| Commit message (Collapse) | Author | Age | Files | Lines |
|---|
| |
|
|
|
|
|
|
|
|
|
|
| |
Evaluated every variable that lives in .data (and globals in .rodata)
in the kernel modules, and constified/eliminated/localised them
appropriately. This means that all read-only data is now actually
read-only data, and, if possible, at file scope. A lot of previously-
global-symbols became inlinable (and inlined!) constants. Probably
not in a big Wowee Performance Moment, but hey.
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Ahelenia Ziemiańska <[email protected]>
Closes #12899
|
| |
|
|
|
|
| |
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Ahelenia Ziemiańska <[email protected]>
Closes #12844
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
While evaluating other assembler implementations it turns out that
the precomputed hash subkey tables vary in size, from 8*16 bytes
(avx2/avx512) up to 48*16 bytes (avx512-vaes), depending on the
implementation.
To be able to handle the size differences later, allocate
`gcm_Htable` dynamically rather then having a fixed size array, and
adapt consumers.
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Attila Fülöp <[email protected]>
Closes #11102
|
| |
|
|
|
|
|
|
|
|
| |
While preparing #9749 some .cfi_{start,end}proc directives
were missed. Add the missing ones.
See upstream https://github.com/openssl/openssl/commit/275a048f
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Attila Fülöp <[email protected]>
Closes #11101
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
There are a couple of x86_64 architectures which support all needed
features to make the accelerated GCM implementation work but the
MOVBE instruction. Those are mainly Intel Sandy- and Ivy-Bridge
and AMD Bulldozer, Piledriver, and Steamroller.
By using MOVBE only if available and replacing it with a MOV
followed by a BSWAP if not, those architectures now benefit from
the new GCM routines and performance is considerably better
compared to the original implementation.
Reviewed-by: Brian Behlendorf <[email protected]>
Reviewed-by: Adam D. Moss <[email protected]>
Signed-off-by: Attila Fülöp <[email protected]>
Followup #9749
Closes #10029
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently SIMD accelerated AES-GCM performance is limited by two
factors:
a. The need to disable preemption and interrupts and save the FPU
state before using it and to do the reverse when done. Due to the
way the code is organized (see (b) below) we have to pay this price
twice for each 16 byte GCM block processed.
b. Most processing is done in C, operating on single GCM blocks.
The use of SIMD instructions is limited to the AES encryption of the
counter block (AES-NI) and the Galois multiplication (PCLMULQDQ).
This leads to the FPU not being fully utilized for crypto
operations.
To solve (a) we do crypto processing in larger chunks while owning
the FPU. An `icp_gcm_avx_chunk_size` module parameter was introduced
to make this chunk size tweakable. It defaults to 32 KiB. This step
alone roughly doubles performance. (b) is tackled by porting and
using the highly optimized openssl AES-GCM assembler routines, which
do all the processing (CTR, AES, GMULT) in a single routine. Both
steps together result in up to 32x reduction of the time spend in
the en/decryption routines, leading up to approximately 12x
throughput increase for large (128 KiB) blocks.
Lastly, this commit changes the default encryption algorithm from
AES-CCM to AES-GCM when setting the `encryption=on` property.
Reviewed-By: Brian Behlendorf <[email protected]>
Reviewed-By: Jason King <[email protected]>
Reviewed-By: Tom Caputi <[email protected]>
Reviewed-By: Richard Laager <[email protected]>
Signed-off-by: Attila Fülöp <[email protected]>
Closes #9749
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
- Add two new module parameters to icp (icp_aes_impl, icp_gcm_impl)
that control the crypto implementation. At the moment there is a
choice between generic and aesni (on platforms that support it).
- This enables support for AES-NI and PCLMULQDQ-NI on AMD Family
15h (bulldozer) and newer CPUs (zen).
- Modify aes_key_t to track what implementation it was generated
with as key schedules generated with various implementations
are not necessarily interchangable.
Reviewed by: Gvozden Neskovic <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Reviewed-by: Tom Caputi <[email protected]>
Reviewed-by: Richard Laager <[email protected]>
Signed-off-by: Nathaniel R. Lewis <[email protected]>
Closes #7102
Closes #7103
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| |
Currently, the ICP contains accelerated assembly code to be
used specifically on CPUs with AES-NI enabled. This code
makes heavy use of the movaps instruction which assumes that
it will be provided aes keys that are 16 byte aligned. This
assumption seems to hold on Illumos, but on Linux some kernel
options such as 'slub_debug=P' will violate it. This patch
changes all instances of this instruction to movups which is
the same except that it can handle unaligned memory.
This patch also adds a few flags which were accidentally never
given to the assembly compiler, resulting in objtool warnings.
Reviewed by: Gvozden Neskovic <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Reviewed-by: Nathaniel R. Lewis <[email protected]>
Signed-off-by: Tom Caputi <[email protected]>
Closes #7065
Closes #7108
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
Properly annotate functions and data section so that objtool does not complain
when CONFIG_STACK_VALIDATION and CONFIG_FRAME_POINTER are enabled.
Pass KERNELCPPFLAGS to assembler.
Use kfpu_begin()/kfpu_end() to protect SIMD regions in Linux kernel.
Reviewed-by: Tom Caputi <[email protected]>
Reviewed-by: Brian Behlendorf <[email protected]>
Signed-off-by: Gvozden Neskovic <[email protected]>
Closes #5872
Closes #5041
|
| |
|
|
|
|
|
|
|
|
|
|
|
|
| |
If there is no explicit note in the .S files, the obj file will mark it
as requiring an executable stack. This is unneeded and causes issues on
hardened systems.
More info:
https://wiki.gentoo.org/wiki/Hardened/GNU_stack_quickstart
Signed-off-by: Jason Zaman <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Closes #4947
Closes #4962
|
|
|
A port of the Illumos Crypto Framework to a Linux kernel module (found
in module/icp). This is needed to do the actual encryption work. We cannot
use the Linux kernel's built in crypto api because it is only exported to
GPL-licensed modules. Having the ICP also means the crypto code can run on
any of the other kernels under OpenZFS. I ended up porting over most of the
internals of the framework, which means that porting over other API calls (if
we need them) should be fairly easy. Specifically, I have ported over the API
functions related to encryption, digests, macs, and crypto templates. The ICP
is able to use assembly-accelerated encryption on amd64 machines and AES-NI
instructions on Intel chips that support it. There are place-holder
directories for similar assembly optimizations for other architectures
(although they have not been written).
Signed-off-by: Tom Caputi <[email protected]>
Signed-off-by: Tony Hutter <[email protected]>
Signed-off-by: Brian Behlendorf <[email protected]>
Issue #4329
|
