[Fwd: Can't activate swap partitions]
No one seemed to answer, so I try again, because I think we have a general problem with the "bit"-functions in include/asm-mips64/bitops.h. I don't think we can make the address 64-bit aligned in these functions, without hurting a lot of drivers which use these functions to operate on 32-bit aligned data.
One of the issues is that "integer" and "long integer" used to be 32-bit (in the 32-bit kernel), but now "long" is 64 bit.
I can see on the ia64 part, they already has taken care of this by saying that bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1). This is the beauty about little endian, the lower 32 bit are located on the same address no matter if you access the data as a 32-bit or 64-bit access. This of course doesn't count for big endian, so the ia64 approach can't be used on a big endian system.
Has anyone any ideas how to handle this without making a lot of changes in the general code and thereby hurting the other architectures.
/Carsten
-------- Original Message --------
Subject: Can't activate swap partitions
Date: Thu, 18 Jan 2001 16:30:15 +0100
From: Carsten Langgaard <[email protected]>
To: [email protected]
I'm having some problems with activating swap partitions when using a 64 bit kernel on a 32 bit userland.
I think I know what the problem is.
The structure of the swap devices is that the first 4096 bytes contains a bitmap. Bits that have been set indicate that the page of memory for which the number in the swap space matches the offset of the bit at the start of the space is available for paging.
The problem is then these bits are being checked, through the test_bit function call, where we read 64 bit at a time, and they where written 32 bit at a time (from the 32 bit kernel).
Note: we are talking about a bigendian system.
A little example to illustrate that I mean:
Set bit 0-53 to 1 and clear bit 54-63 (stored with 32 bit access) ADDR = 0xffffffff;
ADDR+4 = 0x003fffff;
If I read it as two 32 bit words I get the same result (bit 0-53 is set), but if I read it as one 64 bit double-word I get.
ADDR = 0xffffffff0003ffff;
Now bit 0-17 and bit 32-63 is set.
The bottom line is that I don't think we can make the address 64 bit aligned in the "set/clear-and-test" functions in include/asm-mips64/bitops.h, without hurting a lot of drivers which use these functions to operate on hw-defined data-structures.
/Carsten
--
_ _ ____ ___ Carsten Langgaard Mailto:[email protected] |\ /|||___)(___ MIPS Denmark Direct: +45 4486 5527 | \/ ||| ____) Lautrupvang 4B Switch: +45 4486 5555 TECHNOLOGIES 2750 Ballerup Fax...: +45 4486 5556 Denmark http://www.mips.com
Quoted 7 lines · Carsten Langgaard
No one seemed to answer, so I try again, because I think we have a general problem with the "bit"-functions in
include/asm-mips64/bitops.h. I don't think we can make the address 64-bit aligned in these functions, without hurting a lot of drivers which use these functions to operate on 32-bit aligned data.
One of the issues is that "integer" and "long integer" used to be 32-bit (in the 32-bit kernel), but now "long" is 64 bit.
The consensus about bitops is to always use `unsigned long', although IMHO a predefined type (bitops_t) would be better. On e.g. m68k a byte would be sufficient, assumed you don't need more than 8 bits. Or bitsops_t8 for minimum 8 bits, bitsops_t16 for minimum 16 bits, etc....
Quoted 9 lines
I can see on the ia64 part, they already has taken care of this by saying that bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1). This is the beauty about little endian, the lower 32 bit are located on the same address no matter if you access the data as a 32-bit or 64-bit access. This of course doesn't count for big endian, so the ia64 approach can't be used on a big endian system.
Has anyone any ideas how to handle this without making a lot of changes in the general code and thereby hurting the other architectures.
Your problem is special because you have a machine that can run in either 32-bit or 64-bit mode.
Quoted 11 lines
I'm having some problems with activating swap partitions when using a 64 bit kernel on a 32 bit userland.
I think I know what the problem is.
The structure of the swap devices is that the first 4096 bytes contains a bitmap. Bits that have been set indicate that the page of memory for which the number in the swap space matches the offset of the bit at the start of the space is available for paging.
The problem is then these bits are being checked, through thetest_bitfunction call, where we read 64 bit at a time, and they where written 32 bit at a time (from the 32 bit kernel).
Note: we are talking about a bigendian system.
I suggest to put an explicit `mkswap' command in the startup scripts. It's not needed to preserver the swap space contents between successive reboots (you can't switch from 32-bit to 64-bit or vice versa without rebooting, right?).
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven ------------- Sony Software Development Center Europe (SDCE) [email protected] ------------------- Sint-Stevens-Woluwestraat 55 Voice +32-2-7248626 Fax +32-2-7262686 ---------------- B-1130 Brussels, Belgium
Quoted 39 lines · Geert Uytterhoeven
Quoted 7 lines · Carsten Langgaard
No one seemed to answer, so I try again, because I think we have a general problem with the "bit"-functions in
include/asm-mips64/bitops.h. I don't think we can make the address 64-bit aligned in these functions, without hurting a lot of drivers which use these functions to operate on 32-bit aligned data.
One of the issues is that "integer" and "long integer" used to be 32-bit (in the 32-bit kernel), but now "long" is 64 bit.The consensus about bitops is to always use `unsigned long', although IMHO a predefined type (
bitops_t) would be better. On e.g. m68k a byte would be sufficient, assumed you don't need more than 8 bits. Orbitsops_t8for minimum 8 bits,bitsops_t16for minimum 16 bits, etc....Quoted 9 lines
I can see on the ia64 part, they already has taken care of this by saying that bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1). This is the beauty about little endian, the lower 32 bit are located on the same address no matter if you access the data as a 32-bit or 64-bit access. This of course doesn't count for big endian, so the ia64 approach can't be used on a big endian system.
Has anyone any ideas how to handle this without making a lot of changes in the general code and thereby hurting the other architectures.
Your problem is special because you have a machine that can run in either 32-bit or 64-bit mode.
Quoted 11 lines
I'm having some problems with activating swap partitions when using a 64 bit kernel on a 32 bit userland.
I think I know what the problem is.
The structure of the swap devices is that the first 4096 bytes contains a bitmap. Bits that have been set indicate that the page of memory for which the number in the swap space matches the offset of the bit at the start of the space is available for paging.
The problem is then these bits are being checked, through thetest_bitfunction call, where we read 64 bit at a time, and they where written 32 bit at a time (from the 32 bit kernel).
Note: we are talking about a bigendian system.
Quoted 3 lines
I suggest to put an explicit `mkswap' command in the startup scripts. It's not needed to preserver the swap space contents between successive reboots (you can't switch from 32-bit to 64-bit or vice versa without rebooting, right?).
That will probably solve my swap partition problem, but I think there are other potential problems as mention above.
I don't think we can guarantee that data always are 32-bit aligned.
Quoted 9 lines
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven ------------- Sony Software Development Center Europe (SDCE) [email protected] ------------------- Sint-Stevens-Woluwestraat 55 Voice +32-2-7248626 Fax +32-2-7262686 ---------------- B-1130 Brussels, Belgium
--
_ _ ____ ___ Carsten Langgaard Mailto:[email protected] |\ /|||___)(___ MIPS Denmark Direct: +45 4486 5527 | \/ ||| ____) Lautrupvang 4B Switch: +45 4486 5555 TECHNOLOGIES 2750 Ballerup Fax...: +45 4486 5556 Denmark http://www.mips.com
Quoted 39 lines · Geert Uytterhoeven
Quoted 7 lines · Carsten Langgaard
No one seemed to answer, so I try again, because I think we have a general problem with the "bit"-functions in
include/asm-mips64/bitops.h. I don't think we can make the address 64-bit aligned in these functions, without hurting a lot of drivers which use these functions to operate on 32-bit aligned data.
One of the issues is that "integer" and "long integer" used to be 32-bit (in the 32-bit kernel), but now "long" is 64 bit.The consensus about bitops is to always use `unsigned long', although IMHO a predefined type (
bitops_t) would be better. On e.g. m68k a byte would be sufficient, assumed you don't need more than 8 bits. Orbitsops_t8for minimum 8 bits,bitsops_t16for minimum 16 bits, etc....Quoted 9 lines
I can see on the ia64 part, they already has taken care of this by saying that bit 0 is the LSB of addr; bit 32 is the LSB of (addr+1). This is the beauty about little endian, the lower 32 bit are located on the same address no matter if you access the data as a 32-bit or 64-bit access. This of course doesn't count for big endian, so the ia64 approach can't be used on a big endian system.
Has anyone any ideas how to handle this without making a lot of changes in the general code and thereby hurting the other architectures.
Your problem is special because you have a machine that can run in either 32-bit or 64-bit mode.
Quoted 11 lines
I'm having some problems with activating swap partitions when using a 64 bit kernel on a 32 bit userland.
I think I know what the problem is.
The structure of the swap devices is that the first 4096 bytes contains a bitmap. Bits that have been set indicate that the page of memory for which the number in the swap space matches the offset of the bit at the start of the space is available for paging.
The problem is then these bits are being checked, through thetest_bitfunction call, where we read 64 bit at a time, and they where written 32 bit at a time (from the 32 bit kernel).
Note: we are talking about a bigendian system.
Quoted 4 lines
I suggest to put an explicit `mkswap' command in the startup scripts. It's not needed to preserver the swap space contents between successive reboots (you can't switch from 32-bit to 64-bit or vice versa without rebooting, right?).
I tried to use mkswap, but it also fails:
mkswap /dev/hda2
unknown ioctl: 00001260
Quoted 9 lines
Gr{oetje,eeting}s,
Geert
--
Geert Uytterhoeven ------------- Sony Software Development Center Europe (SDCE) [email protected] ------------------- Sint-Stevens-Woluwestraat 55 Voice +32-2-7248626 Fax +32-2-7262686 ---------------- B-1130 Brussels, Belgium
--
_ _ ____ ___ Carsten Langgaard Mailto:[email protected] |\ /|||___)(___ MIPS Denmark Direct: +45 4486 5527 | \/ ||| ____) Lautrupvang 4B Switch: +45 4486 5555 TECHNOLOGIES 2750 Ballerup Fax...: +45 4486 5556 Denmark http://www.mips.com