Implements the "maintenance info amdgpu address-aliases" command by czidev-amd · Pull Request #225 · ROCm/ROCgdb

czidev-amd · 2026-07-21T12:40:54Z

Motivation

Given an expression that evaluates to an address in some address space, show for each address space supported by the current AMDGPU architecture the aliasing address in that space, if any, along with the number of consecutive bytes for which the aliasing is valid.

This is purely a diagnostic aid: it queries amd_dbgapi_convert_address_space for every address space known to the architecture and reports "" for the ones the given address does not alias into.

Test Plan

New show_address_aliases.exp test added.

czidev-amd

Thank you @lumachad for your review, I have reworked a bit this PR to include your comments. One observation is the renaming of the tests to match other "maint" tests naming pattern. Also, I have relaxed the test as I cannot really guarantee the same output across different architectures. Now, the tests is testing if gdb implementation is sane rather than gdbabi output sanity.

czidev-amd · 2026-07-29T11:00:02Z

I have spined a new version including Pedro's and Lancelot's inputs.

aktemur · 2026-07-30T10:43:00Z

+	  /* Not a real alias, NULL just maps to NULL in every
+	     space.  */
+	  if (src_is_null_address)
+	    ui_out->field_string ("valid", "<null>");


IMHO it wouldn't be wrong to print 0 here.

I can do that, let's see if anyone else has a different idea.

lancesix · 2026-08-04T10:47:40Z

+Some address spaces are private to a wave or a lane of a wave, and
+@value{GDBN} converts using whichever wave and lane are currently
+selected.  Not every address is reachable from every space: if it isn't,
+the row for that space reads @samp{<not reachable>}.


One nit, I think that we do an LDS -> global conversion, but this is only valid (ish) while the wave is stopped. As soon as we resume, this address is not valid anymore. Not sure if this should be described, or maybe this is just a dbgapi bug we should address.

It can sometimes be useful to be able to convert address from an address space to another one (if such conversion is possible). This patch proposes to add a maintenance command which shows for a given address (in any address space) the aliasing addresses in other address spaces. It also shows the number of consecutive bytes this conversion is valid for. Here are example outputs on this new command: (gdb) maintenance info amdgpu address-aliases &idx Address space Address Valid for global 0x7ff7f4600c00 0x4 generic generic#0x2000000000030 0x3d0 private_wave private_wave#0xc00 0x4 private_lane private_lane#0x30 0x3d0 local <not reachable> 0x0 (gdb) maintenance info amdgpu address-aliases gidx_p Address space Address Valid for global <not reachable> 0x0 generic generic#0x1000000000000 0x200 private_wave <not reachable> 0x0 private_lane <not reachable> 0x0 local local#0x0 0x100000000 Tested on x86_64-linux + gfx942. Change-Id: I9d118f208589c3db6a2ec966ac221b1a54d0ccb6 Co-authored-by: Claudiu Zissulescu-Ianculescu <claudiu.zissulescu-ianculescu@amd.com>

lumachad

Some automated comments. Take it or leave it. They seem mostly nits to me.

lumachad · 2026-08-21T08:14:49Z

 }

+/* Implements the "maintenance info amdgpu address-aliases" command.
+


GDB function comments should be written in third-person present tense describing what the function does, not "Implements the". Consider:

/* Show for each address space supported by the current AMDGPU architecture the aliases of a given ADDR, and the number of consecutive bytes for which each aliasing is valid. */

"Implements the" reads more like a commit message than a source comment.

lumachad · 2026-08-21T08:14:49Z

+    error (_("amd_dbgapi_address_space_get_info (NULL_ADDRESS) failed (%s)"),
+	   get_status_string (status));
+  bool src_is_null_address = (offset == src_null_address);
+


The outer parentheses around the comparison are unnecessary in C++:

bool src_is_null_address = offset == src_null_address;

lumachad · 2026-08-21T08:14:49Z

+  ui_out->table_header (max_aspace_name_len, ui_left, "aspace",
+			_("Address space"));
+  /* Address form is: <Name>0xNN..NNN.  */
+  ui_out->table_header (max_aspace_name_len + addr_width,


The comment on the next line (/* Address form is: <Name>0xNN..NNN. */) is inaccurate. paspace() emits a # separator between the address-space name and the hex address (e.g., generic#0x2000000000030), and for the default space it emits nothing at all. Suggested wording:

/* Address form is: <name>#0xNN..NNN for a named space, or just 0xNN..NNN for the default (global) space. */

lumachad · 2026-08-21T08:14:50Z

+
+  add_cmd ("address-aliases", class_maintenance,
+	   amdgpu_info_address_aliases_command, _("\
+Show ADDRESS aliases across AMDGPU address spaces.\n\


The commit message headline says "maint amdgpu info address-alias" (singular, and with amdgpu before info), but the registered command is "maintenance info amdgpu address-aliases" (plural, correct order). Please update the commit headline to match, e.g.:

gdb: Add "maint info amdgpu address-aliases" command

lumachad · 2026-08-21T08:14:50Z

@@ -0,0 +1,60 @@
+/* Copyright (C) 2026 Free Software Foundation, Inc.
+   Copyright (C) 2026 Advanced Micro Devices, Inc. All rights reserved.


The filename says maint-print-address-aliases but the command under test is maint info amdgpu address-aliases (note info, not print). Consider renaming both test files to maint-info-amdgpu-address-aliases.{cpp,exp} so they are discoverable by command name.

lumachad · 2026-08-21T08:14:50Z

+    # Count address spaces known to the architecture.
+    set aspaces 0
+    gdb_test_multiple "maint print address-spaces" "" -lbl {
+	-re "\r\n${asname}\[ \t\]+$::decimal\[ \t\]+" {


The address-space count is obtained from maint print address-spaces and then cross-checked against the row count from maint info amdgpu address-aliases. If those two commands ever diverge (e.g. one filters address spaces the other doesn't), the assertion gives a confusing failure with no obvious cause. It would be more robust to count the rows produced by the address-aliases command itself.

lumachad · 2026-08-21T08:14:50Z

+	}
+
+	-re "^\r\n${asname}\[ \t\]+${asname}#*$::hex\[ \t\]+($::hex|<null>)${trailer}" {
+	    # Match address space.


#* matches zero or more # characters, so this branch can fire even when there is no # in the address field, potentially stealing matches from the global branch above. The separator is always exactly # for named spaces, so use a literal #:

-re "^\r\n${asname}\[ \t\]+${asname}#$::hex\[ \t\]+($::hex|<null>)${trailer}" {

czidev-amd requested a review from a team as a code owner July 21, 2026 12:40

github-actions Bot assigned lumachad Jul 21, 2026

czidev-amd force-pushed the users/czissule/address-alias branch from 3d2d94b to d234a73 Compare July 21, 2026 12:45

czidev-amd changed the title ~~Implements the "maintenance info amdgpu address-aliases" comman~~ Implements the "maintenance info amdgpu address-aliases" command Jul 21, 2026

czidev-amd requested review from amd-shahab and lancesix July 21, 2026 14:05

lumachad reviewed Jul 22, 2026

View reviewed changes

lumachad assigned czidev-amd and unassigned lumachad Jul 22, 2026

czidev-amd marked this pull request as draft July 27, 2026 11:39

czidev-amd force-pushed the users/czissule/address-alias branch 2 times, most recently from dbe0775 to 1432895 Compare July 27, 2026 15:08

czidev-amd commented Jul 27, 2026

View reviewed changes

czidev-amd marked this pull request as ready for review July 27, 2026 15:14

palves reviewed Jul 27, 2026

View reviewed changes

lancesix reviewed Jul 27, 2026

View reviewed changes

Comment thread gdb/testsuite/gdb.rocm/maint-print-address-aliases.cpp Outdated

czidev-amd force-pushed the users/czissule/address-alias branch from 1432895 to 2f1c254 Compare July 28, 2026 11:36

czidev-amd mentioned this pull request Jul 29, 2026

gdb: extend maint print address-spaces to show all dbgapi available info #218

Merged

1 task

lumachad reviewed Jul 30, 2026

View reviewed changes

Comment thread gdb/amdgpu-tdep.c Outdated

Comment thread gdb/testsuite/gdb.rocm/maint-print-address-aliases.exp

Comment thread gdb/testsuite/gdb.rocm/maint-print-address-aliases.exp

aktemur reviewed Jul 30, 2026

View reviewed changes

palves reviewed Jul 30, 2026

View reviewed changes

Comment thread gdb/testsuite/gdb.rocm/maint-print-address-aliases.cpp Outdated

Comment thread gdb/testsuite/gdb.rocm/maint-print-address-aliases.exp Outdated

Comment thread gdb/amdgpu-tdep.c Outdated

Comment thread gdb/amdgpu-tdep.c Outdated

lancesix reviewed Aug 4, 2026

View reviewed changes

czidev-amd force-pushed the users/czissule/address-alias branch from 2f1c254 to 9b0c0d2 Compare August 4, 2026 13:39

aktemur reviewed Aug 13, 2026

View reviewed changes

Comment thread gdb/doc/gdb.texinfo

Comment thread gdb/testsuite/gdb.rocm/maint-print-address-aliases.exp Outdated

czidev-amd marked this pull request as draft August 17, 2026 11:10

czidev-amd force-pushed the users/czissule/address-alias branch from 9b0c0d2 to ec37ff2 Compare August 17, 2026 11:16

czidev-amd force-pushed the users/czissule/address-alias branch from ec37ff2 to 866d8c3 Compare August 17, 2026 13:14

czidev-amd marked this pull request as ready for review August 17, 2026 13:16

lumachad reviewed Aug 21, 2026

View reviewed changes

		}

		/* Implements the "maintenance info amdgpu address-aliases" command.

		@@ -0,0 +1,60 @@
		/* Copyright (C) 2026 Free Software Foundation, Inc.
		Copyright (C) 2026 Advanced Micro Devices, Inc. All rights reserved.

Uh oh!

Conversation

czidev-amd commented Jul 21, 2026

Motivation

Test Plan

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

czidev-amd left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

czidev-amd commented Jul 29, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

lumachad left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone