Paging error.

[Shvets04]

I wrote simple paging according to this tutorial, and i have some trouble.
When i run my OS , the text on a screen begins to appear and disappear. I don't know what's going on, but if i don't call switch_page_directory() (see - https://github.com/s04v/locOS/blob/mast ... ing.c#L143). All works correctly.


What is the problem??

+ dump of switch_page_directory

Code: Select all

319:	55                   	push   %ebp
 31a:	89 e5                	mov    %esp,%ebp
 31c:	83 ec 10             	sub    $0x10,%esp
 31f:	8b 45 08             	mov    0x8(%ebp),%eax
 322:	a3 00 00 00 00       	mov    %eax,0x0
			323: R_386_32	current_directory
 327:	8b 45 08             	mov    0x8(%ebp),%eax
 32a:	05 00 10 00 00       	add    $0x1000,%eax
 32f:	0f 22 d8             	mov    %eax,%cr3
 332:	0f 20 c0             	mov    %cr0,%eax
 335:	89 45 fc             	mov    %eax,-0x4(%ebp)
 338:	81 4d fc 00 00 00 80 	orl    $0x80000000,-0x4(%ebp)
 33f:	8b 45 fc             	mov    -0x4(%ebp),%eax
 342:	0f 22 c0             	mov    %eax,%cr0
 345:	c9                   	leave  
 346:	c3                   	ret   

[deleted8917]

Maybe you should ¿check this? James Molloy's Tutorial Known Bugs

[Shvets04]

Maybe you should ¿check this? James Molloy's Tutorial Known Bugs

this doesn't help me.

[MichaelPetch]

Quick glance.If you run you kernel in BOCHSs and stop right after you enable paging and do info tab you get this:

Code: Select all

0x0000000000000000-0x0000000000008fff -> 0x000000000000-0x000000008fff
0x00000000003f0000-0x00000000003f0fff -> 0x000010006000-0x000010006fff
0x00000000003f1000-0x00000000003f1fff -> 0x000000020000-0x000000020fff

Those are your current mappings. Your code fails when it hits the leave instruction at the end of the function. The reason why? If you look at the register dump when it fails it says this:

Code: Select all

00017537854i[CPU0  ] | EAX=e0000011  EBX=00007d8f  ECX=00008000  EDX=00002adf
00017537854i[CPU0  ] | ESP=0008ffa4  EBP=0008ffb4  ESI=00007dca  EDI=0000ffac

What is important is that ESP is 0008ffb4 which is in memory you haven't mapped (the address may be slightly different because of your compiler), thus leave fails trying to access the stack. I highly recommend using BOCHs for debugging this kind of thing.

[Shvets04]

Quick glance.If you run you kernel in BOCHSs and stop right after you enable paging and do info tab you get this:

Code: Select all

0x0000000000000000-0x0000000000008fff -> 0x000000000000-0x000000008fff
0x00000000003f0000-0x00000000003f0fff -> 0x000010006000-0x000010006fff
0x00000000003f1000-0x00000000003f1fff -> 0x000000020000-0x000000020fff

Those are your current mappings. Your code fails when it hits the leave instruction at the end of the function. The reason why? If you look at the register dump when it fails it says this:

Code: Select all

00017537854i[CPU0  ] | EAX=e0000011  EBX=00007d8f  ECX=00008000  EDX=00002adf
00017537854i[CPU0  ] | ESP=0008ffa4  EBP=0008ffb4  ESI=00007dca  EDI=0000ffac

What is important is that ESP is 0008ffb4 which is in memory you haven't mapped (the address may be slightly different because of your compiler), thus leave fails trying to access the stack. I highly recommend using BOCHs for debugging this kind of thing.

I use QEMU.

[MichaelPetch]

I use QEMU too, but I will say this. BOCHs is a better tool IMHO for this kind of issue (paging). It took me a matter of seconds to identify this problem in BOCHs. Whether you run this in QEMU or BOCHs your issue is that the stack isn't in a region of memory you have mapped, thus it fails.

[Shvets04]

I use QEMU too, but I will say this. BOCHs is a better tool IMHO for this kind of issue (paging). It took me a matter of seconds to identify this problem in BOCHs. Whether you run this in QEMU or BOCHs your issue is that the stack isn't in a region of memory you have mapped, thus it fails.

What should i do to fix it?

[MichaelPetch]

You need to create a mapping that includes the region of memory where you have your stack. I also didn't look at your code to determine what mapping you were expecting.What you have seems a bit unusual. The question is - is the mapping I showed from BOCHs what you were expecting to be mapped? If not then you have an issue with mapping already. If it is as you expect then you have to add additional mapping for the stack area.The other alternative is to place the stack in a region of memory that is already mapped.

Since the value of ESP suggests you probably started having your stack grow down from 0x90000 you could start by mapping the 4KiB page at 0x8f000 (which includes 0x8f000 to 0x8ffff) into virtual memory. Just identity map it as a test (virtual address 0x8f000 mapped to physical address 0x8f000). You could also just identity map all the memory in the first megabyte.

[Shvets04]

You need to create a mapping that includes the region of memory where you have your stack. I also didn't look at your code to determine what mapping you were expecting.What you have seems a bit unusual. The question is - is the mapping I showed from BOCHs what you were expecting to be mapped? If not then you have an issue with mapping already. If it is as you expect then you have to add additional mapping for the stack area.The other alternative is to place the stack in a region of memory that is already mapped.

Since the value of ESP suggests you probably started having your stack grow down from 0x90000 you could start by mapping the 4KiB page at 0x8f000 (which includes 0x8f000 to 0x8ffff) into virtual memory. Just identity map it as a test (virtual address 0x8f000 mapped to physical address 0x8f000). You could also just identity map all the memory in the first megabyte.

I started VM at 0x100000. and it works, but when page fault happens I continuously receive messages about page fault.