Re: [PATCH v12 11/20] KVM: xen: allow shared_info to be mapped by fixed HVA

[David Woodhouse <dwmw2@infradead.org>]

[-- Attachment #1: Type: text/plain, Size: 1096 bytes --]

On Thu, 2024-02-08 at 16:51 +0000, Paul Durrant wrote:
> On 08/02/2024 16:48, Sean Christopherson wrote:
> > On Thu, Feb 08, 2024, Paul Durrant wrote:
> > > With the benefit of some sleep, I'm wondering why 0 is a 'ridiculous'
> > > invalid value for a *virtual* address? Surely it's essentially a numerical
> > > cast of the canonically invalid NULL pointer?
> > 
> > It's legal to mmap() virtual address '0', albeit not by default:

Well yes, to make dosemu work. But if you attempt to actually *do* that
in C code, the compiler itself doesn't cope...

$ cat foo.c
int foo(int *bar)
{
    if (bar)
        return 0;
    return *bar;
}
$ gcc -O2 -S -o- foo.c
...
foo:
.LFB0:
	.cfi_startproc
	endbr64
	testq	%rdi, %rdi
	je	.L4
	xorl	%eax, %eax
	ret
	.p2align 4,,10
	.p2align 3
.L4:
	movl	0, %eax
	ud2
	.cfi_endproc
.LFE0:
	.size	foo, .-foo

Note the ud2 instead of actually trying to dereference it.

Using anything except NULL as the "no value" value doesn't make sense
to me. It violates the principle of least surprise and would be a
really bad API.

[-- Attachment #2: smime.p7s --]
[-- Type: application/pkcs7-signature, Size: 5965 bytes --]