On Thu, Feb 20, 2025 at 1:47 PM Dave Hansen dave.hansen@intel.com wrote:
On 2/20/25 13:16, Vishal Annapurve wrote:
Since enabling CONFIG_PARAVIRT_XXL is too bloated for TDX guest like platforms, move HLT and SAFE_HLT paravirt calls under CONFIG_PARAVIRT.
I guess it's just one patch, but doesn't this expose CONFIG_PARAVIRT=y users to what _was_ specific to CONFIG_PARAVIRT_XXL=y? According to the changelog, TDX users shouldn't have to use use PARAVIRT_XXL, so PARAVIRT=y and PARAVIRT_XXL=n must be an *IMPORTANT* configuration for TDX users.
Before this patch, those users would have no way to hit the unsafe-for-TDX pv_native_safe_halt(). After this patch, they will hit it.
Before this patch, those users had access to arch_safe_halt() -> native_safe_halt() path. With this patch, such users can execute arch_safe_halt -> pv_native_safe_halt() -> native_safe_halt(), so this patch doesn't cause any additional regression.
So, there are two possibilities:
- This patch breaks bisection for an important TDX configuration
- This patch's conjecture that PARAVIRT_XXL=n is important for TDX is wrong and it is not necessary in the first place.
What am I missing?