Message-ID: <b206dcaeedbff450713d9da4f27b786cdea69815.camel@intel.com>
Date: Tue, 30 Sep 2025 17:00:24 +0000
From: "Edgecombe, Rick P" <rick.p.edgecombe@...el.com>
To: "kvm@...r.kernel.org" <kvm@...r.kernel.org>, "linux-coco@...ts.linux.dev"
	<linux-coco@...ts.linux.dev>, "Huang, Kai" <kai.huang@...el.com>, "Hansen,
 Dave" <dave.hansen@...el.com>, "Zhao, Yan Y" <yan.y.zhao@...el.com>,
	"dave.hansen@...ux.intel.com" <dave.hansen@...ux.intel.com>, "kas@...nel.org"
	<kas@...nel.org>, "seanjc@...gle.com" <seanjc@...gle.com>, "mingo@...hat.com"
	<mingo@...hat.com>, "linux-kernel@...r.kernel.org"
	<linux-kernel@...r.kernel.org>, "tglx@...utronix.de" <tglx@...utronix.de>,
	"Yamahata, Isaku" <isaku.yamahata@...el.com>, "pbonzini@...hat.com"
	<pbonzini@...hat.com>, "Annapurve, Vishal" <vannapurve@...gle.com>, "Gao,
 Chao" <chao.gao@...el.com>, "bp@...en8.de" <bp@...en8.de>, "x86@...nel.org"
	<x86@...nel.org>
CC: "kirill.shutemov@...ux.intel.com" <kirill.shutemov@...ux.intel.com>
Subject: Re: [PATCH v3 07/16] x86/virt/tdx: Add tdx_alloc/free_page() helpers

On Tue, 2025-09-30 at 08:25 -0700, Dave Hansen wrote:
> On 9/18/25 16:22, Rick Edgecombe wrote:
> ...
> > +/*
> > + * The TDX spec treats the registers like an array, as they are ordered
> > + * in the struct. The array size is limited by the number or registers,
> > + * so define the max size it could be for worst case allocations and sanity
> > + * checking.
> > + */
> > +#define MAX_DPAMT_ARG_SIZE (sizeof(struct tdx_module_args) - \
> > +			    offsetof(struct tdx_module_args, rdx))
> > +
> > +/*
> > + * Treat struct the registers like an array that starts at RDX, per
> > + * TDX spec. Do some sanitychecks, and return an indexable type.
> > + */
> > +static u64 *dpamt_args_array_ptr(struct tdx_module_args *args)
> > +{
> > +	WARN_ON_ONCE(tdx_dpamt_entry_pages() > MAX_DPAMT_ARG_SIZE);
> > +
> > +	/*
> > +	 * FORTIFY_SOUCE could inline this and complain when callers copy
> > +	 * across fields, which is exactly what this is supposed to be
> > +	 * used for. Obfuscate it.
> > +	 */
> > +	return (u64 *)((u8 *)args + offsetof(struct tdx_module_args, rdx));
> > +}
> 
> There are a lot of ways to to all of this jazz to alias an array over
> the top of a bunch of named structure fields. My worry about this
> approach is that it intentionally tries to hide the underlying type from
> the compiler.

I was feeling good about it until I had to add that FORTIFY_SOUCE fix and almost
scrapped it. I should have taken the compiler issues as a bigger hint.

> 
> It could be done with a bunch of union/struct voodoo like 'struct page':
> 
> struct tdx_module_args {
> 	u64 rcx;
> 	union {
> 		struct {
> 			u64 rdx;
> 			u64 r8;
> 			u64 r9;
> 			...
> 		};
> 		u64 array[FOO];
> 	};
> }
> 
> Or a separate structure:
> 
> struct tdx_module_array_args {
> 	u64 rcx;
> 	u64 array[FOO];
> };
> 
> So that you could do something simpler:
> 
> u64 *dpamt_args_array_ptr(struct tdx_module_args *args)
> {
> 	return ((struct tdx_module_array_args *)args)->array;
> }

I think the union would be better, otherwise it gets a bit hidden how tdxcall.S
is plucking things out of the args struct.

Yan was asking if we could reuse some of this for the demote SEAMCALL which also
does an args array, but at a different offset. It would be nice if this thing
supported creating array's at different parts of the args struct without
defining a bunch of extra structs? Would something a little more dynamic still
be clear enough?

struct tdx_module_array_args {
	union {
		struct tdx_module_args args;
		u64 args_array[sizeof(struct tdx_module_args) / sizeof(u64)];
	};
};


#define MAX_TDX_ARG_SIZE(reg) (sizeof(struct tdx_module_args) - \
			       offsetof(struct tdx_module_args, reg))

static u64 *dpamt_args_array_ptr(struct tdx_module_array_args *args)
{
	WARN_ON_ONCE(tdx_dpamt_entry_pages() > MAX_TDX_ARG_SIZE(rdx));

	return &args->args_array[offsetof(struct tdx_module_args, rdx)];
}

Then for huge pages there could be a:
static u64 *demote_args_array_ptr(struct tdx_module_array_args *args)
{
	WARN_ON_ONCE(tdx_dpamt_entry_pages() > MAX_TDX_ARG_SIZE(r12));

	return &args->args_array[offsetof(struct tdx_module_args, r12)];
}

> 
> Along with one of these somewhere:
> 
> BUILD_BUG_ON(sizeof(struct tdx_module_array_args) !=
> 	     sizeof(struct tdx_module_array));
> 
> I personally find the offsetof() tricks to be harder to follow than
> either of those.
> 
> > +static int alloc_pamt_array(u64 *pa_array)
> > +{
> > +	struct page *page;
> > +
> > +	for (int i = 0; i < tdx_dpamt_entry_pages(); i++) {
> > +		page = alloc_page(GFP_KERNEL);
> > +		if (!page)
> > +			return -ENOMEM;
> > +		pa_array[i] = page_to_phys(page);
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +static void free_pamt_array(u64 *pa_array)
> > +{
> > +	for (int i = 0; i < tdx_dpamt_entry_pages(); i++) {
> > +		if (!pa_array[i])
> > +			break;
> > +
> > +		reset_tdx_pages(pa_array[i], PAGE_SIZE);
> 
> One nit: this reset is unnecessary in the error cases here where the
> array never gets handed to the TDX module. Right?

If it was ever passed to the TDX module I'd think we should reset it, but not
all of the error cases do. Maybe add a comment like:

/*
 * Reset pages unconditionally to cover cases
 * where they were passed to the TDX module.
 */

> 
> > +		/*
> > +		 * It might have come from 'prealloc', but this is an error
> > +		 * path. Don't be fancy, just free them. TDH.PHYMEM.PAMT.ADD
> > +		 * only modifies RAX, so the encoded array is still in place.
> > +		 */
> > +		__free_page(phys_to_page(pa_array[i]));
> > +	}
> > +}
> > +
> > +/*
> > + * Add PAMT memory for the given HPA. Return's negative error code
> > + * for kernel side error conditions (-ENOMEM) and 1 for TDX Module
> > + * error. In the case of TDX module error, the return code is stored
> > + * in tdx_err.
> > + */
> > +static u64 tdh_phymem_pamt_add(unsigned long hpa, u64 *pamt_pa_array)
> > +{
> > +	struct tdx_module_args args = {
> > +		.rcx = hpa,
> > +	};
> > +	u64 *args_array = dpamt_args_array_ptr(&args);
> > +
> > +	WARN_ON_ONCE(!IS_ALIGNED(hpa & PAGE_MASK, PMD_SIZE));
> > +
> > +	/* Copy PAMT page PA's into the struct per the TDX ABI */
> > +	memcpy(args_array, pamt_pa_array,
> > +	       tdx_dpamt_entry_pages() * sizeof(*args_array));
> 
> This uses 'sizeof(*args_array)'.
> 
> > +	return seamcall(TDH_PHYMEM_PAMT_ADD, &args);
> > +}
> > +
> > +/* Remove PAMT memory for the given HPA */
> > +static u64 tdh_phymem_pamt_remove(unsigned long hpa, u64 *pamt_pa_array)
> > +{
> > +	struct tdx_module_args args = {
> > +		.rcx = hpa,
> > +	};
> > +	u64 *args_array = dpamt_args_array_ptr(&args);
> > +	u64 ret;
> > +
> > +	WARN_ON_ONCE(!IS_ALIGNED(hpa & PAGE_MASK, PMD_SIZE));
> > +
> > +	ret = seamcall_ret(TDH_PHYMEM_PAMT_REMOVE, &args);
> > +	if (ret)
> > +		return ret;
> > +
> > +	/* Copy PAMT page PA's out of the struct per the TDX ABI */
> > +	memcpy(pamt_pa_array, args_array,
> > +	       tdx_dpamt_entry_pages() * sizeof(u64));
> 
> While this one is sizeof(u64).
> 
> Could we make it consistent, please?

Oops, yes. Should standardize on sizeof(*args_array) I think.

> 
> 
> > +/* Serializes adding/removing PAMT memory */
> > +static DEFINE_SPINLOCK(pamt_lock);
> > +
> > +/* Bump PAMT refcount for the given page and allocate PAMT memory if needed */
> > +int tdx_pamt_get(struct page *page)
> > +{
> > +	unsigned long hpa = ALIGN_DOWN(page_to_phys(page), PMD_SIZE);
> > +	u64 pamt_pa_array[MAX_DPAMT_ARG_SIZE];
> > +	atomic_t *pamt_refcount;
> > +	u64 tdx_status;
> > +	int ret;
> > +
> > +	if (!tdx_supports_dynamic_pamt(&tdx_sysinfo))
> > +		return 0;
> > +
> > +	ret = alloc_pamt_array(pamt_pa_array);
> > +	if (ret)
> > +		return ret;
> > +
> > +	pamt_refcount = tdx_find_pamt_refcount(hpa);
> > +
> > +	scoped_guard(spinlock, &pamt_lock) {
> > +		if (atomic_read(pamt_refcount))
> > +			goto out_free;
> > +
> > +		tdx_status = tdh_phymem_pamt_add(hpa | TDX_PS_2M, pamt_pa_array);
> > +
> > +		if (IS_TDX_SUCCESS(tdx_status)) {
> > +			atomic_inc(pamt_refcount);
> > +		} else {
> > +			pr_err("TDH_PHYMEM_PAMT_ADD failed: %#llx\n", tdx_status);
> > +			goto out_free;
> > +		}
> 
> I'm feeling like the states here are under-commented.
> 
> 	1. PAMT already allocated
> 	2. 'pamt_pa_array' consumed, bump the refcount
> 	3. TDH_PHYMEM_PAMT_ADD failed
> 
> #1 and #3 need to free the allocation.
> 
> Could we add comments to that effect, please?

Ok, maybe it even deserves a little design paragraph comment. I'll have to work
a bit longer on that.

> 
> > +	}
> 
> This might get easier to read if the pr_err() gets dumped in
> tdh_phymem_pamt_add() instead.
> 
> > +	return ret;
> > +out_free:
> > +	free_pamt_array(pamt_pa_array);
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(tdx_pamt_get);
> > +
> > +/*
> > + * Drop PAMT refcount for the given page and free PAMT memory if it is no
> > + * longer needed.
> > + */
> > +void tdx_pamt_put(struct page *page)
> > +{
> > +	unsigned long hpa = ALIGN_DOWN(page_to_phys(page), PMD_SIZE);
> > +	u64 pamt_pa_array[MAX_DPAMT_ARG_SIZE];
> > +	atomic_t *pamt_refcount;
> > +	u64 tdx_status;
> > +
> > +	if (!tdx_supports_dynamic_pamt(&tdx_sysinfo))
> > +		return;
> > +
> > +	hpa = ALIGN_DOWN(hpa, PMD_SIZE);
> > +
> > +	pamt_refcount = tdx_find_pamt_refcount(hpa);
> > +
> > +	scoped_guard(spinlock, &pamt_lock) {
> > +		if (!atomic_read(pamt_refcount))
> > +			return;
> > +
> > +		tdx_status = tdh_phymem_pamt_remove(hpa | TDX_PS_2M, pamt_pa_array);
> > +
> > +		if (IS_TDX_SUCCESS(tdx_status)) {
> > +			atomic_dec(pamt_refcount);
> > +		} else {
> > +			pr_err("TDH_PHYMEM_PAMT_REMOVE failed: %#llx\n", tdx_status);
> > +			return;
> > +		}
> > +	}
> > +
> > +	free_pamt_array(pamt_pa_array);
> > +}
> > +EXPORT_SYMBOL_GPL(tdx_pamt_put);
> 
> It feels like there's some magic in terms of how the entire contents of
> pamt_pa_array[] get zeroed so that this ends up being safe.
> 
> Could that get commented, please?


Good point.

/*
 * pamt_pa_array is populated up to tdx_dpamt_entry_pages() by the TDX
 * module with pages, or remains zero inited. free_pamt_array() can
 * handle either case. Just pass it unconditionally.
 */

> 
> > +/* Allocate a page and make sure it is backed by PAMT memory */
> 
> This comment is giving the "what" but is weak on the "why". Could we add
> this?
> 
> 	This ensures that the page can be used as TDX private
> 	memory and obtain TDX protections.

It sounds like the comment was trying to be imperative. Is it not needed here?
Alternatively:

        Return a page that can be be used as TDX private memory
        and obtain TDX protections.

> 
> > +struct page *tdx_alloc_page(void)
> > +{
> > +	struct page *page;
> > +
> > +	page = alloc_page(GFP_KERNEL);
> > +	if (!page)
> > +		return NULL;
> > +
> > +	if (tdx_pamt_get(page)) {
> > +		__free_page(page);
> > +		return NULL;
> > +	}
> > +
> > +	return page;
> > +}
> > +EXPORT_SYMBOL_GPL(tdx_alloc_page);
> > +
> > +/* Free a page and release its PAMT memory */
> 
> Also:
> 
> 	After this, the page is can no longer be protected by TDX.
> 
> > +void tdx_free_page(struct page *page)
> > +{
> > +	if (!page)
> > +		return;
> > +
> > +	tdx_pamt_put(page);
> > +	__free_page(page);
> > +}
> > +EXPORT_SYMBOL_GPL(tdx_free_page);
> > diff --git a/arch/x86/virt/vmx/tdx/tdx.h b/arch/x86/virt/vmx/tdx/tdx.h
> > index 82bb82be8567..46c4214b79fb 100644
> > --- a/arch/x86/virt/vmx/tdx/tdx.h
> > +++ b/arch/x86/virt/vmx/tdx/tdx.h
> > @@ -46,6 +46,8 @@
> >  #define TDH_PHYMEM_PAGE_WBINVD		41
> >  #define TDH_VP_WR			43
> >  #define TDH_SYS_CONFIG			45
> > +#define TDH_PHYMEM_PAMT_ADD		58
> > +#define TDH_PHYMEM_PAMT_REMOVE		59
> >  
> >  /*
> >   * SEAMCALL leaf:
> 

Powered by blists - more mailing lists