.5 0.954 901.0 aragorn Linux 2.4.9-b 36.9 8.6300 105.6 22.0 578.0 1.345 13.0 aragorn Linux 2.4.9-b 36.9 8.6200 107.3 22.1 591.0 1.340 13.0 aragorn Linux 2.4.13- 37.8 8.5220 105.3 23.1 632.0 1.193 13.0 aragorn Linux 2.4.13- 37.9 8.6200 105.0 23.0 622.0 1.204 13.0K *Local* Communication bandwidths in MB/s - bigger is better ----------------------------------------------------------- Host OS Pipe AF TCP File Mmap Bcopy Bcopy Mem Mem UNIX reread reread (libc) (hand) read write --------- ------------- ---- ---- ---- ------ ------ ------ ------ ---- ----- localhost Darwin 1.3.7 104. 97.8 75.6 158.2 179.4 269.8 126.1 179. 434.1 localhost Darwin 1.3.7 103. 100. 74.8 155.4 180.2 270.8 126.8 179. 434.8 localhost Darwin 1.3.7 104. 102. 74.2 157.2 180.4 266.5 125.2 179. 436.5 dyn168.pl Darwin 1.4 108. 107. 76.1 158.3 180.8 249.3 126.9 180. 438.7 dyn168.pl Darwin 1.4 108. 108. 75.9 161.9 180.9 247.0 127.3 181. 437.4 dyn168.pl Darwin 1.4 109. 108. 75.0 157.7 180.8 248.4 127.2 180. 438.0 localhost Darwin 1.4 108. 110. 76.9 162.4 180.9 246.9 126.5 180. 434.8 localhost Darwin 5.1 108. 111. 75.2 160.7 178.3 243.5 124.5 178. 431.2 localhost Darwin 5.1 108. 108. 74.2 159.2 178.8 246.0 126.2 178. 432.8 localhost Darwin 5.1 109. 111. 70.7 157.1 178.1 245.0 125.4 178. 431.1 aragorn NetBSD 1.5Y 126. 122. 56.0 71.5 177.6 115.7 113.3 177. 433.1 aragorn NetBSD 1.5Y 126. 123. 55.8 72.6 179.9 115.1 115.6 179. 430.4 aragorn NetBSD 1.5Y 127. 121. 58.1 72.3 179.8 116.9 120.7 179. 431.3 aragorn Linux 2.4.9-b 206. 99.2 130. 215.6 180.7 126.6 126.9 180. 385.8 aragorn Linux 2.4.9-b 206. 204. 141. 216.0 179.8 125.8 126.8 180. 231.1 aragorn Linux 2.4.13- 206. 211. 143. 216.1 180.1 126.4 127.4 180. 374.5 aragorn Linux 2.4.13- 206. 209. 118. 214.1 180.7 126.7 128.2 180. 385.4 Memory latencies in nanoseconds - smaller is better (WARNING - may not be correct, check graphs) --------------------------------------------------- Host OS Mhz L1 $ L2 $ Main mem Guesses --------- ------------- ---- ----- ------ -------- ------- localhost Darwin 1.3.7 397 5.015 30.3 128.3 localhost Darwin 1.3.7 397 5.025 30.2 128.1 localhost Darwin 1.3.7 397 5.025 36.0 128.2 dyn168.pl Darwin 1.4 399 5.012 30.2 128.0 dyn168.pl Darwin 1.4 399 5.019 30.3 128.4 dyn168.pl Darwin 1.4 399 5.019 30.2 127.8 localhost Darwin 1.4 400 5.014 30.2 127.8 localhost Darwin 5.1 399 5.012 30.2 129.8 localhost Darwin 5.1 399 5.019 30.2 129.8 localhost Darwin 5.1 399 5.019 30.3 129.7 aragorn NetBSD 1.5Y 400 5.002 30.1 132.8 aragorn NetBSD 1.5Y 400 5.002 30.1 129.7 aragorn NetBSD 1.5Y 400 5.001 30.1 129.7 aragorn Linux 2.4.9-b 399 5.007 30.0 127.9 aragorn Linux 2.4.9-b 399 5.008 30.0 127.9 aragorn Linux 2.4.13- 399 5.008 30.0 129.3 aragorn Linux 2.4.13- 399 5.007 30.1 128.8 make[1]: Leaving directory `/home/rob/src/LMbench/results'

1 0