Re: [PATCH v13 06/13] x86/sgx: Detect Intel SGX

From: Jarkko Sakkinen
Date: Tue Aug 28 2018 - 04:28:21 EST


On Mon, Aug 27, 2018 at 12:53:59PM -0700, Dave Hansen wrote:
> > +config INTEL_SGX_CORE
> > + prompt "Intel SGX core functionality"
> > + def_bool n
> > + depends on X86_64 && CPU_SUP_INTEL
> > + help
> > + Intel Software Guard eXtensions (SGX) is a set of CPU instructions
> > + that allows ring 3 applications to create enclaves, private regions
> > + of memory that are protected, by hardware, from unauthorized access
> > + and/or modification.
>
> This is a bit comma-crazy. Also, considering some of our recent CVE
> fun, I'd probably not claim hardware protection. :)

Agreed :)

> Maybe:
>
> Intel Software Guard eXtensions (SGX) CPU feature that allows
> ring 3 applications to create enclaves: private regions
> of memory that are architecturally protected from unauthorized
> access and/or modification.

Yeah, looks way more better structured.

> > + This option enables kernel recognition of SGX, high-level management
> > + of the Enclave Page Cache (EPC), tracking and writing of SGX Launch
> > + Enclave Hash MSRs, and allows for virtualization of SGX via KVM. By
> > + iteslf, this option does not provide SGX support to userspace.
>
> itself
>
>
> > diff --git a/arch/x86/include/asm/sgx_pr.h b/arch/x86/include/asm/sgx_pr.h
> > new file mode 100644
> > index 000000000000..c68578127620
> > --- /dev/null
> > +++ b/arch/x86/include/asm/sgx_pr.h
> > @@ -0,0 +1,13 @@
> > +// SPDX-License-Identifier: (GPL-2.0 OR BSD-3-Clause)
> > +// Copyright(c) 2016-17 Intel Corporation.
> > +
> > +#ifndef _ASM_X86_SGX_PR_H
> > +#define _ASM_X86_SGX_PR_H
> > +
> > +#include <linux/printk.h>
> > +#include <linux/ratelimit.h>
> > +
> > +#undef pr_fmt
> > +#define pr_fmt(fmt) "sgx: " fmt
> > +
> > +#endif /* _ASM_X86_SGX_PR_H */
>
> I don't think this belongs in a generic header. Generally, we do the
> pr_fmt stuff in .c files, not in headers. If someone includes this
> header directly or indirectly, they'll get a big surprise.
>
> If you *must* have this in a .h file, put it in
> arch/x86/kernel/cpu/intel_sgx.h or something and #include "intel_sgx.h"
> in all the .c files where you want this.

I think for intel_sgx.c (the core part) we could just manually add the
"sgx:" prefix because there are only few log messages. I would move the
definition to drivers/platform/x86/intel_sgx/sgx.h because the prefix
makes sense for all .c files there AFAIK.

> > +static __init int sgx_init(void)
> > +{
> > + unsigned long fc;
> > +
> > + if (!boot_cpu_has(X86_FEATURE_SGX))
> > + return false;
> > +
> > + if (!boot_cpu_has(X86_FEATURE_SGX1))
> > + return false;
> > +
> > + rdmsrl(MSR_IA32_FEATURE_CONTROL, fc);
> > + if (!(fc & FEATURE_CONTROL_LOCKED)) {
> > + pr_info("IA32_FEATURE_CONTROL MSR is not locked\n");
> > + return false;
> > + }
>
> This is a rather crummy error message. Doesn't this keep sgx from
> initializing? Would something like this be more informative?
>
> pr_info("failed init: IA32_FEATURE_CONTROL MSR not locked\n");

What about:

pr_err(FW_BUG "IA32_FEATURE_CONTROL MSR not locked\n");

> > + if (!(fc & FEATURE_CONTROL_SGX_ENABLE)) {
> > + pr_info("disabled by the firmware\n");
> > + return false;
> > + }
> > +
> > + if (!(fc & FEATURE_CONTROL_SGX_LE_WR))
> > + pr_info("IA32_SGXLEPUBKEYHASHn MSRs are not writable\n");
>
> How about something that might help an end user? Perhaps:
>
> pr_warn("launch configuration not available\n");

I think this message is a false flag here in the first place as KVM does
not require writable MSRs. It really should be moved to the driver.

> > + sgx_enabled = true;
> > + sgx_lc_enabled = !!(fc & FEATURE_CONTROL_SGX_LE_WR);
> > + return 0;
> > +}
> > +
> > +arch_initcall(sgx_init);
> >
>
>

/Jarkko