Should it raise an exception when VA exceed the max range of the VA when mmu off? what kind of the that exception?

[rock-ifly]

Should it raise an exception when VA exceed the max range of the VA when mmu off? and what kind of the that exception?

Let's think about a PAW(PA width) = 40, when it is in M mode and mstatus.mprv=0 (MMU off)

1） Can the software use a VA that exceed 40-bits, like VA0=0xAAAA_BBBB_CCCC？ Here the VA0 width is 48bits， exceed the max PA width(40bits).
2) if The answer is NO, then an exception should be raised, the what kind shold we use? access fault, selfdefined fault, or any other determined kind? The RV spec is better to make this clear.

[rock-ifly]

The same case when pte.ppn[55:40] is not full-0, when the PAW=40, should an exception be raised? access fault or any other exception?

[allenjbaum]

for SV48, a VA can be any address such that A[63:47] == either all1s, or all 0s (so, sign extended from the VA MSB. the PAW is completely independent from that; it can be larger or smaller. if A[63:PAW] !=0, then you should get an illegal access fault. (I don't recall if there are any weird cases that some other cause would be reported offhand, and I'm too lazy to look) It is the responsibility of the OS to ensure that the VA->PA mapping defined by the page table never exceeds that PAW limit.

…

On Thu, Dec 5, 2024 at 11:17 PM rock-ifly ***@***.***> wrote: The same case when pte.ppn[55:40] is not full-0, when the PAW=40, should an exception be raised? access fault or any other exception? — Reply to this email directly, view it on GitHub <#1754 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AHPXVJQOTUZD5DORDLT7MTD2EFFQFAVCNFSM6AAAAABTEAWWOCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMRSGI4TSNRTGM> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[rock-ifly]

for SV48, a VA can be any address such that A[63:47] == either all1s, or all 0s (so, sign extended from the VA MSB. the PAW is completely independent from that; it can be larger or smaller. if A[63:PAW] !=0, then you should get an illegal access fault. (I don't recall if there are any weird cases that some other cause would be reported offhand, and I'm too lazy to look) It is the responsibility of the OS to ensure that the VA->PA mapping defined by the page table never exceeds that PAW limit.
…
On Thu, Dec 5, 2024 at 11:17 PM rock-ifly @.> wrote: The same case when pte.ppn[55:40] is not full-0, when the PAW=40, should an exception be raised? access fault or any other exception? — Reply to this email directly, view it on GitHub <#1754 (comment)>, or unsubscribe https://github.com/notifications/unsubscribe-auth/AHPXVJQOTUZD5DORDLT7MTD2EFFQFAVCNFSM6AAAAABTEAWWOCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDKMRSGI4TSNRTGM . You are receiving this because you are subscribed to this thread.Message ID: @.>

Thanks， I see.

1. mmu on, for a Sv39/Sv48/Sv57, sign extenstoin is checked for VA
2. mmu on, OS should ensure the pte.ppn(PA) not exceed the PAW during the PTW process.
3. mmu off, OS should ensure the PA=VA not exceed the PAW

Why not make it clear in Ch12.3.2 vritual address translation prcess in the RV spec, since the PAW is different for every implementation.