Signed-off-by: Angie Chinchilla <[email protected]>

Authors:
Serge Ayoun <[email protected]>
Angie Chinchilla <[email protected]>
Shay Katz-zamir <[email protected]>
Cedric Xing <[email protected]>

Signed-off-by: Angie Chinchilla <[email protected]>

Signed-off-by: Angie Chinchilla <[email protected]>

Signed-off-by: Angie Chinchilla <[email protected]>

Set sgx2 to master:<03435d33de0bcca6c5777f23ac161249b9158f1e>
for rebasing but keep the previous commit history/label.

Rebased 2.0 patch applies to linux-sgx-driver:master
commit-id 03435d3

Authors:
Serge Ayoun <[email protected]>
Angie Chinchilla <[email protected]>
Shay Katz-zamir <[email protected]>
Cedric Xing <[email protected]>

Signed-off-by: Angie Chinchilla <[email protected]>

Signed-off-by: Angie Chinchilla <[email protected]>

Sgx2

Signed-off-by: Zhang Lili [email protected]

Update README

Signed-off-by: Zhang Lili [email protected]

Signed-off-by: Zhang Lili [email protected]

Update README for SUSE on sgx2 branch

Signed-off-by: Zhang Lili [email protected]

Signed-off-by: Zhang Lili [email protected]

Port README.md update from master

Update README.md

Replace the call to mmu_notifier_unregister_no_release() with a call
to the basic mmu_notifier_unregister().  The no_release variant was
not added until kernel 3.17 and using it prevents building the driver
in Debian 8 (and earlier).

Back when usge of the MMU notifier was added to the SGX driver, the
MMU release callback, sgx_mmu_notifier_release(), included acquiring
mmap_sem for read, i.e. down_read(&mm->mmap_sem).  This caused a
deadlock if we called mmu_notifier_unregister() in sgx_encl_release()
because the semaphore is already write-locked by munmap().  Eventually
we stopped acquiring mmap_sem in sgx_mmu_notifier_release(), but kept
using the no_release unregister variant as invoking release on a dying
enclave was unnecessary and added an extra lock/unlock sequence.

TL;DR: Calling mmu_notifier_unregister_no_release() was necessary to
avoid deadlock in an old incarnation of the driver, but the current
driver plays nice with the release variant.

Signed-off-by: Sean Christopherson <[email protected]>

intel_sgx: use mmu_notifier_unregister w/o "no_release"

Signed-off-by: Serge Ayoun <[email protected]>

Removing conflicting compilation flags

…ap_sem) added

isolate_range does not always extract the right vma ptr: code fix

Signed-off-by: Serge Ayoun <[email protected]>

Calling PF handler from isolate_range was not protected: down_read(mm…

Signed-off-by: Angie Chinchilla <[email protected]>

Update License.txt to 3-c BSD/GPL v2, to match src

. Updating sgx_secs structure to reflect configid and configsvn members.
. Releasing SGX_ATTR_RESERVED bit mask 7.

Signed-off-by: Serge Ayoun <[email protected]>

In order to cope with KSS feature:

Signed-off-by: Zhang Lili Z <[email protected]>

Signed-off-by: Serge Ayoun <[email protected]>

vm_insert_pfn replaced by vmf_insert_pfn from kernel 4.20

Signed-off-by: Serge Ayoun <[email protected]>

Enclave virtual address range should never be defined as MAP_PRIVATE.

in success case

Signed-off-by: Serge Ayoun <[email protected]>

Kernel 4.20 and later: vmf_insert_pfn returns VM_FAULT_NOPAGE

For kernel v5.1.x, compiling the out-of-tree 'isgx' driver led to the
following error:
sgx_vma.c:241:11: error: initialization of 'vm_fault_t (*)(struct
vm_fault *)' {aka 'unsigned int (*)(struct vm_fault *)'} from
incompatible pointer type 'int (*)(struct vm_fault *)'
[-Werror=incompatible-pointer-types]

Hence, added return type of unsigned int to sgx_vma_fault for kernel
v5.1.

Signed-off-by: Iyer, Naveen <[email protected]>

sgx_vma: return unsigned int from sgx_vma_fault

Change propagated from master branch.

Signed-off-by: Serge Ayoun <[email protected]>

For kernel 5.1 and higher, sgx_vma_fault returns unsigned int.

Signed-off-by: Serge Ayoun <[email protected]>

Version number is 2.5.0

VM_FAULT_NOPAGE upon success and for kernels 4.20 and later

Signed-off-by: Serge Ayoun <[email protected]>

Branch sgx2 (SGX 1.5 and SGX 2.0) should be used from now.

Signed-off-by: Serge Ayoun <[email protected]>

Branch sgx2 (SGX 1.5 and SGX 2.0) should be used from now.

Signed-off-by: Serge Ayoun <[email protected]>

README file updated: master branch (SGX 1.5) not supported anymore.

README file updated: master branch (SGX 1.5) not supported anymore.

Signed-off-by: Zhang Lili Z <[email protected]>

Fix configid offset in secs.

README file updated

Signed-off-by: Serge Ayoun <[email protected]>

elfutils-libelf-devel is a prerequisite for RedHat 8.0

Signed-off-by: Ayoun Serge <[email protected]>

Ubuntu 18.04 support added to README

apply_to_page_range() API signature was changed
Makefile does not support SUBDIR anymore

Signed-off-by: Ayoun Serge <[email protected]>

Build error fixes for kernel 5.3 and later

Signed-off-by: Huang, Haitao <[email protected]>

Version 2.6

Signed-off-by: Huang, Haitao <[email protected]>

Also updated to refer to DCAP driver for tracking proposed upstream version

Signed-off-by: Huang, Haitao <[email protected]>

Removed mentioning of sgx2 branch and update comments to emphasize differences between kernel, DCAP and this driver.

They have new names in more recent kernels. We check for the old
defines, however, in order to prevent breaking on older kernels.

Also, rename FEATURE_CONTROL_SGX_ENABLE to match the new naming.

Signed-off-by: Nathaniel McCallum <[email protected]>

Rename MSR_IA32_FEATURE_CONTROL and FEATURE_CONTROL_LOCKED

Signed-off-by: Haitao Huang <[email protected]>

Signed-off-by: Haitao Huang <[email protected]>

Minor fixes

Clarify intended usage of this driver

Signed-off-by: Haitao Huang <[email protected]>

Signed-off-by: Haitao Huang <[email protected]>

Fix build for RHEL 8.2

Update README.md

Obviously, this if check is redundant. Remove it.

Signed-off-by: Tianjia Zhang <[email protected]>

Signed-off-by: Tianjia Zhang <[email protected]>

Signed-off-by: Tianjia Zhang <[email protected]>

Fix

Shifting a signed integer value of 1 by 31 or more bits will cause
overflow and can lead to undefined behaviour. Fix this by adding
a UL suffix to ensure an unsigned long is being shifted.

Signed-off-by: Colin Ian King [email protected]

Signed-off-by: Haitao Huang <[email protected]>

Fix signed integer overflow on shift

Note this is used in Intel PSW release 2.11

Signed-off-by: Haitao Huang <[email protected]>

Update version to 2.11

* Fix compilation on 5.8 kernel

Signed-off-by: Don Porter <[email protected]>

* Address review comments

* Address review comments

Signed-off-by: Haitao Huang <[email protected]>

Signed-off-by: Haitao Huang <[email protected]>

reset hashes

Signed-off-by: Haitao Huang <[email protected]>

Update supported OSes by referring to releases

For non-FLC platforms, Linux kernel 5.11 or above disables sgx feature flag
so boot_cpu_has is not usable.

Signed-off-by: Haitao Huang <[email protected]>

Use cpuid instead of boot_cpu_has to check cpu features

Signed-off-by: Zhang Lili [email protected]

Signed-off-by: Haitao Huang <[email protected]>

reuse the patch from intel#151 and fix "shift exponent 32 is too large for 32-bit type 'int'" when loading the isgx module