Skip to contentSkip to conversation

Arm32: MSR to SPSR needs qualification

PATCH Superseded Jan Beulich 3 participants 6 messages Jun 11, 2021

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Arm32: MSR to SPSR needs qualification

The Arm ARM's description of MSR doesn't even allow for plain "SPSR" here, and while gas accepts this, it takes it to mean SPSR_cf. Yet surely all of SPSR wants updating on this path, not just the lowest and highest 8 bits.

Fixes dfcffb128be4 ("xen/arm32: SPSR_hyp/SPSR")
xen/arch/arm/arm32/entry.S +1 -1
@@ -395,7 +395,7 @@ return_to_hypervisor:
         ldr r11, [sp, #UREGS_pc]
         msr ELR_hyp, r11
         ldr r11, [sp, #UREGS_cpsr]
-        msr SPSR, r11
+        msr SPSR_cxsf, r11
 #ifdef CONFIG_ARM32_HARDEN_BRANCH_PREDICTOR
         /*
          * Hardening branch predictor may require to setup a different
#1 lore raw
Quoted 5 lines · On Fri, 11 Jun 2021, 08:55 Jan Beulich, <[email protected]> wrote:

The Arm ARM's description of MSR doesn't even allow for plain "SPSR" here, and while gas accepts this, it takes it to mean SPSR_cf. Yet surely all of SPSR wants updating on this path, not just the lowest and highest 8 bits.

Can you provide a reference to the Arm Arm? This would help to navigate through the 8000 pages.

Cheers,

Quoted 16 lines
Fixes dfcffb128be4 ("xen/arm32: SPSR_hyp/SPSR")
xen/arch/arm/arm32/entry.S +1 -1
@@ -395,7 +395,7 @@ return_to_hypervisor:
         ldr r11, [sp, #UREGS_pc]
         msr ELR_hyp, r11
         ldr r11, [sp, #UREGS_cpsr]
-        msr SPSR, r11
+        msr SPSR_cxsf, r11
 #ifdef CONFIG_ARM32_HARDEN_BRANCH_PREDICTOR
         /*
          * Hardening branch predictor may require to setup a different
#2 lore raw
Quoted 10 lines · Julien Grall
Quoted 5 lines · On Fri, 11 Jun 2021, 08:55 Jan Beulich, <[email protected]> wrote:

The Arm ARM's description of MSR doesn't even allow for plain "SPSR" here, and while gas accepts this, it takes it to mean SPSR_cf. Yet surely all of SPSR wants updating on this path, not just the lowest and highest 8 bits.

Can you provide a reference to the Arm Arm? This would help to navigate through the 8000 pages.

Referencing the instruction page would be enough, I thought (as even I, not being an Arm person, have no difficulty locating it). If it isn't, how is a canonical doc ref supposed to look like on Arm? On x86, we avoid recording document versions, section numbers, or even page numbers in code comments or commit messages (which isn't to say we have none of these, but we try to avoid new ones to appear), as these tend to change with every new version of the doc. Therefore, to me, the offending commit's "ARM DDI 0487D.b page G8-5993" doesn't look like something I wanted to clone from. But if you tell me otherwise, then well - so be it.

Jan

#3 lore raw
Quoted 12 lines · On Fri, 11 Jun 2021, 11:16 Jan Beulich, <[email protected]> wrote:
Quoted 10 lines · Julien Grall
Quoted 5 lines · On Fri, 11 Jun 2021, 08:55 Jan Beulich, <[email protected]> wrote:

The Arm ARM's description of MSR doesn't even allow for plain "SPSR" here, and while gas accepts this, it takes it to mean SPSR_cf. Yet surely all of SPSR wants updating on this path, not just the lowest and highest 8 bits.

Can you provide a reference to the Arm Arm? This would help to navigate through the 8000 pages.

Quoted 10 lines

Referencing the instruction page would be enough, I thought (as even I, not being an Arm person, have no difficulty locating it). If it isn't, how is a canonical doc ref supposed to look like on Arm? On x86, we avoid recording document versions, section numbers, or even page numbers in code comments or commit messages (which isn't to say we have none of these, but we try to avoid new ones to appear), as these tend to change with every new version of the doc. Therefore, to me, the offending commit's "ARM DDI 0487D.b page G8-5993" doesn't look like something I wanted to clone from. But if you tell me otherwise, then well - so be it.

The Arm website provides a link for nearly every revision on the specs. As the wording can change between version, it is useful to know which spec the understanding is based from.

Note that for Arm32 we should quote the Armv7 spec and not the Armv8 one
because we only follow the former (there are a few small differences).

Quoted 4 lines

Jan

#4 lore raw
Quoted 26 lines · Julien Grall
Quoted 12 lines · On Fri, 11 Jun 2021, 11:16 Jan Beulich, <[email protected]> wrote:
Quoted 10 lines · Julien Grall
Quoted 5 lines · On Fri, 11 Jun 2021, 08:55 Jan Beulich, <[email protected]> wrote:

The Arm ARM's description of MSR doesn't even allow for plain "SPSR" here, and while gas accepts this, it takes it to mean SPSR_cf. Yet surely all of SPSR wants updating on this path, not just the lowest and highest 8 bits.

Can you provide a reference to the Arm Arm? This would help to navigate through the 8000 pages.

Quoted 10 lines

Referencing the instruction page would be enough, I thought (as even I, not being an Arm person, have no difficulty locating it). If it isn't, how is a canonical doc ref supposed to look like on Arm? On x86, we avoid recording document versions, section numbers, or even page numbers in code comments or commit messages (which isn't to say we have none of these, but we try to avoid new ones to appear), as these tend to change with every new version of the doc. Therefore, to me, the offending commit's "ARM DDI 0487D.b page G8-5993" doesn't look like something I wanted to clone from. But if you tell me otherwise, then well - so be it.

Quoted 6 lines

The Arm website provides a link for nearly every revision on the specs. As the wording can change between version, it is useful to know which spec the understanding is based from.

Note that for Arm32 we should quote the Armv7 spec and not the Armv8 one
because we only follow the former (there are a few small differences).

Thanks for having me dig out an up-to-date Armv7 spec. I find this puzzling in particular because you didn't care to have the earlier commit provide a v7 doc ref. Initially I did intentionally use (a newer version of) the doc that was pointed at there (which I also think is better structured than the v7 one).

Jan

#5 lore raw
Quoted 34 lines · On Fri, 11 Jun 2021, 15:02 Jan Beulich, <[email protected]> wrote:
Quoted 8 lines · Julien Grall
Quoted 6 lines · On Fri, 11 Jun 2021, 11:16 Jan Beulich, <[email protected]> wrote:
Quoted 5 lines · Julien Grall
Quoted 3 lines · On Fri, 11 Jun 2021, 08:55 Jan Beulich, <[email protected]> wrote:

The Arm ARM's description of MSR doesn't even allow for plain "SPSR" here, and while gas accepts this, it takes it to mean SPSR_cf. Yet surely all of SPSR wants updating on this path, not just the lowest

and

Quoted 18 lines
Quoted 6 lines
Quoted 5 lines
Quoted 2 lines

highest 8 bits.

Can you provide a reference to the Arm Arm? This would help to navigate through the 8000 pages.

Quoted 10 lines

Referencing the instruction page would be enough, I thought (as even I, not being an Arm person, have no difficulty locating it). If it isn't, how is a canonical doc ref supposed to look like on Arm? On x86, we avoid recording document versions, section numbers, or even page numbers in code comments or commit messages (which isn't to say we have none of these, but we try to avoid new ones to appear), as these tend to change with every new version of the doc. Therefore, to me, the offending commit's "ARM DDI 0487D.b page G8-5993" doesn't look like something I wanted to clone from. But if you tell me otherwise, then well - so be it.

Quoted 1 line

The Arm website provides a link for nearly every revision on the specs.

As

Quoted 1 line

the wording can change between version, it is useful to know which spec

the

Quoted 2 lines

understanding is based from.

Quoted 8 lines
Quoted 2 lines

Note that for Arm32 we should quote the Armv7 spec and not the Armv8 one
because we only follow the former (there are a few small differences).

Thanks for having me dig out an up-to-date Armv7 spec. I find this puzzling in particular because you didn't care to have the earlier commit provide a v7 doc ref. Initially I did intentionally use (a newer version of) the doc that was pointed at there (which I also think is better structured than the v7 one).

Well Stefano replied past midnight UK time with the reference and committed nearly afterwards. So I didn't really have time to object...

When I asked for the reference I didn't think I needed to mention it should be the Armv7 as he should know we only support Armv7 for 32-bit.

I didn't bother to reply afterwards. But given there is a bug and you quoted him, I chose to make clear that reference should be Armv7 only.

Cheers,

Quoted 3 lines

Jan