mirror of https://github.com/clearlinux/graphene.git synced 2026-09-06 13:51:28 +00:00

Files

T

Michał Kowalczyk d53729b201 [Pal] Rework manifest loading

This is a major refactor of the way manifests are loaded and handled,
which will be followed by a complete rework of the loader code (which
will include e.g. centralized config).

Changes/fixes:
- Huge part of manifest handling was refactored and untangled.
- Starting without a manifest is now disallowed. This was actually
  accidentally broken for some time and no one complained. It also makes
  little sense in practice and in Graphene's overall design, e.g. it
  conflicts with protected argv.
- Now we only allow starting by giving the executable, not manifest (the
  magic resolution logic was removed).
- Now manifests are sent over pipes between parent and children, instead
  of children finding and loading them on their own. This is a
  preparation for the upcoming centralized manifests change.
- Previously manifests were parsed 2 times on Linux and 3 times on
  Linux-SGX (by untrusted PAL, trusted PAL and LibOS). This is now
  fixed.
- The common `pal_main()` now requires that the backend-specific PAL
  loader loads the manifest before calling it. SGX code already has to
  do it (for proper initialization), so let's unify this interface for
  all PALs.
- Fix for a PAL crash when manifest size was divisible by page size
  (sic!). NULL termination was missing, but most of the time the padding
  to page size saved Graphene from crashing.

2020-12-05 01:46:03 +01:00

1.9 KiB

Raw Permalink Blame History

PyTorch

This directory contains steps and artifacts to run a PyTorch sample workload on Graphene. Specifically, this example uses a pre-trained model for image classification. We tested this on Ubuntu 16.04 and 18.04. The sample uses the package versions of Python (3.5 and 3.6) coming with each distribution.

The workload reads an image from a file on disk input.jpg and runs the classifier to detect what object is present in the image. For the image shipping with the workload, the output should be

[('Labrador retriever', 41.58518600463867), ('golden retriever', 16.59165382385254), ('Saluki, gazelle hound', 16.286855697631836), ('whippet', 2.853914976119995), ('Ibizan hound, Ibizan Podenco', 2.3924756050109863)]

With high probability (41%) the classifier detected the image to contain a Labrador retriever.

Pre-requisites

The following steps should suffice to run the workload on a stock Ubuntu 16.04 and 18.04 installation.

sudo apt install libnss-mdns libnss-myhostname to install additional DNS-resolver libraries.
sudo apt install python3-pip lsb-release to install pip and lsb_release. The former is required to install additional Python packages while the latter is used by the Makefile.
pip3 install --user torchvision pillow to install the torchvision and pillow Python packages and their dependencies (usually in $HOME/.local). WARNING: This downloads several hundred megabytes of data!
make download_model to download and save the pre-trained model. WARNING: this downloads about 200MB of data!

Build

Run make to build the non-SGX version and make SGX=1 to build the SGX version.

Run

Execute any one of the following commands to run the workload

natively: python3 pytorchexample.py
Graphene w/o SGX: ./pal_loader ./python3 ./pytorchexample.py
Graphene with SGX: SGX=1 ./pal_loader ./python3 ./pytorchexample.py

1.9 KiB Raw Permalink Blame History

PyTorch

Pre-requisites

Build

Run

1.9 KiB

Raw Permalink Blame History