Failing to run a program due to: reason: eat_violation

[angerman]

So I'm trying to run a haskell program compiled against muslc through noah. Using --output, I see the following repeating forever:

[48232:36034161] reason: ept_violation
[48232:36034161] guest-physical address = 0x7fc0000000
[48232:36034161] exit qualification = 0x83
[48232:36034161] guest linear address = 0x22c039e
[48232:36034161] reason: ept_violation
[48232:36034161] guest-physical address = 0x7fc0000000
[48232:36034161] exit qualification = 0x83
[48232:36034161] guest linear address = 0x22c039e
[48232:36034161] reason: ept_violation
[48232:36034161] guest-physical address = 0x7fc0000000
[48232:36034161] exit qualification = 0x83
[48232:36034161] guest linear address = 0x22c039e
[48232:36034161] reason: ept_violation
[48232:36034161] guest-physical address = 0x7fc0000000
[48232:36034161] exit qualification = 0x83
[48232:36034161] guest linear address = 0x22c039e
[48232:36034161] reason: ept_violation
[48232:36034161] guest-physical address = 0x7fc0000000
[48232:36034161] exit qualification = 0x83
[48232:36034161] guest linear address = 0x22c039e

I don't mind getting into this and fixing it. What am I looking for?

[angerman]

Looking at the VM EXITS (27-10 Vol. 3C), Table 27-7. Exit Qualification for EPT Violations, we find:
that 0x83 which is

12 11 10  9  8    7  6  5  4  3  2  1  0
 0  0  0  0  0    1  0  0  0  0  0  1  1 

tells us that we have should probably look at note 1, which states:

If accessed and dirty flags for EPT are enabled, processor accesses to guest paging-structure entries are treated as writes with regard to EPT violations (see Section 28.2.3.2). If such an access causes an EPT violation, the processor sets both bit 0 and bit 1 of the exit qualification.

Looking at

noah/src/main.c

Lines 335 to 347 in b61c85c

	int verify = 0;
	if (qual & (1 << 0)) {
	verify = VERIFY_READ;
	} else if (qual & (1 << 1)) {
	verify = VERIFY_WRITE;
	} else if (qual & (1 << 2)) {
	verify = VERIFY_EXEC;
	}
	if (!addr_ok(gladdr, verify)) {
	printk("page fault: caused by guest linear address 0x%llx\n", gladdr);
	send_signal(getpid(), LINUX_SIGSEGV);
	}

it seems as if we assume that either bits 0, 1 or 2 are set. At least one is set and never more than two.

[angerman]

Upon further investigation I believe this might actually be the right approach, to verify READ, and silently ignore/discard the WRITE flag.

However a comment might be helpful. Also I wonder if one could simply set the dirty flag to off and thus reduce the number of times this code is called.

[roolebo]

@angerman What's the status of rflags immediately after VMLAUNCH/VMRESUME?