problem with shmat () and > 1GB memory in kernel 2.2.17

problem with shmat () and > 1GB memory in kernel 2.2.17

[Rolf Fokkens]

> Hi!
> 
> Recently we installed extra memory in our Oracle-on-Linux database server,
> it now has 1.25 GB. I installed a 2.2.17 kernel with the 2GB option
> enabled. I rebooted the machine (a Compaq Proliant 5500 dual PII 450MHz)
> and noticed that one of the databases wasn't able to start. After
> installing a 2.2.17 kernel without the 2GB option averything worked fine.
> 
> As said, one of the databases didn't start, it was the one with SID
> "linux" (this may be relevant to Oracle insiders), the other six databases
> started OK. After some reboots it appeared that espcially this database
> wouldn't start. Not being able to look into the source I did an strace on
> the server process that got into trouble, the text was something like
> "Unable to attach shared memory, OS Error 22". The strace generated a log
> which showed this:
> 
> shmget(IPC_PRIVATE, 33554432, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 4872
> shmat(4872, 0x29400000, 0)              = -1 EINVAL (Invalid argument)
> 
> This confirmed the problem of not being able to attach the shared memory.
> I also did an strace on this process on the 2.2.17 kernel without the 2GB
> option enabled, and it showd:
> 
> shmget(IPC_PRIVATE, 33554432, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2568
> shmat(2568, 0x29400000, 0)              = 0x29400000
> 
> The failing shmat isn't the first one, but the ones before all succeed,
> some lines from the strace log:
> 
> shmget(IPC_PRIVATE, 49152, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 4865
> shmat(4865, 0x20000000, 0)              = 0x20000000
> shmget(IPC_PRIVATE, 26214400, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 4866
> shmat(4866, 0x20400000, 0)              = 0x20400000
> shmget(IPC_PRIVATE, 23068672, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 4867
> shmat(4867, 0x22000000, 0)              = 0x22000000
> shmget(IPC_PRIVATE, 19922944, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 4868
> shmat(4868, 0x23800000, 0)              = 0x23800000
> shmget(IPC_PRIVATE, 17825792, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 4869
> shmat(4869, 0x24c00000, 0)              = 0x24c00000
> shmget(IPC_PRIVATE, 29360128, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 4870
> shmat(4870, 0x26000000, 0)              = 0x26000000
> shmget(IPC_PRIVATE, 22020096, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 4871
> shmat(4871, 0x27c00000, 0)              = 0x27c00000
> shmget(IPC_PRIVATE, 33554432, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 4872
> shmat(4872, 0x29400000, 0)              = -1 EINVAL (Invalid argument)
> 
> On the 1GB kernel things go like this:
> 
> shmget(IPC_PRIVATE, 49152, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2561
> shmat(2561, 0x20000000, 0)              = 0x20000000
> shmget(IPC_PRIVATE, 26214400, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2562
> shmat(2562, 0x20400000, 0)              = 0x20400000
> shmget(IPC_PRIVATE, 23068672, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2563
> shmat(2563, 0x22000000, 0)              = 0x22000000
> shmget(IPC_PRIVATE, 19922944, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2564
> shmat(2564, 0x23800000, 0)              = 0x23800000
> shmget(IPC_PRIVATE, 17825792, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2565
> shmat(2565, 0x24c00000, 0)              = 0x24c00000
> shmget(IPC_PRIVATE, 29360128, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2566
> shmat(2566, 0x26000000, 0)              = 0x26000000
> shmget(IPC_PRIVATE, 22020096, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2567
> shmat(2567, 0x27c00000, 0)              = 0x27c00000
> shmget(IPC_PRIVATE, 33554432, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2568
> shmat(2568, 0x29400000, 0)              = 0x29400000
> shmget(IPC_PRIVATE, 32505856, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2569
> shmat(2569, 0x2b400000, 0)              = 0x2b400000
> shmget(3779348124, 13291520, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 2570
> shmat(2570, 0x2d400000, 0)              = 0x2d400000
> 
> The only thing I see is that shmaddr (shmat's second argument) climbs
> above 0x28000000. This is confirmed by another database that runs fine on
> the 2GB kernel:
> 
> shmget(IPC_PRIVATE, 49152, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 7303
> shmat(7303, 0x20000000, 0)              = 0x20000000
> shmget(IPC_PRIVATE, 20971520, IPC_CREAT|IPC_EXCL|0x1a0|0640) = 7304
> shmat(7304, 0x20400000, 0)              = 0x20400000
> shmget(2522853243, 0, 0)                = -1 ENOENT (No such file or
> directory)
> shmat(7305, 0x21800000, 0)              = 0x21800000
> shmat(7305, 0, 0)                       = 0x2ac4a000
> shmat(7305, 0x21800000, 0)              = 0x21800000
> shmat(7303, 0x20000000, 0)              = 0x20000000
> shmat(7304, 0x20400000, 0)              = 0x20400000
> 
> All shmaddr's are below 0x28000000 and everything is fine. This may be
> accidental, but it's all I see.
> 
> Because the machine is a production system I installed the 1GB 2.2.17
> kernel, but I saved the strace logs, so If any you want's then I can mail
> them.
> 
> Rolf
> 
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/

Re: problem with shmat () and > 1GB memory in kernel 2.2.17

[Christoph Rohland]

Hi Rolf,

On Wed, 8 Nov 2000, Rolf Fokkens wrote:
>> Recently we installed extra memory in our Oracle-on-Linux database
>> server, it now has 1.25 GB. I installed a 2.2.17 kernel with the
>> 2GB option enabled. I rebooted the machine (a Compaq Proliant 5500
>> dual PII 450MHz) and noticed that one of the databases wasn't able
>> to start. After installing a 2.2.17 kernel without the 2GB option
>> averything worked fine.

You should not use the 2GB option. Use the bigmem patch and you will
have 3GB userspace with 4GB main memory instead of 2GB user space with
2GB memory.

Greetings
		Christoph

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
Please read the FAQ at http://www.tux.org/lkml/