Skip to contentSkip to conversation

consolidate do_bug_frame() / bug_fn_t

PATCH In review Jan Beulich 3 participants 8 messages Jan 9, 2024 – Jan 25, 2024

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 consolidate do_bug_frame() / bug_fn_t

The type not being used in do_bug_frame() is suspicious. Apparently that's solely because the type uses a pointer-to-const parameter, when run_in_exception_handler() wants functions taking pointer-to-non- const. Drop the const, in turn requiring Arm's do_bug_frame() to also have its const dropped. This then brings that function also closer to the common one, with Arm's use of vaddr_t remaining as a difference.

---
This is an alternative proposal to
https://lists.xen.org/archives/html/xen-devel/2023-12/msg01385.html,
albeit without paving a road towards Andrew's desire of getting rid of show_execution_state_nonconst() again. Retaining (and propagating) the const would imply the need to cast away the const-ness somewhere on (at least) the path to invoking gdb stub code. Personally I'm averse to such casting away of const-ness ...

xen/arch/arm/include/asm/traps.h +1 -1
@@ -69,7 +69,7 @@ void do_cp(struct cpu_user_regs *regs, c
 void do_trap_smc(struct cpu_user_regs *regs, const union hsr hsr);
 void do_trap_hvc_smccc(struct cpu_user_regs *regs);
 
-int do_bug_frame(const struct cpu_user_regs *regs, vaddr_t pc);
+int do_bug_frame(struct cpu_user_regs *regs, vaddr_t pc);
 
 void noreturn do_unexpected_trap(const char *msg,
                                  const struct cpu_user_regs *regs);
xen/arch/arm/traps.c +1 -1
@@ -1202,7 +1202,7 @@ void do_unexpected_trap(const char *msg,
     panic("CPU%d: Unexpected Trap: %s\n", smp_processor_id(), msg);
 }
 
-int do_bug_frame(const struct cpu_user_regs *regs, vaddr_t pc)
+int do_bug_frame(struct cpu_user_regs *regs, vaddr_t pc)
 {
     const struct bug_frame *bug = NULL;
     const char *prefix = "", *filename, *predicate;
xen/common/bug.c +1 -5
@@ -63,14 +63,10 @@ int do_bug_frame(struct cpu_user_regs *r
 
     if ( id == BUGFRAME_run_fn )
     {
-        void (*fn)(struct cpu_user_regs *) = bug_ptr(bug);
+        bug_fn_t *fn = bug_ptr(bug);
 
         fn(regs);
 
-        /* Re-enforce consistent types, because of the casts involved. */
-        if ( false )
-            run_in_exception_handler(fn);
-
         return id;
     }
 
xen/include/xen/bug.h +2 -3
@@ -101,12 +101,11 @@ struct bug_frame {
 #endif
 
 struct cpu_user_regs;
-typedef void bug_fn_t(const struct cpu_user_regs *regs);
+typedef void bug_fn_t(struct cpu_user_regs *regs);
 
 #ifndef run_in_exception_handler
 
-static void always_inline run_in_exception_handler(
-    void (*fn)(struct cpu_user_regs *regs))
+static void always_inline run_in_exception_handler(bug_fn_t *fn)
 {
     BUG_FRAME(BUGFRAME_run_fn, 0, fn, 0, NULL);
 }
#1 lore raw

Hi Jan, Andrew,

I managed to get back to read the mailing list and noticed this patch.

Is it still relevant and needs to be reviewed?

Are there any outstanding disagreements between maintainers on the approach to take here? Or should I just go ahead and review it?

Quoted 75 lines · Jan Beulich

The type not being used in do_bug_frame() is suspicious. Apparently that's solely because the type uses a pointer-to-const parameter, when run_in_exception_handler() wants functions taking pointer-to-non- const. Drop the const, in turn requiring Arm's do_bug_frame() to also have its const dropped. This then brings that function also closer to the common one, with Arm's use of vaddr_t remaining as a difference.

---
This is an alternative proposal to
https://lists.xen.org/archives/html/xen-devel/2023-12/msg01385.html,
albeit without paving a road towards Andrew's desire of getting rid of show_execution_state_nonconst() again. Retaining (and propagating) the const would imply the need to cast away the const-ness somewhere on (at least) the path to invoking gdb stub code. Personally I'm averse to such casting away of const-ness ...

xen/arch/arm/include/asm/traps.h +1 -1
@@ -69,7 +69,7 @@ void do_cp(struct cpu_user_regs *regs, c
 void do_trap_smc(struct cpu_user_regs *regs, const union hsr hsr);
 void do_trap_hvc_smccc(struct cpu_user_regs *regs);
 
-int do_bug_frame(const struct cpu_user_regs *regs, vaddr_t pc);
+int do_bug_frame(struct cpu_user_regs *regs, vaddr_t pc);
 
 void noreturn do_unexpected_trap(const char *msg,
                                  const struct cpu_user_regs *regs);
xen/arch/arm/traps.c +1 -1
@@ -1202,7 +1202,7 @@ void do_unexpected_trap(const char *msg,
     panic("CPU%d: Unexpected Trap: %s\n", smp_processor_id(), msg);
 }
 
-int do_bug_frame(const struct cpu_user_regs *regs, vaddr_t pc)
+int do_bug_frame(struct cpu_user_regs *regs, vaddr_t pc)
 {
     const struct bug_frame *bug = NULL;
     const char *prefix = "", *filename, *predicate;
xen/common/bug.c +1 -5
@@ -63,14 +63,10 @@ int do_bug_frame(struct cpu_user_regs *r
 
     if ( id == BUGFRAME_run_fn )
     {
-        void (*fn)(struct cpu_user_regs *) = bug_ptr(bug);
+        bug_fn_t *fn = bug_ptr(bug);
 
         fn(regs);
 
-        /* Re-enforce consistent types, because of the casts involved. */
-        if ( false )
-            run_in_exception_handler(fn);
-
         return id;
     }
 
xen/include/xen/bug.h +2 -3
@@ -101,12 +101,11 @@ struct bug_frame {
 #endif
 
 struct cpu_user_regs;
-typedef void bug_fn_t(const struct cpu_user_regs *regs);
+typedef void bug_fn_t(struct cpu_user_regs *regs);
 
 #ifndef run_in_exception_handler
 
-static void always_inline run_in_exception_handler(
-    void (*fn)(struct cpu_user_regs *regs))
+static void always_inline run_in_exception_handler(bug_fn_t *fn)
 {
     BUG_FRAME(BUGFRAME_run_fn, 0, fn, 0, NULL);
 }
#2 lore raw
Quoted 6 lines · Stefano Stabellini

I managed to get back to read the mailing list and noticed this patch.

Is it still relevant and needs to be reviewed?

Are there any outstanding disagreements between maintainers on the approach to take here? Or should I just go ahead and review it?

It is still relevant from my pov, and everything that may be controversial is said ...

Quoted 8 lines · Jan Beulich
Quoted 7 lines

The type not being used in do_bug_frame() is suspicious. Apparently that's solely because the type uses a pointer-to-const parameter, when run_in_exception_handler() wants functions taking pointer-to-non- const. Drop the const, in turn requiring Arm's do_bug_frame() to also have its const dropped. This then brings that function also closer to the common one, with Arm's use of vaddr_t remaining as a difference.

Quoted 9 lines
Quoted 9 lines

---
This is an alternative proposal to
https://lists.xen.org/archives/html/xen-devel/2023-12/msg01385.html,
albeit without paving a road towards Andrew's desire of getting rid of show_execution_state_nonconst() again. Retaining (and propagating) the const would imply the need to cast away the const-ness somewhere on (at least) the path to invoking gdb stub code. Personally I'm averse to such casting away of const-ness ...

... here. Without Andrew commenting, I'm afraid it's not actually clear whether he objects to this approach, or is meaning to tolerate it silently, or actually views it as a step in a good direction, even if not quite getting where earlier on he thought we may want to move to.

Jan

#3 lore raw
Quoted 9 lines · Jan Beulich
Quoted 6 lines · Stefano Stabellini

I managed to get back to read the mailing list and noticed this patch.

Is it still relevant and needs to be reviewed?

Are there any outstanding disagreements between maintainers on the approach to take here? Or should I just go ahead and review it?

It is still relevant from my pov, and everything that may be controversial is said ...

BUGFRAME_* cannot legitimately modify the interrupted context.  Two are fatal paths, and other two are side-effect-less as far as C can tell.

So the infrastructure ought to take a const pointer.

The reason why this pointer is non-const is to do with the interaction of the serial and keyhandler infrastructures.  Because we're adjusting that for other reasons, I was hoping it would subsequently be easy to switch Xen to being properly const in this regard.

Turns out it is:


https://gitlab.com/xen-project/people/andyhhp/xen/-/commit/4f857075005da1d28632e4f9198c2e7d0f404b9a

with a couple of caveats.  (Only the buster-gcc-ibt run failed, so I've got some cf_check-ing to adjust, but all the other builds worked fine).

To make the serial code compile, I ended up having to revert patch 2 of the regs series, which I believe is safe to do following patch 3-5 which un-plumb the regs pointer deeper in the call chain.  If this is turns out to be true, then the patch ought to be added and reverted in the same series so it isn't left hanging about after the fact.

The _$X_poll() functions are used in timer context, which means there's an outer regs context already latched, and that's arguably a better context to use anyway for 'd'.

This in turn allows us to remove a #UD from a fast(ish) path, and remove some per-cpu or static variables which are just used for non-standard parameter passing because run_in_exception_handler() doesn't let you pass any.

This leaves the '%' debugger infrastructure.  Being a debugger, it's making arbitrary changes anyway and I'd much rather cast away constness for a debugger, than to keep everything else mutable when it oughtn't to be.

If absolutely nothing else, registration and handling '%' ought to be from x86 code rather than common code, which would remove the do_debugger_trap_fatal() layering violation.

But, the more I look into the gdbstub the more I'm convinced that it doesn't work.  For example, this gem:

/* Resuming after we've stopped used to work, but more through luck than any actual intention.  It doesn't at the moment. */

From c/s b69f92f3012 in July 2004, and more specifically the commit which added the gdbstub functionality to begin with.  I.e. it doesn't appear to have ever supported more than "poke around in the crashed state".  In the 2 decades that noone has fixed this, we've gained far better technologies for doing this, such as running it in a VM.

I am going to submit some patches deleting gdbstub.  It clearly had not much value to begin with, and is not definitely not worth the problems it is creating in adjacent code these days.

~Andrew

#4 lore raw
Quoted 30 lines · Andrew Cooper
Quoted 9 lines · Jan Beulich
Quoted 6 lines · Stefano Stabellini

I managed to get back to read the mailing list and noticed this patch.

Is it still relevant and needs to be reviewed?

Are there any outstanding disagreements between maintainers on the approach to take here? Or should I just go ahead and review it?

It is still relevant from my pov, and everything that may be controversial is said ...

BUGFRAME_* cannot legitimately modify the interrupted context.  Two are fatal paths, and other two are side-effect-less as far as C can tell.

So the infrastructure ought to take a const pointer.

The reason why this pointer is non-const is to do with the interaction of the serial and keyhandler infrastructures.  Because we're adjusting that for other reasons, I was hoping it would subsequently be easy to switch Xen to being properly const in this regard.

Turns out it is:


https://gitlab.com/xen-project/people/andyhhp/xen/-/commit/4f857075005da1d28632e4f9198c2e7d0f404b9a

with a couple of caveats.  (Only the buster-gcc-ibt run failed, so I've got some cf_check-ing to adjust, but all the other builds worked fine).

Quoted 5 lines

To make the serial code compile, I ended up having to revert patch 2 of the regs series, which I believe is safe to do following patch 3-5 which un-plumb the regs pointer deeper in the call chain.  If this is turns out to be true, then the patch ought to be added and reverted in the same series so it isn't left hanging about after the fact.

Hmm, I'm not sure I see how reverting that would end up working. However, aiui you need to revert primarily for the non-const-ness of the pointers involved in [gs]et_irq_regs(). I wonder whether, if we followed your underlying thought here, those shouldn't be const-ified then anyway.

Quoted 3 lines

The _$X_poll() functions are used in timer context, which means there's an outer regs context already latched, and that's arguably a better context to use anyway for 'd'.

If the timer happens to run on an idle vCPU, what "outer regs context" would we have there?

Quoted 26 lines

This in turn allows us to remove a #UD from a fast(ish) path, and remove some per-cpu or static variables which are just used for non-standard parameter passing because run_in_exception_handler() doesn't let you pass any.

This leaves the '%' debugger infrastructure.  Being a debugger, it's making arbitrary changes anyway and I'd much rather cast away constness for a debugger, than to keep everything else mutable when it oughtn't to be.

If absolutely nothing else, registration and handling '%' ought to be from x86 code rather than common code, which would remove the do_debugger_trap_fatal() layering violation.

But, the more I look into the gdbstub the more I'm convinced that it doesn't work.  For example, this gem:

/* Resuming after we've stopped used to work, but more through luck than any actual intention.  It doesn't at the moment. */

From c/s b69f92f3012 in July 2004, and more specifically the commit which added the gdbstub functionality to begin with.  I.e. it doesn't appear to have ever supported more than "poke around in the crashed state".  In the 2 decades that noone has fixed this, we've gained far better technologies for doing this, such as running it in a VM.

Quoted 3 lines

I am going to submit some patches deleting gdbstub.  It clearly had not much value to begin with, and is not definitely not worth the problems it is creating in adjacent code these days.

All fine. Still I wonder whether in the meantime this patch isn't an improvement on its own, and hence whether the const couldn't sensibly be added subsequently.

Jan

#5 lore raw

Answering out of order:

Quoted 3 lines · Jan Beulich

All fine. Still I wonder whether in the meantime this patch isn't an improvement on its own, and hence whether the const couldn't sensibly be added subsequently.

We have a while until 4.19 goes out.  I would prefer to try and get this untangled properly, because half of your patch is in the opposite direction for getting the const-ness working.

If we start to hit rc1 and the const-ness isn't complete, we can revisit.

Regarding the removal of gdb-stub.  I'd like to get that done, and rebase the remainder of the IRQ series over it, because it will reduce the churn in your IRQ series.  I'm happy to do the rebase if you want.

Quoted 11 lines · Andrew Cooper
Quoted 5 lines

To make the serial code compile, I ended up having to revert patch 2 of the regs series, which I believe is safe to do following patch 3-5 which un-plumb the regs pointer deeper in the call chain.  If this is turns out to be true, then the patch ought to be added and reverted in the same series so it isn't left hanging about after the fact.

Hmm, I'm not sure I see how reverting that would end up working. However, aiui you need to revert primarily for the non-const-ness of the pointers involved in [gs]et_irq_regs(). I wonder whether, if we followed your underlying thought here, those shouldn't be const-ified then anyway.

Quoted 5 lines
Quoted 3 lines

The _$X_poll() functions are used in timer context, which means there's an outer regs context already latched, and that's arguably a better context to use anyway for 'd'.

If the timer happens to run on an idle vCPU, what "outer regs context" would we have there?

The only reason the serial infrastructure was setting up a fake IRQ context was because it was using run_in_exception_handler().

But I (think I have) removed that fully (and it simplifies things more than I was hoping).

With '%' deleted, it's only 'd' that cares, isn't it?  And that's "dump the current regs", rather than wanting something else.

~Andrew

#6 lore raw
Quoted 11 lines · Andrew Cooper

Answering out of order:

Quoted 3 lines · Jan Beulich

All fine. Still I wonder whether in the meantime this patch isn't an improvement on its own, and hence whether the const couldn't sensibly be added subsequently.

We have a while until 4.19 goes out.  I would prefer to try and get this untangled properly, because half of your patch is in the opposite direction for getting the const-ness working.

Quoted 6 lines

If we start to hit rc1 and the const-ness isn't complete, we can revisit.

Regarding the removal of gdb-stub.  I'd like to get that done, and rebase the remainder of the IRQ series over it, because it will reduce the churn in your IRQ series.  I'm happy to do the rebase if you want.

Shouldn't be overly difficult, so I guess I can as well do it before (eventually) re-submitting.

Quoted 17 lines
Quoted 11 lines · Andrew Cooper
Quoted 5 lines

To make the serial code compile, I ended up having to revert patch 2 of the regs series, which I believe is safe to do following patch 3-5 which un-plumb the regs pointer deeper in the call chain.  If this is turns out to be true, then the patch ought to be added and reverted in the same series so it isn't left hanging about after the fact.

Hmm, I'm not sure I see how reverting that would end up working. However, aiui you need to revert primarily for the non-const-ness of the pointers involved in [gs]et_irq_regs(). I wonder whether, if we followed your underlying thought here, those shouldn't be const-ified then anyway.

Quoted 5 lines
Quoted 3 lines

The _$X_poll() functions are used in timer context, which means there's an outer regs context already latched, and that's arguably a better context to use anyway for 'd'.

If the timer happens to run on an idle vCPU, what "outer regs context" would we have there?

Quoted 7 lines

The only reason the serial infrastructure was setting up a fake IRQ context was because it was using run_in_exception_handler().

But I (think I have) removed that fully (and it simplifies things more than I was hoping).

With '%' deleted, it's only 'd' that cares, isn't it?

Yes.

Quoted 1 line

  And that's "dump the current regs", rather than wanting something else.

Question remains though: What are "current regs" when on an idle vCPU? But perhaps that will clarify itself once I see how you "removed that fully".

Jan

#7 lore raw
In reply to Andrew Cooper
Quoted 30 lines · Andrew Cooper
Quoted 9 lines · Jan Beulich
Quoted 6 lines · Stefano Stabellini

I managed to get back to read the mailing list and noticed this patch.

Is it still relevant and needs to be reviewed?

Are there any outstanding disagreements between maintainers on the approach to take here? Or should I just go ahead and review it?

It is still relevant from my pov, and everything that may be controversial is said ...

BUGFRAME_* cannot legitimately modify the interrupted context.  Two are fatal paths, and other two are side-effect-less as far as C can tell.

So the infrastructure ought to take a const pointer.

The reason why this pointer is non-const is to do with the interaction of the serial and keyhandler infrastructures.  Because we're adjusting that for other reasons, I was hoping it would subsequently be easy to switch Xen to being properly const in this regard.

Turns out it is:


https://gitlab.com/xen-project/people/andyhhp/xen/-/commit/4f857075005da1d28632e4f9198c2e7d0f404b9a

with a couple of caveats.  (Only the buster-gcc-ibt run failed, so I've got some cf_check-ing to adjust, but all the other builds worked fine).

Quoted 5 lines

To make the serial code compile, I ended up having to revert patch 2 of the regs series, which I believe is safe to do following patch 3-5 which un-plumb the regs pointer deeper in the call chain.  If this is turns out to be true, then the patch ought to be added and reverted in the same series so it isn't left hanging about after the fact.

Looking further into this, I can't see how reverting ought to be possible, even less so specifically after patch 5. Patches 4 and 5 merely eliminate now unused parameters. Hence imo if it could be reverted after 5, it also ought to be fine to revert after 3. Which in turn it would mean it's not needed at all. Which I simply cannot see (yet?).

Jan