Skip to contentSkip to conversation

target/arm: Some CONFIG_TCG code movement

PATCH Superseded Fabiano Rosas 4 participants 20 messages Dec 16, 2022 – Dec 20, 2022 Arm/Amlogic Meson SoC support

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

cover target/arm: Some CONFIG_TCG code movement

Hi,

This is the second round of rebasing the patches from:
https://lore.kernel.org/r/[email protected]

These are the simpler ones that move code under CONFIG_TCG/tcg_enabled. No new directories or files.

Claudio Fontana (5):
target/arm: only build psci for TCG
target/arm: rename handle_semihosting to tcg_handle_semihosting target/arm: wrap semihosting and psci calls with tcg_enabled target/arm: wrap call to aarch64_sve_change_el in tcg_enabled() target/arm: only perform TCG cpu and machine inits if TCG enabled

 target/arm/cpu.c       | 31 ++++++++++++---------
 target/arm/helper.c    | 45 +++++++++++++++++--------------
 target/arm/kvm.c       | 18 ++++++-------
 target/arm/kvm_arm.h   |  3 +--
 target/arm/machine.c   | 61 +++++++++++++++++++++++-------------------
 target/arm/meson.build |  5 +++-
 6 files changed, 91 insertions(+), 72 deletions(-)
Signature · 1 line
2.35.3
#1 lore raw

PATCH 1/5 target/arm: only build psci for TCG

From: Claudio Fontana <[email protected]>

---
Originally from:
[RFC v14 09/80] target/arm: only build psci for TCG
https://lore.kernel.org/r/[email protected]

---
 target/arm/meson.build | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
target/arm/meson.build +4 -1
index 87e911b27f..26e425418f 100644
@@ -61,10 +61,13 @@ arm_softmmu_ss.add(files(
   'arm-powerctl.c',
   'machine.c',
   'monitor.c',
-  'psci.c',
   'ptw.c',
 ))
 
+arm_softmmu_ss.add(when: 'CONFIG_TCG', if_true: files(
+  'psci.c',
+))
+
 subdir('hvf')
 
 target_arch += {'arm': arm_ss}
Signature · 1 line
2.35.3
#2 lore raw

Re: [PATCH 1/5] target/arm: only build psci for TCG

Hi Claudio,

If the PSCI implementation becomes TCG only, can we also move to a tcg accel directory? It slowly gets super confusing to keep track of which files are supposed to be generic target code and which ones TCG specific.

Alex

Quoted 39 lines

Am 16.12.2022 um 22:37 schrieb Fabiano Rosas <[email protected]>:

From: Claudio Fontana <[email protected]>

---
Originally from:
[RFC v14 09/80] target/arm: only build psci for TCG
https://lore.kernel.org/r/[email protected]
---
target/arm/meson.build | 5 ++++-

1 file changed, 4 insertions(+), 1 deletion(-)
target/arm/meson.build +4 -1
index 87e911b27f..26e425418f 100644
@@ -61,10 +61,13 @@ arm_softmmu_ss.add(files(
  'arm-powerctl.c',
  'machine.c',
  'monitor.c',
-  'psci.c',
  'ptw.c',
 ))
 
+arm_softmmu_ss.add(when: 'CONFIG_TCG', if_true: files(
+  'psci.c',
+))
+
 subdir('hvf')
 
 target_arch += {'arm': arm_ss}
Signature · 1 line
2.35.3
#3 lore raw

Re: [PATCH 1/5] target/arm: only build psci for TCG

Quoted 4 lines · Alexander Graf

Hi Claudio,

If the PSCI implementation becomes TCG only, can we also move to a tcg accel directory? It slowly gets super confusing to keep track of which files are supposed to be generic target code and which ones TCG specific> Alex

Hi Alex, Fabiano, Peter and all,

that was the plan but at the time of:

https://lore.kernel.org/all/[email protected]/

Peter mentioned that HVF AArch64 might use that code too:

https://lists.gnu.org/archive/html/qemu-devel/2021-03/msg00509.html

so from v2 to v3 the series changed to not move the code under tcg/ , expecting HVF to be reusing that code "soon".

I see that your hvf code in hvf/ implements psci, is there some plan to reuse pieces from the tcg implementation now?

Ciao,

Claudio

Quoted 40 lines
Quoted 40 lines

Am 16.12.2022 um 22:37 schrieb Fabiano Rosas <[email protected]>:

From: Claudio Fontana <[email protected]>

---
Originally from:
[RFC v14 09/80] target/arm: only build psci for TCG
https://lore.kernel.org/r/[email protected]
---
target/arm/meson.build | 5 ++++-

1 file changed, 4 insertions(+), 1 deletion(-)
target/arm/meson.build +4 -1
index 87e911b27f..26e425418f 100644
@@ -61,10 +61,13 @@ arm_softmmu_ss.add(files(
  'arm-powerctl.c',
  'machine.c',
  'monitor.c',
-  'psci.c',
  'ptw.c',
 ))
 
+arm_softmmu_ss.add(when: 'CONFIG_TCG', if_true: files(
+  'psci.c',
+))
+
 subdir('hvf')
 
 target_arch += {'arm': arm_ss}
Signature · 1 line
2.35.3
#4 lore raw

Re: [PATCH 1/5] target/arm: only build psci for TCG

Hey Claudio,

Quoted 12 lines · Claudio Fontana
Quoted 4 lines · Alexander Graf

Hi Claudio,

If the PSCI implementation becomes TCG only, can we also move to a tcg accel directory? It slowly gets super confusing to keep track of which files are supposed to be generic target code and which ones TCG specific> Alex

Hi Alex, Fabiano, Peter and all,

that was the plan but at the time of:

https://lore.kernel.org/all/[email protected]/

Quoted 7 lines

Peter mentioned that HVF AArch64 might use that code too:

https://lists.gnu.org/archive/html/qemu-devel/2021-03/msg00509.html

so from v2 to v3 the series changed to not move the code under tcg/ , expecting HVF to be reusing that code "soon".

I see that your hvf code in hvf/ implements psci, is there some plan to reuse pieces from the tcg implementation now?

I originally reused the PSCI code in earlier versions of my hvf patch set, but then we realized that some functions like remote CPU reset are wired in a TCG specific view of the world with full target CPU register ownership. So if we want to actually share it, we'll need to abstract it up a level.

Hence I'd suggest to move it to a TCG directory for now and then later move it back into a generic helper if we want / need to. The code just simply isn't generic yet.

Or alternatively, you create a patch (set) to actually merge the 2 implementations into a generic one again which then can live at a generic place :)

Alex

#5 lore raw

Re: [PATCH 1/5] target/arm: only build psci for TCG

Ciao Alex,

Quoted 33 lines · Alexander Graf

Hey Claudio,

Quoted 12 lines · Claudio Fontana
Quoted 4 lines · Alexander Graf

Hi Claudio,

If the PSCI implementation becomes TCG only, can we also move to a tcg accel directory? It slowly gets super confusing to keep track of which files are supposed to be generic target code and which ones TCG specific> Alex

Hi Alex, Fabiano, Peter and all,

that was the plan but at the time of:

https://lore.kernel.org/all/[email protected]/

Quoted 7 lines

Peter mentioned that HVF AArch64 might use that code too:

https://lists.gnu.org/archive/html/qemu-devel/2021-03/msg00509.html

so from v2 to v3 the series changed to not move the code under tcg/ , expecting HVF to be reusing that code "soon".

I see that your hvf code in hvf/ implements psci, is there some plan to reuse pieces from the tcg implementation now?

I originally reused the PSCI code in earlier versions of my hvf patch set, but then we realized that some functions like remote CPU reset are wired in a TCG specific view of the world with full target CPU register ownership. So if we want to actually share it, we'll need to abstract it up a level.

Hence I'd suggest to move it to a TCG directory for now and then later move it back into a generic helper if we want / need to. The code just simply isn't generic yet.

Quoted 6 lines

Or alternatively, you create a patch (set) to actually merge the 2 implementations into a generic one again which then can live at a generic place :)

Alex

Thanks for the clarification, I'll leave the choice up to Fabiano now, since he is working on the series currently :-)

Ciao,

Claudio

#6 lore raw

Re: [PATCH 1/5] target/arm: only build psci for TCG

Quoted 41 lines · Claudio Fontana

Ciao Alex,

Quoted 38 lines · Alexander Graf

Hey Claudio,

Quoted 12 lines · Claudio Fontana
Quoted 4 lines · Alexander Graf

Hi Claudio,

If the PSCI implementation becomes TCG only, can we also move to a tcg accel directory? It slowly gets super confusing to keep track of which files are supposed to be generic target code and which ones TCG specific> Alex

Hi Alex, Fabiano, Peter and all,

that was the plan but at the time of:

https://lore.kernel.org/all/[email protected]/

Quoted 7 lines

Peter mentioned that HVF AArch64 might use that code too:

https://lists.gnu.org/archive/html/qemu-devel/2021-03/msg00509.html

so from v2 to v3 the series changed to not move the code under tcg/ , expecting HVF to be reusing that code "soon".

I see that your hvf code in hvf/ implements psci, is there some plan to reuse pieces from the tcg implementation now?

I originally reused the PSCI code in earlier versions of my hvf patch set, but then we realized that some functions like remote CPU reset are wired in a TCG specific view of the world with full target CPU register ownership. So if we want to actually share it, we'll need to abstract it up a level.

Hence I'd suggest to move it to a TCG directory for now and then later move it back into a generic helper if we want / need to. The code just simply isn't generic yet.

Or alternatively, you create a patch (set) to actually merge the 2 implementations into a generic one again which then can live at a generic place :)

Quoted 7 lines
Quoted 1 line

Alex

Thanks for the clarification, I'll leave the choice up to Fabiano now, since he is working on the series currently :-)

Ciao,

Claudio

Hello, thank you all for the comments.

I like the idea of merging the two implementations. However, I won't get to it anytime soon. There's still ~70 patches in the original series that I need to understand, rebase and test, including the introduction of the tcg directory.

I'd say we merge this as is now, since this patch has no dependencies. Later when I introduce the tcg directory I can move the code there along with the other tcg-only files. I'll take note to come back to the PSCI code as well.

#7 lore raw

Re: [PATCH 1/5] target/arm: only build psci for TCG

Hey Fabiano,

Quoted 46 lines · Fabiano Rosas
Quoted 44 lines · Claudio Fontana

Ciao Alex,

Quoted 31 lines · Alexander Graf

Hey Claudio,

Quoted 11 lines · Claudio Fontana
Quoted 4 lines · Alexander Graf

Hi Claudio,

If the PSCI implementation becomes TCG only, can we also move to a tcg accel directory? It slowly gets super confusing to keep track of which files are supposed to be generic target code and which ones TCG specific> Alex

Hi Alex, Fabiano, Peter and all,

that was the plan but at the time of:

https://lore.kernel.org/all/[email protected]/

Quoted 7 lines

Peter mentioned that HVF AArch64 might use that code too:

https://lists.gnu.org/archive/html/qemu-devel/2021-03/msg00509.html

so from v2 to v3 the series changed to not move the code under tcg/ , expecting HVF to be reusing that code "soon".

I see that your hvf code in hvf/ implements psci, is there some plan to reuse pieces from the tcg implementation now?

I originally reused the PSCI code in earlier versions of my hvf patch set, but then we realized that some functions like remote CPU reset are wired in a TCG specific view of the world with full target CPU register ownership. So if we want to actually share it, we'll need to abstract it up a level.

Hence I'd suggest to move it to a TCG directory for now and then later move it back into a generic helper if we want / need to. The code just simply isn't generic yet.

Quoted 6 lines

Or alternatively, you create a patch (set) to actually merge the 2 implementations into a generic one again which then can live at a generic place :)

Alex

Thanks for the clarification, I'll leave the choice up to Fabiano now, since he is working on the series currently :-)

Ciao,

Quoted 7 lines
Quoted 1 line

Claudio

Hello, thank you all for the comments.

I like the idea of merging the two implementations. However, I won't get to it anytime soon. There's still ~70 patches in the original series that I need to understand, rebase and test, including the introduction of the tcg directory.

Sure, I am definitely fine with leaving them separate for now as well :).

Quoted 4 lines

I'd say we merge this as is now, since this patch has no dependencies. Later when I introduce the tcg directory I can move the code there along with the other tcg-only files. I'll take note to come back to the PSCI code as well.

I'm confused about the patch ordering :). Why is it easier to move the psci.c compilation target from generic to an if(CONFIG_TCG) only to later move it into a tcg/ directory? Wouldn't it be easier to create a tcg/ directory from the start and just put it there?

The current approach just looks like duplicate effort to me.

Alex

#8 lore raw

Re: [PATCH 1/5] target/arm: only build psci for TCG

Quoted 61 lines · Alexander Graf

Hey Fabiano,

Quoted 46 lines · Fabiano Rosas
Quoted 44 lines · Claudio Fontana

Ciao Alex,

Quoted 31 lines · Alexander Graf

Hey Claudio,

Quoted 11 lines · Claudio Fontana
Quoted 4 lines · Alexander Graf

Hi Claudio,

If the PSCI implementation becomes TCG only, can we also move to a tcg accel directory? It slowly gets super confusing to keep track of which files are supposed to be generic target code and which ones TCG specific> Alex

Hi Alex, Fabiano, Peter and all,

that was the plan but at the time of:

https://lore.kernel.org/all/[email protected]/

Quoted 7 lines

Peter mentioned that HVF AArch64 might use that code too:

https://lists.gnu.org/archive/html/qemu-devel/2021-03/msg00509.html

so from v2 to v3 the series changed to not move the code under tcg/ , expecting HVF to be reusing that code "soon".

I see that your hvf code in hvf/ implements psci, is there some plan to reuse pieces from the tcg implementation now?

I originally reused the PSCI code in earlier versions of my hvf patch set, but then we realized that some functions like remote CPU reset are wired in a TCG specific view of the world with full target CPU register ownership. So if we want to actually share it, we'll need to abstract it up a level.

Hence I'd suggest to move it to a TCG directory for now and then later move it back into a generic helper if we want / need to. The code just simply isn't generic yet.

Quoted 6 lines

Or alternatively, you create a patch (set) to actually merge the 2 implementations into a generic one again which then can live at a generic place :)

Alex

Thanks for the clarification, I'll leave the choice up to Fabiano now, since he is working on the series currently :-)

Ciao,

Quoted 7 lines
Quoted 1 line

Claudio

Hello, thank you all for the comments.

I like the idea of merging the two implementations. However, I won't get to it anytime soon. There's still ~70 patches in the original series that I need to understand, rebase and test, including the introduction of the tcg directory.

Sure, I am definitely fine with leaving them separate for now as well :).

Quoted 8 lines
Quoted 4 lines

I'd say we merge this as is now, since this patch has no dependencies. Later when I introduce the tcg directory I can move the code there along with the other tcg-only files. I'll take note to come back to the PSCI code as well.

I'm confused about the patch ordering :). Why is it easier to move the psci.c compilation target from generic to an if(CONFIG_TCG) only to later move it into a tcg/ directory?

It's a simple patch, so the overhead didn't cross my mind. But you are right, this could go directly into tcg/ without having to put it under CONFIG_TCG first.

Quoted 2 lines

Wouldn't it be easier to create a
tcg/ directory from the start and just put it there?

I don't know about "from the start". At this point we cannot have a single commit moving everything into the tcg/ directory because some files still contain tcg + non-tcg code. We would end up with several commits moving files into tcg/ along the history, which I think results in a poor experience when inspecting the log later on (git blame and so on). So my idea was to split as much as I can upfront and only later move everything into the directory.

I'm also rebasing this series [1] from 2021, which means I'd rather have small chunks of code moved under CONFIG_TCG that I can build-test with --disable-tcg (even though the build doesn't finish, I can see the number of errors going down), than to move non-tcg code into tcg/ and then pull it out later like in the original series.

1- https://lore.kernel.org/r/[email protected]

But hey, that's just my reasoning, no strong feelings about it.

#9 lore raw

Re: [PATCH 1/5] target/arm: only build psci for TCG

Quoted 66 lines · Fabiano Rosas
Quoted 64 lines · Alexander Graf

Hey Fabiano,

Quoted 46 lines · Fabiano Rosas
Quoted 44 lines · Claudio Fontana

Ciao Alex,

Quoted 31 lines · Alexander Graf

Hey Claudio,

Quoted 11 lines · Claudio Fontana
Quoted 4 lines · Alexander Graf

Hi Claudio,

If the PSCI implementation becomes TCG only, can we also move to a tcg accel directory? It slowly gets super confusing to keep track of which files are supposed to be generic target code and which ones TCG specific> Alex

Hi Alex, Fabiano, Peter and all,

that was the plan but at the time of:

https://lore.kernel.org/all/[email protected]/

Quoted 7 lines

Peter mentioned that HVF AArch64 might use that code too:

https://lists.gnu.org/archive/html/qemu-devel/2021-03/msg00509.html

so from v2 to v3 the series changed to not move the code under tcg/ , expecting HVF to be reusing that code "soon".

I see that your hvf code in hvf/ implements psci, is there some plan to reuse pieces from the tcg implementation now?

I originally reused the PSCI code in earlier versions of my hvf patch set, but then we realized that some functions like remote CPU reset are wired in a TCG specific view of the world with full target CPU register ownership. So if we want to actually share it, we'll need to abstract it up a level.

Hence I'd suggest to move it to a TCG directory for now and then later move it back into a generic helper if we want / need to. The code just simply isn't generic yet.

Quoted 6 lines

Or alternatively, you create a patch (set) to actually merge the 2 implementations into a generic one again which then can live at a generic place :)

Alex

Thanks for the clarification, I'll leave the choice up to Fabiano now, since he is working on the series currently :-)

Ciao,

Quoted 7 lines
Quoted 1 line

Claudio

Hello, thank you all for the comments.

I like the idea of merging the two implementations. However, I won't get to it anytime soon. There's still ~70 patches in the original series that I need to understand, rebase and test, including the introduction of the tcg directory.

Sure, I am definitely fine with leaving them separate for now as well :).

Quoted 4 lines

I'd say we merge this as is now, since this patch has no dependencies. Later when I introduce the tcg directory I can move the code there along with the other tcg-only files. I'll take note to come back to the PSCI code as well.

Quoted 6 lines
Quoted 3 lines

I'm confused about the patch ordering :). Why is it easier to move the psci.c compilation target from generic to an if(CONFIG_TCG) only to later move it into a tcg/ directory?

It's a simple patch, so the overhead didn't cross my mind. But you are right, this could go directly into tcg/ without having to put it under CONFIG_TCG first.

I'm sure more like this will follow, and it will be a lot easier on everyone if the pattern is going to be "move tcg specific code to tcg/ and leave generic code in the main directory".

Quoted 6 lines
Quoted 3 lines

Wouldn't it be easier to create a
tcg/ directory from the start and just put it there?

I don't know about "from the start". At this point we cannot have a single commit moving everything into the tcg/ directory because some files still contain tcg + non-tcg code.

Yes, the only thing the initial commit at the start would do is create the directory and ninja config, pretty much nothing else. All follow-on commits then split the currently entangled code into tcg/ once code is clearly separate :).

I believe this was also the approach Claudio took in his patch set last year, and I find it very reasonable. It allows you to stop at any point mid-way.

Quoted 5 lines

We would end up with several
commits moving files into tcg/ along the history, which I think results in a poor experience when inspecting the log later on (git blame and so on). So my idea was to split as much as I can upfront and only later move everything into the directory.

Quite the opposite: Please make sure to move everything slowly at a digestible pace. There is no way you will get 100 patches in at once. Make sure you can cut off at any point in between.

What you basically want is to move from "target/arm is tcg+generic code" to "target/arm is generic, target/arm/tcg is tcg code". You will be in a transitional phase along the way whatever you do, so just make it "target/arm is tcg+generic code, target/arm/tcg is tcg code" while things are in flight and have a final commit that indicates the conversion is done.

Quoted 5 lines

I'm also rebasing this series [1] from 2021, which means I'd rather have small chunks of code moved under CONFIG_TCG that I can build-test with --disable-tcg (even though the build doesn't finish, I can see the number of errors going down), than to move non-tcg code into tcg/ and then pull it out later like in the original series.

I think we're saying the same thing. Please don't move non-tcg code into tcg/. Just move files that are absolutely clearly TCG into tcg/ right from the start. The psci.c is a good example for that. translate*.c and op-helper.c would be another.

Alex

Quoted 3 lines

1- https://lore.kernel.org/r/[email protected]

But hey, that's just my reasoning, no strong feelings about it.

#10 lore raw

Re: [PATCH 1/5] target/arm: only build psci for TCG

Quoted 11 lines · Alexander Graf
Quoted 6 lines
Quoted 3 lines

I'm confused about the patch ordering :). Why is it easier to move the psci.c compilation target from generic to an if(CONFIG_TCG) only to later move it into a tcg/ directory?

It's a simple patch, so the overhead didn't cross my mind. But you are right, this could go directly into tcg/ without having to put it under CONFIG_TCG first.

I'm sure more like this will follow, and it will be a lot easier on everyone if the pattern is going to be "move tcg specific code to tcg/ and leave generic code in the main directory".

Ok, so I'll drop this patch from this series and resend it along with the rest of the code movement to the tcg/ directory.

Quoted 3 lines

Quite the opposite: Please make sure to move everything slowly at a digestible pace. There is no way you will get 100 patches in at once. Make sure you can cut off at any point in between.

I meant having code under CONFIG_TCG first and later moving to tcg/. So we separate moving the code from figuring out if it should be moved. There was no implication of speed, size or indigestibility =).

Quoted 16 lines

What you basically want is to move from "target/arm is tcg+generic code" to "target/arm is generic, target/arm/tcg is tcg code". You will be in a transitional phase along the way whatever you do, so just make it "target/arm is tcg+generic code, target/arm/tcg is tcg code" while things are in flight and have a final commit that indicates the conversion is done.

Quoted 5 lines

I'm also rebasing this series [1] from 2021, which means I'd rather have small chunks of code moved under CONFIG_TCG that I can build-test with --disable-tcg (even though the build doesn't finish, I can see the number of errors going down), than to move non-tcg code into tcg/ and then pull it out later like in the original series.

Quoted 4 lines

I think we're saying the same thing. Please don't move non-tcg code into tcg/. Just move files that are absolutely clearly TCG into tcg/ right from the start. The psci.c is a good example for that. translate*.c and op-helper.c would be another.

Yeah, I think we agree. Thanks for taking the time to spell it out.

#11 lore raw

PATCH 2/5 target/arm: rename handle_semihosting to tcg_handle_semihosting

In reply to Fabiano Rosas

From: Claudio Fontana <[email protected]>

make it clearer from the name that this is a tcg-only function.

---
This function moves elsewhere in the original series, but the name change doesn't need to wait.

Originally from:
[RFC v14 38/80] target/arm: rename handle_semihosting to tcg_handle_semihosting
https://lore.kernel.org/r/[email protected]

---
 target/arm/helper.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
target/arm/helper.c +2 -2
index d8c8223ec3..45bf164a07 100644
@@ -10252,7 +10252,7 @@ static void arm_cpu_do_interrupt_aarch64(CPUState *cs)
  * trapped to the hypervisor in KVM.
  */
 #ifdef CONFIG_TCG
-static void handle_semihosting(CPUState *cs)
+static void tcg_handle_semihosting(CPUState *cs)
 {
     ARMCPU *cpu = ARM_CPU(cs);
     CPUARMState *env = &cpu->env;
@@ -10313,7 +10313,7 @@ void arm_cpu_do_interrupt(CPUState *cs)
      */
 #ifdef CONFIG_TCG
     if (cs->exception_index == EXCP_SEMIHOST) {
-        handle_semihosting(cs);
+        tcg_handle_semihosting(cs);
         return;
     }
 #endif
Signature · 1 line
2.35.3
Reviewed-by on v1 2/5
#12 lore raw

Re: [PATCH 2/5] target/arm: rename handle_semihosting to tcg_handle_semihosting

Quoted 6 lines · Fabiano Rosas

From: Claudio Fontana <[email protected]>

make it clearer from the name that this is a tcg-only function.

Quoted 7 lines

---
This function moves elsewhere in the original series, but the name change doesn't need to wait.

Originally from:
[RFC v14 38/80] target/arm: rename handle_semihosting to tcg_handle_semihosting
https://lore.kernel.org/r/[email protected]

r~

#13 lore raw

PATCH 3/5 target/arm: wrap semihosting and psci calls with tcg_enabled

In reply to Fabiano Rosas

From: Claudio Fontana <[email protected]>

for "all" builds (tcg + kvm), we want to avoid doing
the psci and semihosting checks if tcg is built-in, but not enabled.

---
Originally from:
[RFC v14 39/80] target/arm: replace CONFIG_TCG with tcg_enabled
https://lore.kernel.org/r/[email protected]

---
 target/arm/helper.c | 31 +++++++++++++++++--------------
 1 file changed, 17 insertions(+), 14 deletions(-)
target/arm/helper.c +17 -14
index 45bf164a07..9d0a53cb00 100644
@@ -26,6 +26,7 @@
 #include "sysemu/cpus.h"
 #include "sysemu/cpu-timers.h"
 #include "sysemu/kvm.h"
+#include "sysemu/tcg.h"
 #include "qemu/range.h"
 #include "qapi/qapi-commands-machine-target.h"
 #include "qapi/error.h"
@@ -10300,23 +10301,25 @@ void arm_cpu_do_interrupt(CPUState *cs)
                       env->exception.syndrome);
     }
 
-    if (arm_is_psci_call(cpu, cs->exception_index)) {
-        arm_handle_psci_call(cpu);
-        qemu_log_mask(CPU_LOG_INT, "...handled as PSCI call\n");
-        return;
-    }
+    if (tcg_enabled()) {
+        if (arm_is_psci_call(cpu, cs->exception_index)) {
+            arm_handle_psci_call(cpu);
+            qemu_log_mask(CPU_LOG_INT, "...handled as PSCI call\n");
+            return;
+        }
 
-    /*
-     * Semihosting semantics depend on the register width of the code
-     * that caused the exception, not the target exception level, so
-     * must be handled here.
-     */
+        /*
+         * Semihosting semantics depend on the register width of the code
+         * that caused the exception, not the target exception level, so
+         * must be handled here.
+         */
 #ifdef CONFIG_TCG
-    if (cs->exception_index == EXCP_SEMIHOST) {
-        tcg_handle_semihosting(cs);
-        return;
-    }
+        if (cs->exception_index == EXCP_SEMIHOST) {
+            tcg_handle_semihosting(cs);
+            return;
+        }
 #endif
+    }
 
     /* Hooks may change global state so BQL should be held, also the
      * BQL needs to be held for any modification of
Signature · 1 line
2.35.3
Reviewed-by on v1 3/5
#14 lore raw

Re: [PATCH 3/5] target/arm: wrap semihosting and psci calls with tcg_enabled

target/arm/helper.c void arm_cpu_do_interrupt(CPUState *cs) · line 10305

Quoted 7 lines · Fabiano Rosas
target/arm/helper.c +2 -5
-    if (arm_is_psci_call(cpu, cs->exception_index)) {
-        arm_handle_psci_call(cpu);
-        qemu_log_mask(CPU_LOG_INT, "...handled as PSCI call\n");
-        return;
-    }
+    if (tcg_enabled()) {
+        if (arm_is_psci_call(cpu, cs->exception_index)) {

This could be

    if (tcg_enabled() && arm_is_psci_call(...))

because...

Quoted 6 lines
target/arm/helper.c +1 -5
-    /*
-     * Semihosting semantics depend on the register width of the code
-     * that caused the exception, not the target exception level, so
-     * must be handled here.
-     */
+        /*

target/arm/helper.c void arm_cpu_do_interrupt(CPUState *cs) · line 10317

Quoted 10 lines
target/arm/helper.c +5 -4
+         * Semihosting semantics depend on the register width of the code
+         * that caused the exception, not the target exception level, so
+         * must be handled here.
+         */
  #ifdef CONFIG_TCG
-    if (cs->exception_index == EXCP_SEMIHOST) {
-        tcg_handle_semihosting(cs);
-        return;
-    }
+        if (cs->exception_index == EXCP_SEMIHOST) {

If you were able to replace the ifdef, that would be one thing, but since you aren't I don't think this requires a separate check. There is no way to generate EXCP_SEMIHOST except via TCG.

I guess I don't insist, since you're working toward Claudio's much larger patch set, so

r~

#15 lore raw

Re: [PATCH 3/5] target/arm: wrap semihosting and psci calls with tcg_enabled

Quoted 32 lines · Richard Henderson

target/arm/helper.c void arm_cpu_do_interrupt(CPUState *cs) · line 10305

Quoted 7 lines · Fabiano Rosas
target/arm/helper.c +2 -5
-    if (arm_is_psci_call(cpu, cs->exception_index)) {
-        arm_handle_psci_call(cpu);
-        qemu_log_mask(CPU_LOG_INT, "...handled as PSCI call\n");
-        return;
-    }
+    if (tcg_enabled()) {
+        if (arm_is_psci_call(cpu, cs->exception_index)) {

This could be

     if (tcg_enabled() && arm_is_psci_call(...))

because...

Quoted 6 lines
target/arm/helper.c +1 -5
-    /*
-     * Semihosting semantics depend on the register width of the code
-     * that caused the exception, not the target exception level, so
-     * must be handled here.
-     */
+        /*

target/arm/helper.c void arm_cpu_do_interrupt(CPUState *cs) · line 10317

Quoted 10 lines
target/arm/helper.c +5 -4
+         * Semihosting semantics depend on the register width of the code
+         * that caused the exception, not the target exception level, so
+         * must be handled here.
+         */
  #ifdef CONFIG_TCG
-    if (cs->exception_index == EXCP_SEMIHOST) {
-        tcg_handle_semihosting(cs);
-        return;
-    }
+        if (cs->exception_index == EXCP_SEMIHOST) {
Quoted 3 lines

If you were able to replace the ifdef, that would be one thing, but since you aren't I don't think this requires a separate check. There is no way to generate EXCP_SEMIHOST except via TCG.

Right, I had to keep the ifdef that was being removed in the original patch because tcg_handle_semihosting had moved elsewhere in Claudio's series.

Quoted 4 lines

I guess I don't insist, since you're working toward Claudio's much larger patch set, so

If you don't mind I'll leave like this then, unless this comes to a v2.

Thank you

#16 lore raw

PATCH 4/5 target/arm: wrap call to aarch64_sve_change_el in tcg_enabled()

In reply to Fabiano Rosas

From: Claudio Fontana <[email protected]>

---
Originally from:
[RFC v14 42/80] target/arm: wrap call to aarch64_sve_change_el in tcg_enabled()
https://lore.kernel.org/r/[email protected]

---
 target/arm/helper.c | 12 +++++++-----
 1 file changed, 7 insertions(+), 5 deletions(-)
target/arm/helper.c +7 -5
index 9d0a53cb00..92624e2910 100644
@@ -10066,11 +10066,13 @@ static void arm_cpu_do_interrupt_aarch64(CPUState *cs)
     unsigned int cur_el = arm_current_el(env);
     int rt;
 
-    /*
-     * Note that new_el can never be 0.  If cur_el is 0, then
-     * el0_a64 is is_a64(), else el0_a64 is ignored.
-     */
-    aarch64_sve_change_el(env, cur_el, new_el, is_a64(env));
+    if (tcg_enabled()) {
+        /*
+         * Note that new_el can never be 0.  If cur_el is 0, then
+         * el0_a64 is is_a64(), else el0_a64 is ignored.
+         */
+        aarch64_sve_change_el(env, cur_el, new_el, is_a64(env));
+    }
 
     if (cur_el < new_el) {
         /* Entry vector offset depends on whether the implemented EL
Signature · 1 line
2.35.3
#17 lore raw

PATCH 5/5 target/arm: only perform TCG cpu and machine inits if TCG enabled

In reply to Fabiano Rosas

From: Claudio Fontana <[email protected]>

of note, cpreg lists were previously initialized by TCG first, and then thrown away and replaced with the data coming from KVM.

Now we just initialize once, either for TCG or for KVM.

[moved arm_cpu_register_gdb_regs_for_features out of tcg_enabled]

---
We still need to tell GDB about the register sets even when running with KVM.

Originally from:
[RFC v14 16/80] target/arm: only perform TCG cpu and machine inits if TCG enabled
https://lore.kernel.org/r/[email protected]

---
 target/arm/cpu.c     | 31 ++++++++++++----------
 target/arm/kvm.c     | 18 ++++++-------
 target/arm/kvm_arm.h |  3 +--
 target/arm/machine.c | 61 ++++++++++++++++++++++++--------------------
 4 files changed, 62 insertions(+), 51 deletions(-)
target/arm/cpu.c +18 -13
index 38d066c294..a0c77ba153 100644
@@ -523,9 +523,11 @@ static void arm_cpu_reset(DeviceState *dev)
     }
 #endif
 
-    hw_breakpoint_update_all(cpu);
-    hw_watchpoint_update_all(cpu);
-    arm_rebuild_hflags(env);
+    if (tcg_enabled()) {
+        hw_breakpoint_update_all(cpu);
+        hw_watchpoint_update_all(cpu);
+        arm_rebuild_hflags(env);
+    }
 }
 
 #ifndef CONFIG_USER_ONLY
@@ -1597,6 +1599,7 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         }
     }
 
+#ifdef CONFIG_TCG
     {
         uint64_t scale;
 
@@ -1622,7 +1625,8 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         cpu->gt_timer[GTIMER_HYPVIRT] = timer_new(QEMU_CLOCK_VIRTUAL, scale,
                                                   arm_gt_hvtimer_cb, cpu);
     }
-#endif
+#endif /* CONFIG_TCG */
+#endif /* !CONFIG_USER_ONLY */
 
     cpu_exec_realizefn(cs, &local_err);
     if (local_err != NULL) {
@@ -1940,17 +1944,16 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         unset_feature(env, ARM_FEATURE_PMU);
     }
     if (arm_feature(env, ARM_FEATURE_PMU)) {
-        pmu_init(cpu);
-
-        if (!kvm_enabled()) {
+        if (tcg_enabled()) {
+            pmu_init(cpu);
             arm_register_pre_el_change_hook(cpu, &pmu_pre_el_change, 0);
             arm_register_el_change_hook(cpu, &pmu_post_el_change, 0);
-        }
 
 #ifndef CONFIG_USER_ONLY
-        cpu->pmu_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, arm_pmu_timer_cb,
-                cpu);
+            cpu->pmu_timer = timer_new_ns(QEMU_CLOCK_VIRTUAL, arm_pmu_timer_cb,
+                                          cpu);
 #endif
+        }
     } else {
         cpu->isar.id_aa64dfr0 =
             FIELD_DP64(cpu->isar.id_aa64dfr0, ID_AA64DFR0, PMUVER, 0);
@@ -2046,10 +2049,12 @@ static void arm_cpu_realizefn(DeviceState *dev, Error **errp)
         set_feature(env, ARM_FEATURE_VBAR);
     }
 
-    register_cp_regs_for_features(cpu);
-    arm_cpu_register_gdb_regs_for_features(cpu);
+    if (tcg_enabled()) {
+        register_cp_regs_for_features(cpu);
+        init_cpreg_list(cpu);
+    }
 
-    init_cpreg_list(cpu);
+    arm_cpu_register_gdb_regs_for_features(cpu);
 
 #ifndef CONFIG_USER_ONLY
     MachineState *ms = MACHINE(qdev_get_machine());
target/arm/kvm.c +9 -9
index f022c644d2..2f01c26f54 100644
@@ -438,9 +438,11 @@ static uint64_t *kvm_arm_get_cpreg_ptr(ARMCPU *cpu, uint64_t regidx)
     return &cpu->cpreg_values[res - cpu->cpreg_indexes];
 }
 
-/* Initialize the ARMCPU cpreg list according to the kernel's
- * definition of what CPU registers it knows about (and throw away
- * the previous TCG-created cpreg list).
+/*
+ * Initialize the ARMCPU cpreg list according to the kernel's
+ * definition of what CPU registers it knows about.
+ *
+ * The parallel for TCG is init_cpreg_list()
  */
 int kvm_arm_init_cpreg_list(ARMCPU *cpu)
 {
@@ -482,12 +484,10 @@ int kvm_arm_init_cpreg_list(ARMCPU *cpu)
         arraylen++;
     }
 
-    cpu->cpreg_indexes = g_renew(uint64_t, cpu->cpreg_indexes, arraylen);
-    cpu->cpreg_values = g_renew(uint64_t, cpu->cpreg_values, arraylen);
-    cpu->cpreg_vmstate_indexes = g_renew(uint64_t, cpu->cpreg_vmstate_indexes,
-                                         arraylen);
-    cpu->cpreg_vmstate_values = g_renew(uint64_t, cpu->cpreg_vmstate_values,
-                                        arraylen);
+    cpu->cpreg_indexes = g_new(uint64_t, arraylen);
+    cpu->cpreg_values = g_new(uint64_t, arraylen);
+    cpu->cpreg_vmstate_indexes = g_new(uint64_t, arraylen);
+    cpu->cpreg_vmstate_values = g_new(uint64_t, arraylen);
     cpu->cpreg_array_len = arraylen;
     cpu->cpreg_vmstate_array_len = arraylen;
 
target/arm/kvm_arm.h +1 -2
index 99017b635c..41de2a7cf1 100644
@@ -70,8 +70,7 @@ void kvm_arm_register_device(MemoryRegion *mr, uint64_t devid, uint64_t group,
  * @cpu: ARMCPU
  *
  * Initialize the ARMCPU cpreg list according to the kernel's
- * definition of what CPU registers it knows about (and throw away
- * the previous TCG-created cpreg list).
+ * definition of what CPU registers it knows about.
  *
  * Returns: 0 if success, else < 0 error code
  */
target/arm/machine.c +34 -27

105 lines not on this page · raw message

Signature · 1 line
2.35.3
Reviewed-by on v1 5/5
#18 lore raw

Re: [PATCH 5/5] target/arm: only perform TCG cpu and machine inits if TCG enabled

Quoted 6 lines · Fabiano Rosas
target/arm/machine.c +0 -6
-    /*
-     * Misaligned thumb pc is architecturally impossible.
-     * We have an assert in thumb_tr_translate_insn to verify this.
-     * Fail an incoming migrate to avoid this assert.
-     */
-    if (!is_a64(env) && env->thumb && (env->regs[15] & 1)) {

target/arm/machine.c static int cpu_post_load(void *opaque, int version_id) · line 841

Quoted 10 lines
target/arm/machine.c +8 -2
-        return -1;
-    }
+        /*
+         * Misaligned thumb pc is architecturally impossible.
+         * We have an assert in thumb_tr_translate_insn to verify this.
+         * Fail an incoming migrate to avoid this assert.
+         */
+        if (!is_a64(env) && env->thumb && (env->regs[15] & 1)) {
+            return -1;
+        }

This is a sanity check rejecting malformed vmsave. While hw virt won't have the same assert as mentioned in the comment, it won't be happy and will raise some sort of cpu exception later. I think it's better to reject the bad vmload early. I suppose we could expand the comment to that effect, so that it doesn't appear to be wholly tcg inspired.

Otherwise,

r~

#19 lore raw

Re: [PATCH 5/5] target/arm: only perform TCG cpu and machine inits if TCG enabled

Quoted 18 lines · Richard Henderson
Quoted 6 lines · Fabiano Rosas
target/arm/machine.c +0 -6
-    /*
-     * Misaligned thumb pc is architecturally impossible.
-     * We have an assert in thumb_tr_translate_insn to verify this.
-     * Fail an incoming migrate to avoid this assert.
-     */
-    if (!is_a64(env) && env->thumb && (env->regs[15] & 1)) {

target/arm/machine.c static int cpu_post_load(void *opaque, int version_id) · line 841

Quoted 10 lines
target/arm/machine.c +8 -2
-        return -1;
-    }
+        /*
+         * Misaligned thumb pc is architecturally impossible.
+         * We have an assert in thumb_tr_translate_insn to verify this.
+         * Fail an incoming migrate to avoid this assert.
+         */
+        if (!is_a64(env) && env->thumb && (env->regs[15] & 1)) {
+            return -1;
+        }
Quoted 4 lines

This is a sanity check rejecting malformed vmsave. While hw virt won't have the same assert as mentioned in the comment, it won't be happy and will raise some sort of cpu exception later. I think it's better to reject the bad vmload early. I suppose we could expand the comment to that effect, so that it doesn't appear to be wholly tcg inspired.

I see, I'll leave it out of tcg_enabled() and update the comment.

Quoted 5 lines

Otherwise,

r~