Skip to contentSkip to conversation

target/arm: Use signed quantity to represent VMSAv8-64 translation level

PATCH Superseded Ard Biesheuvel 2 participants 5 messages Nov 21, 2022 – Nov 22, 2022

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 target/arm: Use signed quantity to represent VMSAv8-64 translation level

The LPA2 extension implements 52-bit virtual addressing for 4k and 16k translation granules, and for the former, this means an additional level of translation is needed. This means we start counting at -1 instead of 0 when doing a walk, and so 'level' is now a signed quantity, and should be typed as such. So turn it from uint32_t into int32_t.

Cc Peter Maydell <[email protected]>
Cc Philippe Mathieu-Daudé <[email protected]>
Cc Richard Henderson <[email protected]>
---
 target/arm/ptw.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
target/arm/ptw.c +1 -1
index 3745ac9723..6d6992580a 100644
@@ -1172,7 +1172,7 @@ static bool get_phys_addr_lpae(CPUARMState *env, S1Translate *ptw,
     ARMCPU *cpu = env_archcpu(env);
     ARMMMUIdx mmu_idx = ptw->in_mmu_idx;
     bool is_secure = ptw->in_secure;
-    uint32_t level;
+    int32_t level;
     ARMVAParameters param;
     uint64_t ttbr;
     hwaddr descaddr, indexmask, indexmask_grainsize;
Signature · 1 line
2.35.1
#1 lore raw
Quoted 7 lines · Ard Biesheuvel

The LPA2 extension implements 52-bit virtual addressing for 4k and 16k translation granules, and for the former, this means an additional level of translation is needed. This means we start counting at -1 instead of 0 when doing a walk, and so 'level' is now a signed quantity, and should be typed as such. So turn it from uint32_t into int32_t.

Does this cause any visible wrong behaviour, or is it just a cleanup thing ?

thanks
-- PMM

#2 lore raw
Quoted 10 lines · Peter Maydell
Quoted 7 lines · Ard Biesheuvel

The LPA2 extension implements 52-bit virtual addressing for 4k and 16k translation granules, and for the former, this means an additional level of translation is needed. This means we start counting at -1 instead of 0 when doing a walk, and so 'level' is now a signed quantity, and should be typed as such. So turn it from uint32_t into int32_t.

Quoted 3 lines

Does this cause any visible wrong behaviour, or is it just a cleanup thing ?

No, 5 level paging is completely broken because of this, given that the 'level < 3' tests give the wrong result for (uint32_t)-1

#3 lore raw
Quoted 12 lines · Ard Biesheuvel
Quoted 10 lines · Peter Maydell
Quoted 7 lines · Ard Biesheuvel

The LPA2 extension implements 52-bit virtual addressing for 4k and 16k translation granules, and for the former, this means an additional level of translation is needed. This means we start counting at -1 instead of 0 when doing a walk, and so 'level' is now a signed quantity, and should be typed as such. So turn it from uint32_t into int32_t.

Quoted 6 lines
Quoted 3 lines

Does this cause any visible wrong behaviour, or is it just a cleanup thing ?

No, 5 level paging is completely broken because of this, given that the 'level < 3' tests give the wrong result for (uint32_t)-1

Right, thanks. This seems like a bug worth fixing for 7.2.

We should make 'uint32_t startlevel' also an int32_t for consistency, I think, given that it is also sometimes negative, though in that case it doesn't get used in any comparisons so it's not going to cause wrong behaviour.

thanks
-- PMM

#4 lore raw
Quoted 14 lines · Peter Maydell
Quoted 12 lines · Ard Biesheuvel
Quoted 10 lines · Peter Maydell
Quoted 7 lines · Ard Biesheuvel

The LPA2 extension implements 52-bit virtual addressing for 4k and 16k translation granules, and for the former, this means an additional level of translation is needed. This means we start counting at -1 instead of 0 when doing a walk, and so 'level' is now a signed quantity, and should be typed as such. So turn it from uint32_t into int32_t.

Quoted 9 lines
Quoted 6 lines
Quoted 3 lines

Does this cause any visible wrong behaviour, or is it just a cleanup thing ?

No, 5 level paging is completely broken because of this, given that the 'level < 3' tests give the wrong result for (uint32_t)-1

Right, thanks. This seems like a bug worth fixing for 7.2.

Indeed. And the other patch I sent is needed too if you want to run with LPA2

'target/arm: Limit LPA2 effective output address when TCR.DS == 0'

In case it is useful, I have a WIP kernel branch here which can be built with 52-bit virtual addressing for 4k or 16k pages.

https://git.kernel.org/pub/scm/linux/kernel/git/ardb/linux.git/log/?h=arm64-4k-lpa2

Quoted 5 lines

We should make 'uint32_t startlevel' also an int32_t for consistency, I think, given that it is also sometimes negative, though in that case it doesn't get used in any comparisons so it's not going to cause wrong behaviour.

Indeed. I'll send a v2 and fold that in.