1.. SPDX-License-Identifier: GPL-2.0
2
3=====================
4AMD Memory Encryption
5=====================
6
7Secure Memory Encryption (SME) and Secure Encrypted Virtualization (SEV) are
8features found on AMD processors.
9
10SME provides the ability to mark individual pages of memory as encrypted using
11the standard x86 page tables.  A page that is marked encrypted will be
12automatically decrypted when read from DRAM and encrypted when written to
13DRAM.  SME can therefore be used to protect the contents of DRAM from physical
14attacks on the system.
15
16SEV enables running encrypted virtual machines (VMs) in which the code and data
17of the guest VM are secured so that a decrypted version is available only
18within the VM itself. SEV guest VMs have the concept of private and shared
19memory. Private memory is encrypted with the guest-specific key, while shared
20memory may be encrypted with hypervisor key. When SME is enabled, the hypervisor
21key is the same key which is used in SME.
22
23A page is encrypted when a page table entry has the encryption bit set (see
24below on how to determine its position).  The encryption bit can also be
25specified in the cr3 register, allowing the PGD table to be encrypted. Each
26successive level of page tables can also be encrypted by setting the encryption
27bit in the page table entry that points to the next table. This allows the full
28page table hierarchy to be encrypted. Note, this means that just because the
29encryption bit is set in cr3, doesn't imply the full hierarchy is encrypted.
30Each page table entry in the hierarchy needs to have the encryption bit set to
31achieve that. So, theoretically, you could have the encryption bit set in cr3
32so that the PGD is encrypted, but not set the encryption bit in the PGD entry
33for a PUD which results in the PUD pointed to by that entry to not be
34encrypted.
35
36When SEV is enabled, instruction pages and guest page tables are always treated
37as private. All the DMA operations inside the guest must be performed on shared
38memory. Since the memory encryption bit is controlled by the guest OS when it
39is operating in 64-bit or 32-bit PAE mode, in all other modes the SEV hardware
40forces the memory encryption bit to 1.
41
42Support for SME and SEV can be determined through the CPUID instruction. The
43CPUID function 0x8000001f reports information related to SME::
44
45	0x8000001f[eax]:
46		Bit[0] indicates support for SME
47		Bit[1] indicates support for SEV
48	0x8000001f[ebx]:
49		Bits[5:0]  pagetable bit number used to activate memory
50			   encryption
51		Bits[11:6] reduction in physical address space, in bits, when
52			   memory encryption is enabled (this only affects
53			   system physical addresses, not guest physical
54			   addresses)
55
56If support for SME is present, MSR 0xc00100010 (MSR_AMD64_SYSCFG) can be used to
57determine if SME is enabled and/or to enable memory encryption::
58
59	0xc0010010:
60		Bit[23]   0 = memory encryption features are disabled
61			  1 = memory encryption features are enabled
62
63If SEV is supported, MSR 0xc0010131 (MSR_AMD64_SEV) can be used to determine if
64SEV is active::
65
66	0xc0010131:
67		Bit[0]	  0 = memory encryption is not active
68			  1 = memory encryption is active
69
70Linux relies on BIOS to set this bit if BIOS has determined that the reduction
71in the physical address space as a result of enabling memory encryption (see
72CPUID information above) will not conflict with the address space resource
73requirements for the system.  If this bit is not set upon Linux startup then
74Linux itself will not set it and memory encryption will not be possible.
75
76The state of SME in the Linux kernel can be documented as follows:
77
78	- Supported:
79	  The CPU supports SME (determined through CPUID instruction).
80
81	- Enabled:
82	  Supported and bit 23 of MSR_AMD64_SYSCFG is set.
83
84	- Active:
85	  Supported, Enabled and the Linux kernel is actively applying
86	  the encryption bit to page table entries (the SME mask in the
87	  kernel is non-zero).
88
89SME can also be enabled and activated in the BIOS. If SME is enabled and
90activated in the BIOS, then all memory accesses will be encrypted and it
91will not be necessary to activate the Linux memory encryption support.
92
93If the BIOS merely enables SME (sets bit 23 of the MSR_AMD64_SYSCFG),
94then memory encryption can be enabled by supplying mem_encrypt=on on the
95kernel command line.  However, if BIOS does not enable SME, then Linux
96will not be able to activate memory encryption, even if configured to do
97so by default or the mem_encrypt=on command line parameter is specified.
98
99Secure Nested Paging (SNP)
100==========================
101
102SEV-SNP introduces new features (SEV_FEATURES[1:63]) which can be enabled
103by the hypervisor for security enhancements. Some of these features need
104guest side implementation to function correctly. The below table lists the
105expected guest behavior with various possible scenarios of guest/hypervisor
106SNP feature support.
107
108+-----------------+---------------+---------------+------------------+
109| Feature Enabled | Guest needs   | Guest has     | Guest boot       |
110| by the HV       | implementation| implementation| behaviour        |
111+=================+===============+===============+==================+
112|      No         |      No       |      No       |     Boot         |
113|                 |               |               |                  |
114+-----------------+---------------+---------------+------------------+
115|      No         |      Yes      |      No       |     Boot         |
116|                 |               |               |                  |
117+-----------------+---------------+---------------+------------------+
118|      No         |      Yes      |      Yes      |     Boot         |
119|                 |               |               |                  |
120+-----------------+---------------+---------------+------------------+
121|      Yes        |      No       |      No       | Boot with        |
122|                 |               |               | feature enabled  |
123+-----------------+---------------+---------------+------------------+
124|      Yes        |      Yes      |      No       | Graceful boot    |
125|                 |               |               | failure          |
126+-----------------+---------------+---------------+------------------+
127|      Yes        |      Yes      |      Yes      | Boot with        |
128|                 |               |               | feature enabled  |
129+-----------------+---------------+---------------+------------------+
130
131More details in AMD64 APM[1] Vol 2: 15.34.10 SEV_STATUS MSR
132
133Reverse Map Table (RMP)
134=======================
135
136The RMP is a structure in system memory that is used to ensure a one-to-one
137mapping between system physical addresses and guest physical addresses. Each
138page of memory that is potentially assignable to guests has one entry within
139the RMP.
140
141The RMP table can be either contiguous in memory or a collection of segments
142in memory.
143
144Contiguous RMP
145--------------
146
147Support for this form of the RMP is present when support for SEV-SNP is
148present, which can be determined using the CPUID instruction::
149
150	0x8000001f[eax]:
151		Bit[4] indicates support for SEV-SNP
152
153The location of the RMP is identified to the hardware through two MSRs::
154
155        0xc0010132 (RMP_BASE):
156                System physical address of the first byte of the RMP
157
158        0xc0010133 (RMP_END):
159                System physical address of the last byte of the RMP
160
161Hardware requires that RMP_BASE and (RPM_END + 1) be 8KB aligned, but SEV
162firmware increases the alignment requirement to require a 1MB alignment.
163
164The RMP consists of a 16KB region used for processor bookkeeping followed
165by the RMP entries, which are 16 bytes in size. The size of the RMP
166determines the range of physical memory that the hypervisor can assign to
167SEV-SNP guests. The RMP covers the system physical address from::
168
169        0 to ((RMP_END + 1 - RMP_BASE - 16KB) / 16B) x 4KB.
170
171The current Linux support relies on BIOS to allocate/reserve the memory for
172the RMP and to set RMP_BASE and RMP_END appropriately. Linux uses the MSR
173values to locate the RMP and determine the size of the RMP. The RMP must
174cover all of system memory in order for Linux to enable SEV-SNP.
175
176Segmented RMP
177-------------
178
179Segmented RMP support is a new way of representing the layout of an RMP.
180Initial RMP support required the RMP table to be contiguous in memory.
181RMP accesses from a NUMA node on which the RMP doesn't reside
182can take longer than accesses from a NUMA node on which the RMP resides.
183Segmented RMP support allows the RMP entries to be located on the same
184node as the memory the RMP is covering, potentially reducing latency
185associated with accessing an RMP entry associated with the memory. Each
186RMP segment covers a specific range of system physical addresses.
187
188Support for this form of the RMP can be determined using the CPUID
189instruction::
190
191        0x8000001f[eax]:
192                Bit[23] indicates support for segmented RMP
193
194If supported, segmented RMP attributes can be found using the CPUID
195instruction::
196
197        0x80000025[eax]:
198                Bits[5:0]  minimum supported RMP segment size
199                Bits[11:6] maximum supported RMP segment size
200
201        0x80000025[ebx]:
202                Bits[9:0]  number of cacheable RMP segment definitions
203                Bit[10]    indicates if the number of cacheable RMP segments
204                           is a hard limit
205
206To enable a segmented RMP, a new MSR is available::
207
208        0xc0010136 (RMP_CFG):
209                Bit[0]     indicates if segmented RMP is enabled
210                Bits[13:8] contains the size of memory covered by an RMP
211                           segment (expressed as a power of 2)
212
213The RMP segment size defined in the RMP_CFG MSR applies to all segments
214of the RMP. Therefore each RMP segment covers a specific range of system
215physical addresses. For example, if the RMP_CFG MSR value is 0x2401, then
216the RMP segment coverage value is 0x24 => 36, meaning the size of memory
217covered by an RMP segment is 64GB (1 << 36). So the first RMP segment
218covers physical addresses from 0 to 0xF_FFFF_FFFF, the second RMP segment
219covers physical addresses from 0x10_0000_0000 to 0x1F_FFFF_FFFF, etc.
220
221When a segmented RMP is enabled, RMP_BASE points to the RMP bookkeeping
222area as it does today (16K in size). However, instead of RMP entries
223beginning immediately after the bookkeeping area, there is a 4K RMP
224segment table (RST). Each entry in the RST is 8-bytes in size and represents
225an RMP segment::
226
227        Bits[19:0]  mapped size (in GB)
228                    The mapped size can be less than the defined segment size.
229                    A value of zero, indicates that no RMP exists for the range
230                    of system physical addresses associated with this segment.
231        Bits[51:20] segment physical address
232                    This address is left shift 20-bits (or just masked when
233                    read) to form the physical address of the segment (1MB
234                    alignment).
235
236The RST can hold 512 segment entries but can be limited in size to the number
237of cacheable RMP segments (CPUID 0x80000025_EBX[9:0]) if the number of cacheable
238RMP segments is a hard limit (CPUID 0x80000025_EBX[10]).
239
240The current Linux support relies on BIOS to allocate/reserve the memory for
241the segmented RMP (the bookkeeping area, RST, and all segments), build the RST
242and to set RMP_BASE, RMP_END, and RMP_CFG appropriately. Linux uses the MSR
243values to locate the RMP and determine the size and location of the RMP
244segments. The RMP must cover all of system memory in order for Linux to enable
245SEV-SNP.
246
247More details in the AMD64 APM Vol 2, section "15.36.3 Reverse Map Table",
248docID: 24593.
249
250Secure VM Service Module (SVSM)
251===============================
252
253SNP provides a feature called Virtual Machine Privilege Levels (VMPL) which
254defines four privilege levels at which guest software can run. The most
255privileged level is 0 and numerically higher numbers have lesser privileges.
256More details in the AMD64 APM Vol 2, section "15.35.7 Virtual Machine
257Privilege Levels", docID: 24593.
258
259When using that feature, different services can run at different protection
260levels, apart from the guest OS but still within the secure SNP environment.
261They can provide services to the guest, like a vTPM, for example.
262
263When a guest is not running at VMPL0, it needs to communicate with the software
264running at VMPL0 to perform privileged operations or to interact with secure
265services. An example fur such a privileged operation is PVALIDATE which is
266*required* to be executed at VMPL0.
267
268In this scenario, the software running at VMPL0 is usually called a Secure VM
269Service Module (SVSM). Discovery of an SVSM and the API used to communicate
270with it is documented in "Secure VM Service Module for SEV-SNP Guests", docID:
27158019.
272
273(Latest versions of the above-mentioned documents can be found by using
274a search engine like duckduckgo.com and typing in:
275
276  site:amd.com "Secure VM Service Module for SEV-SNP Guests", docID: 58019
277
278for example.)
279