Skip to contentSkip to conversation

system: Define ram_addr_t to be always uint64_t

PATCH Superseded Peter Xu 4 participants 6 messages Aug 14 – Aug 18

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore

Peter Xu

author
#0 lore raw

PATCH 1/1 system: Define ram_addr_t to be always uint64_t

QEMU's 32bit host support was deprecated since 10.0 and removed in 11.0, at least the system emulation part. Now it's safe to move ram_addr_t completely over to uint64_t.

It should be almost the same as uintptr_t as before for !Xen, except that on some systems (like MacOS) uintptr_t and uint64_t can be typed slightly differently, causing unnecessary compiler warnings when use them in a mixture way.

Hopefully, this change also makes it clear that ram_addr_t is never used as a host pointer in any form, but only an internal QEMU integer based address space for allocating ramblocks.

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

Cc Paolo Bonzini <[email protected]>
Cc Philippe Mathieu-Daudé <[email protected]>
---
 include/system/ram_addr.h | 6 ------
 1 file changed, 6 deletions(-)
include/system/ram_addr.h +0 -6
index 129f6b8757..06caca1506 100644
@@ -15,15 +15,9 @@
 #define RAM_ADDR_H
 
 /* address in the RAM (different from a physical address) */
-#if defined(CONFIG_XEN_BACKEND)
 typedef uint64_t ram_addr_t;
 #  define RAM_ADDR_MAX UINT64_MAX
 #  define RAM_ADDR_FMT "%" PRIx64
-#else
-typedef uintptr_t ram_addr_t;
-#  define RAM_ADDR_MAX UINTPTR_MAX
-#  define RAM_ADDR_FMT "%" PRIxPTR
-#endif
 
 #define DIRTY_MEMORY_VGA       0
 #define DIRTY_MEMORY_CODE      1
Signature · 1 line
2.54.0
Reviewed-by on v1 1/1
#1 lore raw
Quoted 2 lines · Peter Xu

QEMU's 32bit host support was deprecated since 10.0 and removed in 11.0, at least the system emulation part

(see commit 372ec46b9f "meson: Reject 32-bit hosts")

Quoted 8 lines

. Now it's safe to move ram_addr_t
completely over to uint64_t.

It should be almost the same as uintptr_t as before for !Xen, except that on some systems (like MacOS) uintptr_t and uint64_t can be typed slightly differently, causing unnecessary compiler warnings when use them in a mixture way.

Quoted 5 lines

Hopefully, this change also makes it clear that ram_addr_t is never used as a host pointer in any form, but only an internal QEMU integer based address space for allocating ramblocks.

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

With the commit sha no need to link to that thread IMO.

Quoted 14 lines
Cc Paolo Bonzini <[email protected]>
Cc Philippe Mathieu-Daudé <[email protected]>
---
  include/system/ram_addr.h | 6 ------
  1 file changed, 6 deletions(-)
include/system/ram_addr.h +0 -0
index 129f6b8757..06caca1506 100644

include/system/ram_addr.h line 17

Quoted 3 lines
include/system/ram_addr.h +0 -0
@@ -15,15 +15,9 @@
  #define RAM_ADDR_H
 
  /* address in the RAM (different from a physical address) */

While here we could describe a bit more:

/*
* ram_addr_t - Offset in QEMU's internal RAM address space (not a
guest physical address).
*/

include/system/ram_addr.h line 23

Quoted 12 lines
include/system/ram_addr.h +0 -6
-#if defined(CONFIG_XEN_BACKEND)
  typedef uint64_t ram_addr_t;
  #  define RAM_ADDR_MAX UINT64_MAX
  #  define RAM_ADDR_FMT "%" PRIx64
-#else
-typedef uintptr_t ram_addr_t;
-#  define RAM_ADDR_MAX UINTPTR_MAX
-#  define RAM_ADDR_FMT "%" PRIxPTR
-#endif
 
  #define DIRTY_MEMORY_VGA       0
  #define DIRTY_MEMORY_CODE      1

Thanks!

author
#2 lore raw
Quoted 5 lines · Philippe Mathieu-Daudé
Quoted 2 lines · Peter Xu

QEMU's 32bit host support was deprecated since 10.0 and removed in 11.0, at least the system emulation part

(see commit 372ec46b9f "meson: Reject 32-bit hosts")

The "at least the system emulation part" was almost for cf634dfcd8f that we re-enabled tools for 32bits.

I'll reference both commit IDs when repost.

Quoted 8 lines
Quoted 8 lines

. Now it's safe to move ram_addr_t
completely over to uint64_t.

It should be almost the same as uintptr_t as before for !Xen, except that on some systems (like MacOS) uintptr_t and uint64_t can be typed slightly differently, causing unnecessary compiler warnings when use them in a mixture way.

Quoted 7 lines
Quoted 5 lines

Hopefully, this change also makes it clear that ram_addr_t is never used as a host pointer in any form, but only an internal QEMU integer based address space for allocating ramblocks.

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

With the commit sha no need to link to that thread IMO.

I wish to keep this to provide a reference to the MacOS issue I mentioned.

Quoted 20 lines
Quoted 15 lines
Cc Paolo Bonzini <[email protected]>
Cc Philippe Mathieu-Daudé <[email protected]>
---
  include/system/ram_addr.h | 6 ------
  1 file changed, 6 deletions(-)
include/system/ram_addr.h +0 -0
index 129f6b8757..06caca1506 100644

include/system/ram_addr.h line 16

Quoted 2 lines
include/system/ram_addr.h +0 -0
@@ -15,15 +15,9 @@
  #define RAM_ADDR_H
  /* address in the RAM (different from a physical address) */

While here we could describe a bit more:

Quoted 4 lines

/*
* ram_addr_t - Offset in QEMU's internal RAM address space (not a guest
physical address).
*/

Sure!

Signature · 1 line
Peter Xu
Reviewed-by on v1 1/1
#3 lore raw
In reply to Peter Xu
Quoted 5 lines · Peter Xu

QEMU's 32bit host support was deprecated since 10.0 and removed in 11.0, at least the system emulation part. Now it's safe to move ram_addr_t completely over to uint64_t.

Quoted 8 lines

It should be almost the same as uintptr_t as before for !Xen, except that on some systems (like MacOS) uintptr_t and uint64_t can be typed slightly differently, causing unnecessary compiler warnings when use them in a mixture way.

Hopefully, this change also makes it clear that ram_addr_t is never used as a host pointer in any form, but only an internal QEMU integer based address space for allocating ramblocks.

I've thought for a while that we ought to do this even if we hadn't dropped 32-bit host support. Having ram_addr_t be 64-bit should work fine even on 32-bit hosts (as evidenced by the fact that we forced it that way when Xen was compiled in), it was just a performance thing to use 32-bit values here.

Having it be 32-bit sometimes was always an irritating source of "whoops, doesn't compile on 32-bit hosts" bugs and other oddities. There are likely various places we can clean up now where we previously were working around this (e.g. in hw/arm/vexpress.c:a15_daughterboard_init()).

There are also a few ifdefs on HOST_LONG_BITS == 32, which (where they're not relevant to the tools or guest-agent) I guess we could drop.

-- PMM

author
#4 lore raw
Quoted 27 lines · Peter Maydell
Quoted 5 lines · Peter Xu

QEMU's 32bit host support was deprecated since 10.0 and removed in 11.0, at least the system emulation part. Now it's safe to move ram_addr_t completely over to uint64_t.

Quoted 8 lines

It should be almost the same as uintptr_t as before for !Xen, except that on some systems (like MacOS) uintptr_t and uint64_t can be typed slightly differently, causing unnecessary compiler warnings when use them in a mixture way.

Hopefully, this change also makes it clear that ram_addr_t is never used as a host pointer in any form, but only an internal QEMU integer based address space for allocating ramblocks.

I've thought for a while that we ought to do this even if we hadn't dropped 32-bit host support. Having ram_addr_t be 64-bit should work fine even on 32-bit hosts (as evidenced by the fact that we forced it that way when Xen was compiled in), it was just a performance thing to use 32-bit values here.

Having it be 32-bit sometimes was always an irritating source of "whoops, doesn't compile on 32-bit hosts" bugs and other oddities. There are likely various places we can clean up now where we previously were working around this (e.g. in hw/arm/vexpress.c:a15_daughterboard_init()).

Quoted 3 lines

There are also a few ifdefs on HOST_LONG_BITS == 32, which (where they're not relevant to the tools or guest-agent) I guess we could drop.

Indeed. Let me see whether I can touch those in one go.

Quoted 2 lines

Thanks,

Signature · 1 line
Peter Xu
Reviewed-by on v1 1/1
#5 lore raw
In reply to Peter Xu
Quoted 15 lines · Peter Xu

QEMU's 32bit host support was deprecated since 10.0 and removed in 11.0, at least the system emulation part. Now it's safe to move ram_addr_t completely over to uint64_t.

It should be almost the same as uintptr_t as before for !Xen, except that on some systems (like MacOS) uintptr_t and uint64_t can be typed slightly differently, causing unnecessary compiler warnings when use them in a mixture way.

Hopefully, this change also makes it clear that ram_addr_t is never used as a host pointer in any form, but only an internal QEMU integer based address space for allocating ramblocks.

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

Quoted 7 lines
Cc Paolo Bonzini<[email protected]>
Cc Philippe Mathieu-Daudé<[email protected]>
---
  include/system/ram_addr.h | 6 ------
  1 file changed, 6 deletions(-)

r~