Bug in EPC paging

Bug in EPC paging

[Jethro Beekman]

[-- Attachment #1: Type: text/plain, Size: 607 bytes --]

A user is reporting CPU soft lockup with Intel SDK version 2.1 and 2.5 
driver. This only happens when using SGX in conjunction with DPDK (Data 
Plane Development Kit) in the same process. The process is running two 
enclaves that together are bigger than the EPC.

Dmesg output:

2.1 driver: 
https://github.com/YangZhou1997/dpdk-sgx-crash/blob/master/dmesg-full.log

2.5 driver: 
https://github.com/YangZhou1997/dpdk-sgx-crash/blob/master/dmesg-full-2.5.0.log

Enclave memory layout: 
https://github.com/YangZhou1997/dpdk-sgx-crash/blob/master/sgxs.info

-- 
Jethro Beekman | Fortanix


[-- Attachment #2: S/MIME Cryptographic Signature --]
[-- Type: application/pkcs7-signature, Size: 3990 bytes --]

Re: Bug in EPC paging

[Sean Christopherson]

On Sat, Jul 27, 2019 at 11:16:01PM +0000, Jethro Beekman wrote:
> A user is reporting CPU soft lockup with Intel SDK version 2.1 and 2.5
> driver. This only happens when using SGX in conjunction with DPDK (Data
> Plane Development Kit) in the same process. The process is running two
> enclaves that together are bigger than the EPC.
> 
> Dmesg output:
> 
> 2.1 driver:
> https://github.com/YangZhou1997/dpdk-sgx-crash/blob/master/dmesg-full.log
> 
> 2.5 driver: https://github.com/YangZhou1997/dpdk-sgx-crash/blob/master/dmesg-full-2.5.0.log

The log from 2.5 doesn't show any SGX activity, rather it's generating
remote TLB flushes during CoW.  That would imply this is a generic remote
TLB flush (or SMP function call IPIs) issue and not an EPC bug.

> 
> Enclave memory layout:
> https://github.com/YangZhou1997/dpdk-sgx-crash/blob/master/sgxs.info
> 
> -- 
> Jethro Beekman | Fortanix
>