On 03/02/2018 06:30 PM, Will Deacon wrote:
On Fri, Mar 02, 2018 at 05:02:32PM +0800, Alex Shi wrote:
As testing the spectre bug fix, that's a good question. I also asked this question to original patch authors, like Marc. They said they just figure out these patches could block spectre or meltdown issue. From my side, I just reproduced the process internal spectre. But all fix on arm can not resolve the user space internal spectre. It can block from user to kernel or kernel to user spectre according the code purose. So I believe these patch could do their job. And arm cpu would drop the spectre branches if it has 20+ 'nop' instructions...
Since this is archived on a public list and I don't want people to rely on this, no, you cannot rely on "20+ 'nop' instructions" to work around spectre on arm/arm64. It might prevent a particular PoC working on a particular SoC, but it's fragile at best.
Thanks for comments, Will!
Yes, I full understand the difference between SoCs. Thanks for point it out!
Regards Alex