From: Alan Hayward <Alan.Hayward@arm.com>
To: Luis Machado <luis.machado@linaro.org>
Cc: "gdb-patches\\@sourceware.org" <gdb-patches@sourceware.org>,
"jhb@freebsd.org" <jhb@FreeBSD.org>,
"david.spickett@linaro.org" <david.spickett@linaro.org>,
Catalin Marinas <Catalin.Marinas@arm.com>, nd <nd@arm.com>
Subject: Re: [PATCH] [AArch64] MTE corefile support
Date: Fri, 21 May 2021 15:12:49 +0000 [thread overview]
Message-ID: <8214C923-1DC2-4996-B070-FF248FE1669C@arm.com> (raw)
In-Reply-To: <20210518202047.3492211-1-luis.machado@linaro.org>
> On 18 May 2021, at 21:20, Luis Machado <luis.machado@linaro.org> wrote:
>
> Teach GDB how to dump memory tags when using the gcore command and how
> to read them back from a core file generated via gcore or the kernel.
>
> Each tagged memory range (listed in /proc/<pid>/smaps) gets dumped to its
> own NT_MEMTAG note. A section named ".memtag" is created for each of those
> when reading the core file back.
>
> Dumping memory tags
> -
>
> When using the gcore command to dump a core file, GDB will go through the maps
> in /proc/<pid>/smaps looking for tagged ranges. Each of those entries gets
> passed to an arch-specific gdbarch hook that generates a vector of blobs of
> memory tag data that are blindly put into a NT_MEMTAG note.
>
> The vector is used because we may have, in the future, multiple tag types for
> a particular memory range.
>
> Each of the NT_MEMTAG notes have a generic header and a arch-specific header,
> like so:
>
> struct tag_dump_header
> {
> uint16_t format; // Only NT_MEMTAG_TYPE_AARCH_MTE at present
> uint64_t start_vma;
> uint64_t end_vma;
> };
>
> struct tag_dump_mte
> {
> uint16_t granule_byte_size;
> uint16_t tag_bit_size;
> uint16_t __unused;
> };
>
> The only bits meant to be generic are the tag_dump_format, start_vma and
> end_vma fields.
>
> The format-specific data is supposed to be opaque and only useful for the
> arch-specific code.
>
> We can extend the format in the future to make room for other memory tag
> layouts.
>
This is a wider question than this patch - but is there someplace this format will
be documented? Ideally either Linux or GDB needs something written down in
text format.
A few nits below too, but mostly looks fine.
> Reading memory tags
> -
>
> When reading a core file that contains NT_MEMTAG entries, GDB will use
> a different approach to check for tagged memory range. Rather than looking
> at /proc/<pid>/smaps, it will now look for ".memtag" sections with the right
> memory range.
>
> When reading tags, GDB will now use the core target's implementation of
> fetch_memtags (store_memtags doesn't exist for core targets). Then the data
> is fed into an arch-specific hook that will decode the memory tag format and
> return a vector of tags.
>
> I've added a test to exercise writing and reading of memory tags in core
> files.
>
> gdb/ChangeLog:
>
> YYYY-MM-DD Luis Machado <luis.machado@linaro.org>
>
> * aarch64-linux-tdep.c: Include elf/common.h.
> (MAX_TAGS_TO_TRANSFER): New constant.
> (aarch64_linux_create_memtag_notes_from_range): New function.
> (aarch64_linux_decode_memtag_note): Likewise.
> (aarch64_linux_init_abi): Register new core file hooks.
> * arch/aarch64-mte-linux.h (AARCH64_MTE_TAG_BIT_SIZE): New constant.
> (struct tag_dump_header): New struct.
> (struct tag_dump_mte): New struct.
> (MEMTAG_NOTE_HEADER_SIZE): New constant.
> * corelow.c (core_target) <supports_memory_tagging, fetch_memtags>: New
> method overrides.
> * gdbarch.c: Regenerate.
> * gdbarch.h: Likewise.
> * gdbarch.sh (create_memtag_notes_from_range): New hook.
> (decode_memtag_note): Likewise.
> * linux-tdep.c (linux_address_in_memtag_page): Renamed to...
> (linux_process_address_in_memtag_page): ... this.
> (linux_core_file_address_in_memtag_page): New function.
> (linux_address_in_memtag_page): Likewise.
> (linux_make_memtag_corefile_notes): Likewise.
> (linux_make_corefile_notes): Handle memory tag notes.
> * NEWS: Mention core file support for memory tagging.
>
> gdb/doc/ChangeLog:
>
> YYYY-MM-DD Luis Machado <luis.machado@linaro.org>
>
> * gdb.texinfo (AArch64 Memory Tagging Extension): Mention support
> for memory tagging in core files.
>
> gdb/testsuite/ChangeLog:
>
> YYYY-MM-DD Luis Machado <luis.machado@linaro.org>
>
> * gdb.arch/aarch64-mte-gcore.c: New file.
> * gdb.arch/aarch64-mte-gcore.exp: New file.
> ---
> gdb/NEWS | 4 +
> gdb/aarch64-linux-tdep.c | 179 +++++++++++++++++++
> gdb/arch/aarch64-mte-linux.h | 29 +++
> gdb/corelow.c | 88 +++++++++
> gdb/doc/gdb.texinfo | 4 +
> gdb/gdbarch.c | 64 +++++++
> gdb/gdbarch.h | 16 ++
> gdb/gdbarch.sh | 6 +
> gdb/linux-tdep.c | 141 ++++++++++++++-
> gdb/testsuite/gdb.arch/aarch64-mte-gcore.c | 93 ++++++++++
> gdb/testsuite/gdb.arch/aarch64-mte-gcore.exp | 115 ++++++++++++
> 11 files changed, 736 insertions(+), 3 deletions(-)
> create mode 100644 gdb/testsuite/gdb.arch/aarch64-mte-gcore.c
> create mode 100644 gdb/testsuite/gdb.arch/aarch64-mte-gcore.exp
>
> diff --git a/gdb/NEWS b/gdb/NEWS
> index ab678acec8b..58b9f739d4f 100644
> --- a/gdb/NEWS
> +++ b/gdb/NEWS
> @@ -3,6 +3,10 @@
>
> *** Changes since GDB 10
>
> +* GDB now supports dumping memory tag data for AArch64 MTE. It also supports
> + reading memory tag data for AArch64 MTE from core files generated by
> + the gcore command or the Linux kernel.
> +
> * GDB now supports general memory tagging functionality if the underlying
> architecture supports the proper primitives and hooks. Currently this is
> enabled only for AArch64 MTE.
> diff --git a/gdb/aarch64-linux-tdep.c b/gdb/aarch64-linux-tdep.c
> index e9761ed2189..663d0e1a215 100644
> --- a/gdb/aarch64-linux-tdep.c
> +++ b/gdb/aarch64-linux-tdep.c
> @@ -53,6 +53,8 @@
>
> #include "gdbsupport/selftest.h"
>
> +#include "elf/common.h"
> +
> /* Signal frame handling.
>
> +------------+ ^
> @@ -1779,6 +1781,172 @@ aarch64_linux_report_signal_info (struct gdbarch *gdbarch,
> }
> }
>
> +/* Maximum number of tags to request. */
> +#define MAX_TAGS_TO_TRANSFER 1024
> +
> +/* AArch64 Linux implementation of the aarch64_create_memtag_notes_from_range
> + gdbarch hook. Create core file notes for memory tags. */
> +
> +static std::vector<gdb::byte_vector>
> +aarch64_linux_create_memtag_notes_from_range (struct gdbarch *gdbarch,
> + CORE_ADDR start_address,
> + CORE_ADDR end_address)
> +{
> + /* We only handle MTE tags for now. */
> +
> + /* Figure out how many tags we need to store in this memory range. */
> + int granules = aarch64_mte_get_tag_granules (start_address,
> + end_address - start_address,
> + AARCH64_MTE_GRANULE_SIZE);
> +
> + /* Vector of memory tag notes. Add the MTE note (we only have MTE tags
> + at the moment). */
> + std::vector<gdb::byte_vector> notes (1);
> +
> + /* If there are no tag granules to fetch, just return. */
> + if (granules == 0)
> + return notes;
> +
> + /* Adjust the MTE note size to hold the header + tags. */
> + notes[0].resize (MEMTAG_NOTE_HEADER_SIZE + granules);
> +
> + CORE_ADDR address = start_address;
> + /* Vector of tags. */
> + gdb::byte_vector tags;
> +
> + while (granules > 0)
> + {
> + /* Transfer tags in chunks. */
> + gdb::byte_vector tags_read;
> + size_t xfer_len
> + = (granules >= MAX_TAGS_TO_TRANSFER)?
> + MAX_TAGS_TO_TRANSFER * AARCH64_MTE_GRANULE_SIZE :
> + granules * AARCH64_MTE_GRANULE_SIZE;
> +
> + if (!target_fetch_memtags (address, xfer_len, tags_read,
> + static_cast<int> (memtag_type::allocation)))
> + {
> + warning (_("Failed to read MTE tags from memory range [%s,%s]."),
> + phex_nz (start_address, sizeof (start_address)),
> + phex_nz (end_address, sizeof (end_address)));
> + notes.resize (0);
> + return notes;
> + }
> +
> + /* Transfer over the tags that have been read. */
> + tags.insert (tags.end (), tags_read.begin (), tags_read.end ());
> +
> + /* Adjust the remaining granules and starting address. */
> + granules -= tags_read.size ();
> + address += tags_read.size () * AARCH64_MTE_GRANULE_SIZE;
> + }
> +
> + /* Create the header. */
> + enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
> + gdb_byte *buf = notes[0].data ();
> +
> + /* Generic header. */
> + /* Tag dump format */
> + store_unsigned_integer (buf, sizeof (uint16_t), byte_order,
> + NT_MEMTAG_TYPE_AARCH_MTE);
> + buf += sizeof (uint16_t);
> +
> + /* Start address */
> + store_unsigned_integer (buf, sizeof (uint64_t), byte_order, start_address);
> + buf += sizeof (uint64_t);
> +
> + /* End address */
> + store_unsigned_integer (buf, sizeof (uint64_t), byte_order, end_address);
> + buf += sizeof (uint64_t);
> +
> + /* MTE-specific header. */
> + /* Granule byte size */
> + store_unsigned_integer (buf, sizeof (uint16_t), byte_order,
> + AARCH64_MTE_GRANULE_SIZE);
> + buf += sizeof (uint16_t);
> +
> + /* Tag bit size */
> + store_unsigned_integer (buf, sizeof (uint16_t), byte_order,
> + AARCH64_MTE_TAG_BIT_SIZE);
> + buf += sizeof (uint16_t);
> +
> + /* Unused value */
> + store_unsigned_integer (buf, sizeof (uint16_t), byte_order, 0);
> +
> + /* Store the tags. */
> + memcpy (notes[0].data () + MEMTAG_NOTE_HEADER_SIZE, tags.data (),
> + tags.size ());
> +
> + return notes;
> +}
> +
> +/* AArch64 Linux implementation of the decode_memtag_note gdbarch
> + hook. Decode a memory tag note and return the request tags. */
> +
> +static gdb::byte_vector
> +aarch64_linux_decode_memtag_note (struct gdbarch *gdbarch,
> + gdb::array_view <const gdb_byte> note,
> + CORE_ADDR address, size_t length)
> +{
> + gdb::byte_vector tags;
> +
> + /* Read the generic header. */
> + enum bfd_endian byte_order = gdbarch_byte_order (gdbarch);
> + struct tag_dump_header header;
> + const gdb_byte *buf = note.data ();
> +
> + header.format
> + = extract_unsigned_integer (buf, sizeof (uint16_t), byte_order);
> + buf += sizeof (uint16_t);
> +
> + header.start_vma
> + = extract_unsigned_integer (buf, sizeof (uint64_t), byte_order);
> + buf += sizeof (uint64_t);
> +
> + header.end_vma
> + = extract_unsigned_integer (buf, sizeof (uint64_t), byte_order);
> + buf += sizeof (uint64_t);
> +
> + /* Sanity check */
> + if (header.format != NT_MEMTAG_TYPE_AARCH_MTE)
> + {
> + warning (_("Unexpected memory tag note format (%x).\n"), header.format);
> + return tags;
> + }
> +
> + /* Calculate how many granules we need to skip to get to the granule of
> + ADDRESS. Align both the start address and the requested address
> + so it is easier to get the number of granules to skip. This way we
> + don't need to consider cases where ADDRESS falls in the middle of a
> + tag granule range. */
> + CORE_ADDR aligned_start_address
> + = align_down (header.start_vma, AARCH64_MTE_GRANULE_SIZE);
> + CORE_ADDR aligned_address = align_down (address, AARCH64_MTE_GRANULE_SIZE);
> +
> + int skipped_granules
> + = aarch64_mte_get_tag_granules (aligned_start_address,
> + aligned_address - aligned_start_address,
> + AARCH64_MTE_GRANULE_SIZE);
> +
> + /* The amount of memory tag granules we need to fetch. */
> + int granules
> + = aarch64_mte_get_tag_granules (address, length, AARCH64_MTE_GRANULE_SIZE);
> +
> + /* If there are no tag granules to decode, just return. */
> + if (granules == 0)
> + return tags;
> +
> + /* Point to the block of data that contains the first granule we are
> + interested in. */
> + const gdb_byte *tags_data = note.data () + sizeof (header) + skipped_granules;
> +
> + /* Read the tag granules. */
> + for (unsigned int i = 0; i < granules; i++)
> + tags.push_back (tags_data[i]);
> +
> + return tags;
> +}
> +
> static void
> aarch64_linux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
> {
> @@ -1862,6 +2030,17 @@ aarch64_linux_init_abi (struct gdbarch_info info, struct gdbarch *gdbarch)
>
> set_gdbarch_report_signal_info (gdbarch,
> aarch64_linux_report_signal_info);
> +
> + /* Core file helpers. */
> +
> + /* Core file helper to create memory tag notes for a particular range of
> + addresses. */
> + set_gdbarch_create_memtag_notes_from_range (gdbarch,
> + aarch64_linux_create_memtag_notes_from_range);
> +
> + /* Core file helper to decode a memory tag note. */
> + set_gdbarch_decode_memtag_note (gdbarch,
> + aarch64_linux_decode_memtag_note);
> }
>
> /* Initialize the aarch64_linux_record_tdep. */
> diff --git a/gdb/arch/aarch64-mte-linux.h b/gdb/arch/aarch64-mte-linux.h
> index 2aa97eb861a..a4a26628847 100644
> --- a/gdb/arch/aarch64-mte-linux.h
> +++ b/gdb/arch/aarch64-mte-linux.h
> @@ -32,6 +32,7 @@
>
> /* We have one tag per 16 bytes of memory. */
> #define AARCH64_MTE_GRANULE_SIZE 16
> +#define AARCH64_MTE_TAG_BIT_SIZE 4
> #define AARCH64_MTE_LOGICAL_TAG_START_BIT 56
> #define AARCH64_MTE_LOGICAL_MAX_VALUE 0xf
>
> @@ -71,4 +72,32 @@ extern CORE_ADDR aarch64_mte_set_ltag (CORE_ADDR address, CORE_ADDR tag);
> It is always possible to get the logical tag. */
> extern CORE_ADDR aarch64_mte_get_ltag (CORE_ADDR address);
>
> +/* Headers for the NT_MEMTAG notes. */
> +
> +/* Generic NT_MEMTAG header. */
> +struct tag_dump_header
> +{
> + /* Tag format. */
> + uint16_t format;
> + /* Start address of the tagged range. */
> + uint64_t start_vma;
> + /* End address of the tagged range. */
> + uint64_t end_vma;
> +};
> +
> +/* MTE-specific NT_MEMTAG header. */
> +struct tag_dump_mte
> +{
> + /* Size of the tag granule in bytes. */
> + uint16_t granule_byte_size;
> + /* Size of the tag in bits. */
> + uint16_t tag_bit_size;
> + /* Reserved field for the future. */
> + uint16_t __unused;
> +};
> +
> +/* Memory tag note header size. Includes both the generic and the
> + arch-specific parts. */
> +#define MEMTAG_NOTE_HEADER_SIZE (2 + 8 + 8 + 2 + 2 + 2)
This should compile down to a constant value: ?
#define MEMTAG_NOTE_HEADER_SIZE (sizeof(tag_dump_header)+sizeof(tag_dump_mte))
> +
> #endif /* ARCH_AARCH64_LINUX_H */
> diff --git a/gdb/corelow.c b/gdb/corelow.c
> index 452b4dd4f9a..33d16b7220a 100644
> --- a/gdb/corelow.c
> +++ b/gdb/corelow.c
> @@ -100,6 +100,13 @@ class core_target final : public process_stratum_target
>
> bool info_proc (const char *, enum info_proc_what) override;
>
> + bool supports_memory_tagging () override;
> +
> + /* Core file implementation of fetch_memtags. Fetch the memory tags from
> + core file notes. */
> + bool fetch_memtags (CORE_ADDR address, size_t len,
> + gdb::byte_vector &tags, int type) override;
> +
> /* A few helpers. */
>
> /* Getter, see variable definition. */
> @@ -1115,6 +1122,87 @@ core_target::info_proc (const char *args, enum info_proc_what request)
> return true;
> }
>
> +/* Implementation of the "supports_memory_tagging" target_ops method. */
> +
> +bool
> +core_target::supports_memory_tagging ()
> +{
> + /* Look for memory tag notes. If they exist, that means this core file
> + supports memory tagging. */
> + if (bfd_get_section_by_name (core_bfd, ".memtag") == nullptr)
> + return false;
> +
> + return true;
> +}
> +
> +/* Implementation of the "fetch_memtags" target_ops method. */
> +
> +bool
> +core_target::fetch_memtags (CORE_ADDR address, size_t len,
> + gdb::byte_vector &tags, int type)
> +{
> + struct gdbarch *gdbarch = target_gdbarch ();
> +
> + /* Make sure we have a way to decode the memory tag notes. */
> + if (!gdbarch_decode_memtag_note_p (gdbarch))
> + warning (_("gdbarch_decode_memtag_note not implemented for this "
> + "architecture."));
> +
> + asection *section
> + = bfd_get_section_by_name (core_bfd, ".memtag");
> +
> + /* Remove the top byte for the memory range check. */
> + address = address_significant (gdbarch, address);
> +
> + /* Go through all the memtag sections and figure out if ADDRESS
> + falls within one of the memory ranges that contain tags. */
> + while (section != nullptr)
> + {
> + size_t note_size = bfd_section_size (section);
> +
> + /* If the note is smaller than the size of the header, this core note
> + is malformed. */
> + if (note_size < 2 * sizeof (uint64_t) + sizeof (uint16_t))
I’m not sure why 64+16 ?
> + {
> + warning (_("malformed core note - too short for header"));
> + return false;
> + }
> +
> + gdb::byte_vector note (note_size);
> +
> + /* Fetch the contents of this particular memtag note. */
> + if (!bfd_get_section_contents (core_bfd, section,
> + note.data (), 0, note_size))
> + {
> + warning (_("could not get core note contents."));
> + return false;
> + }
> +
> + /* Read the generic header of the note. Thos contains the format,
> + start address and end address. */
> + uint64_t start_address
> + = bfd_get_64 (core_bfd, note.data () + sizeof (uint16_t));
> + uint64_t end_address
> + = bfd_get_64 (core_bfd, note.data () + sizeof (uint16_t)
> + + sizeof (uint64_t));
> +
> + /* Is the address within [start_address, end_address)? */
> + if (address >= start_address
> + && address < end_address)
> + {
> + /* Decode the memory tag note and return the tags. */
> + tags = gdbarch_decode_memtag_note (gdbarch, note, address, len);
> + return true;
> + }
> +
> + /* The requested address lies outside this particular memtag note. Keep
> + looking and get the next section. */
> + section = bfd_get_next_section_by_name (core_bfd, section);
> + }
> +
> + return false;
> +}
> +
> /* Get a pointer to the current core target. If not connected to a
> core target, return NULL. */
>
> diff --git a/gdb/doc/gdb.texinfo b/gdb/doc/gdb.texinfo
> index 56f37eb2288..738cb3b56eb 100644
> --- a/gdb/doc/gdb.texinfo
> +++ b/gdb/doc/gdb.texinfo
> @@ -25257,6 +25257,10 @@ options that can be controlled at runtime and emulates the @code{prctl}
> option @code{PR_SET_TAGGED_ADDR_CTRL}. For further information, see the
> documentation in the Linux kernel.
>
> +@value{GDBN} supports dumping memory tag data to core files through the
> +@command{gcore} command and reading memory tag data from core files generated
> +by the @command{gcore} command or the Linux kernel.
> +
> @node i386
> @subsection x86 Architecture-specific Issues
>
> diff --git a/gdb/gdbarch.c b/gdb/gdbarch.c
> index 208cf4b5aaa..de384da2e9a 100644
> --- a/gdb/gdbarch.c
> +++ b/gdb/gdbarch.c
> @@ -283,6 +283,8 @@ struct gdbarch
> gdbarch_iterate_over_regset_sections_ftype *iterate_over_regset_sections;
> gdbarch_make_corefile_notes_ftype *make_corefile_notes;
> gdbarch_find_memory_regions_ftype *find_memory_regions;
> + gdbarch_create_memtag_notes_from_range_ftype *create_memtag_notes_from_range;
> + gdbarch_decode_memtag_note_ftype *decode_memtag_note;
> gdbarch_core_xfer_shared_libraries_ftype *core_xfer_shared_libraries;
> gdbarch_core_xfer_shared_libraries_aix_ftype *core_xfer_shared_libraries_aix;
> gdbarch_core_pid_to_str_ftype *core_pid_to_str;
> @@ -667,6 +669,8 @@ verify_gdbarch (struct gdbarch *gdbarch)
> /* Skip verify of iterate_over_regset_sections, has predicate. */
> /* Skip verify of make_corefile_notes, has predicate. */
> /* Skip verify of find_memory_regions, has predicate. */
> + /* Skip verify of create_memtag_notes_from_range, has predicate. */
> + /* Skip verify of decode_memtag_note, has predicate. */
> /* Skip verify of core_xfer_shared_libraries, has predicate. */
> /* Skip verify of core_xfer_shared_libraries_aix, has predicate. */
> /* Skip verify of core_pid_to_str, has predicate. */
> @@ -925,6 +929,18 @@ gdbarch_dump (struct gdbarch *gdbarch, struct ui_file *file)
> fprintf_unfiltered (file,
> "gdbarch_dump: core_xfer_siginfo = <%s>\n",
> host_address_to_string (gdbarch->core_xfer_siginfo));
> + fprintf_unfiltered (file,
> + "gdbarch_dump: gdbarch_create_memtag_notes_from_range_p() = %d\n",
> + gdbarch_create_memtag_notes_from_range_p (gdbarch));
> + fprintf_unfiltered (file,
> + "gdbarch_dump: create_memtag_notes_from_range = <%s>\n",
> + host_address_to_string (gdbarch->create_memtag_notes_from_range));
> + fprintf_unfiltered (file,
> + "gdbarch_dump: gdbarch_decode_memtag_note_p() = %d\n",
> + gdbarch_decode_memtag_note_p (gdbarch));
> + fprintf_unfiltered (file,
> + "gdbarch_dump: decode_memtag_note = <%s>\n",
> + host_address_to_string (gdbarch->decode_memtag_note));
> fprintf_unfiltered (file,
> "gdbarch_dump: decr_pc_after_break = %s\n",
> core_addr_to_string_nz (gdbarch->decr_pc_after_break));
> @@ -3898,6 +3914,54 @@ set_gdbarch_find_memory_regions (struct gdbarch *gdbarch,
> gdbarch->find_memory_regions = find_memory_regions;
> }
>
> +bool
> +gdbarch_create_memtag_notes_from_range_p (struct gdbarch *gdbarch)
> +{
> + gdb_assert (gdbarch != NULL);
> + return gdbarch->create_memtag_notes_from_range != NULL;
> +}
> +
> +std::vector<gdb::byte_vector>
> +gdbarch_create_memtag_notes_from_range (struct gdbarch *gdbarch, CORE_ADDR start_address, CORE_ADDR end_address)
> +{
> + gdb_assert (gdbarch != NULL);
> + gdb_assert (gdbarch->create_memtag_notes_from_range != NULL);
> + if (gdbarch_debug >= 2)
> + fprintf_unfiltered (gdb_stdlog, "gdbarch_create_memtag_notes_from_range called\n");
> + return gdbarch->create_memtag_notes_from_range (gdbarch, start_address, end_address);
> +}
> +
> +void
> +set_gdbarch_create_memtag_notes_from_range (struct gdbarch *gdbarch,
> + gdbarch_create_memtag_notes_from_range_ftype create_memtag_notes_from_range)
> +{
> + gdbarch->create_memtag_notes_from_range = create_memtag_notes_from_range;
> +}
> +
> +bool
> +gdbarch_decode_memtag_note_p (struct gdbarch *gdbarch)
> +{
> + gdb_assert (gdbarch != NULL);
> + return gdbarch->decode_memtag_note != NULL;
> +}
> +
> +gdb::byte_vector
> +gdbarch_decode_memtag_note (struct gdbarch *gdbarch, gdb::array_view<const gdb_byte> note, CORE_ADDR address, size_t length)
> +{
> + gdb_assert (gdbarch != NULL);
> + gdb_assert (gdbarch->decode_memtag_note != NULL);
> + if (gdbarch_debug >= 2)
> + fprintf_unfiltered (gdb_stdlog, "gdbarch_decode_memtag_note called\n");
> + return gdbarch->decode_memtag_note (gdbarch, note, address, length);
> +}
> +
> +void
> +set_gdbarch_decode_memtag_note (struct gdbarch *gdbarch,
> + gdbarch_decode_memtag_note_ftype decode_memtag_note)
> +{
> + gdbarch->decode_memtag_note = decode_memtag_note;
> +}
> +
> bool
> gdbarch_core_xfer_shared_libraries_p (struct gdbarch *gdbarch)
> {
> diff --git a/gdb/gdbarch.h b/gdb/gdbarch.h
> index 7157e5596fd..80e244624de 100644
> --- a/gdb/gdbarch.h
> +++ b/gdb/gdbarch.h
> @@ -980,6 +980,22 @@ typedef int (gdbarch_find_memory_regions_ftype) (struct gdbarch *gdbarch, find_m
> extern int gdbarch_find_memory_regions (struct gdbarch *gdbarch, find_memory_region_ftype func, void *data);
> extern void set_gdbarch_find_memory_regions (struct gdbarch *gdbarch, gdbarch_find_memory_regions_ftype *find_memory_regions);
>
> +/* Create memory tag core file notes given a range of addresses. */
> +
> +extern bool gdbarch_create_memtag_notes_from_range_p (struct gdbarch *gdbarch);
> +
> +typedef std::vector<gdb::byte_vector> (gdbarch_create_memtag_notes_from_range_ftype) (struct gdbarch *gdbarch, CORE_ADDR start_address, CORE_ADDR end_address);
> +extern std::vector<gdb::byte_vector> gdbarch_create_memtag_notes_from_range (struct gdbarch *gdbarch, CORE_ADDR start_address, CORE_ADDR end_address);
> +extern void set_gdbarch_create_memtag_notes_from_range (struct gdbarch *gdbarch, gdbarch_create_memtag_notes_from_range_ftype *create_memtag_notes_from_range);
> +
> +/* Decode a memory tag note and return the tags that it contains. */
> +
> +extern bool gdbarch_decode_memtag_note_p (struct gdbarch *gdbarch);
> +
> +typedef gdb::byte_vector (gdbarch_decode_memtag_note_ftype) (struct gdbarch *gdbarch, gdb::array_view<const gdb_byte> note, CORE_ADDR address, size_t length);
> +extern gdb::byte_vector gdbarch_decode_memtag_note (struct gdbarch *gdbarch, gdb::array_view<const gdb_byte> note, CORE_ADDR address, size_t length);
> +extern void set_gdbarch_decode_memtag_note (struct gdbarch *gdbarch, gdbarch_decode_memtag_note_ftype *decode_memtag_note);
> +
> /* Read offset OFFSET of TARGET_OBJECT_LIBRARIES formatted shared libraries list from
> core file into buffer READBUF with length LEN. Return the number of bytes read
> (zero indicates failure).
> diff --git a/gdb/gdbarch.sh b/gdb/gdbarch.sh
> index 43e51341f97..7c5eed0780c 100755
> --- a/gdb/gdbarch.sh
> +++ b/gdb/gdbarch.sh
> @@ -745,6 +745,12 @@ M;gdb::unique_xmalloc_ptr<char>;make_corefile_notes;bfd *obfd, int *note_size;ob
> # Find core file memory regions
> M;int;find_memory_regions;find_memory_region_ftype func, void *data;func, data
>
> +# Create memory tag core file notes given a range of addresses.
> +M;std::vector<gdb::byte_vector>;create_memtag_notes_from_range;CORE_ADDR start_address, CORE_ADDR end_address;start_address, end_address
> +
> +# Decode a memory tag note and return the tags that it contains.
> +M;gdb::byte_vector;decode_memtag_note;gdb::array_view<const gdb_byte> note, CORE_ADDR address, size_t length;note, address, length
> +
> # Read offset OFFSET of TARGET_OBJECT_LIBRARIES formatted shared libraries list from
> # core file into buffer READBUF with length LEN. Return the number of bytes read
> # (zero indicates failure).
> diff --git a/gdb/linux-tdep.c b/gdb/linux-tdep.c
> index 927e69bf1e1..f4ff57dec87 100644
> --- a/gdb/linux-tdep.c
> +++ b/gdb/linux-tdep.c
> @@ -1438,10 +1438,11 @@ parse_smaps_data (const char *data,
> return smaps;
> }
>
> -/* See linux-tdep.h. */
> +/* Helper that checks if an address is in a memory tag page for a live
> + process. */
>
> -bool
> -linux_address_in_memtag_page (CORE_ADDR address)
> +static bool
> +linux_process_address_in_memtag_page (CORE_ADDR address)
> {
> if (current_inferior ()->fake_pid_p)
> return false;
> @@ -1473,6 +1474,137 @@ linux_address_in_memtag_page (CORE_ADDR address)
> return false;
> }
>
> +/* Helper that checks if an address is in a memory tag page for a core file
> + process. */
> +
> +static bool
> +linux_core_file_address_in_memtag_page (CORE_ADDR address)
> +{
> + if (core_bfd == nullptr)
> + return false;
> +
> + asection *section
> + = bfd_get_section_by_name (core_bfd, ".memtag");
> +
> + /* Go through all the memtag sections and figure out if ADDRESS
> + falls within one of the memory ranges that contain tags. */
> + while (section != nullptr)
> + {
> + size_t note_size = bfd_section_size (section);
> +
> + /* If the note is smaller than the size of the header, this core note
> + is malformed. */
> + if (note_size < 2 * sizeof (uint64_t) + sizeof (uint16_t))
> + {
> + warning (_("malformed core note - too short for header"));
> + return false;
> + }
> +
> + gdb::byte_vector note (note_size);
> +
> + /* Fetch the contents of this particular memtag note. */
> + if (!bfd_get_section_contents (core_bfd, section,
> + note.data (), 0, note_size))
> + {
> + warning (_("could not get core note contents."));
> + return false;
> + }
> +
> + /* Read the generic header of the note. Those contain the format,
> + start address and end address. */
> + uint64_t start_address
> + = bfd_get_64 (core_bfd, note.data () + sizeof (uint16_t));
> + uint64_t end_address
> + = bfd_get_64 (core_bfd, note.data () + sizeof (uint16_t)
> + + sizeof (uint64_t));
> +
> + /* Is the address within [start_address, end_address)? */
> + if (address >= start_address
> + && address < end_address)
> + return true;
> +
> + /* The requested address lies outside this particular memtag note. Keep
> + looking and get the next section. */
> + section = bfd_get_next_section_by_name (core_bfd, section);
> + }
> +
> + return false;
> +}
> +
> +/* See linux-tdep.h. */
> +
> +bool
> +linux_address_in_memtag_page (CORE_ADDR address)
> +{
> + if (!target_has_execution ())
> + return linux_core_file_address_in_memtag_page (address);
> +
> + return linux_process_address_in_memtag_page (address);
> +}
> +
> +/* For each memory map entry that has memory tagging enabled, create a new
> + core file note that contains all of its memory tags. Save the data to
> + NOTE_DATA and update NOTE_SIZE accordingly. */
> +
> +static void
> +linux_make_memtag_corefile_notes (struct gdbarch *gdbarch, bfd *obfd,
> + gdb::unique_xmalloc_ptr<char> ¬e_data,
> + int *note_size)
> +{
> + if (current_inferior ()->fake_pid_p)
> + return;
> +
> + /* If the architecture doesn't have a hook to return memory tag notes,
> + there is nothing left to do. */
> + if (!gdbarch_create_memtag_notes_from_range_p (gdbarch))
> + return;
> +
> + pid_t pid = current_inferior ()->pid;
> +
> + std::string smaps_file = string_printf ("/proc/%d/smaps", pid);
> +
> + gdb::unique_xmalloc_ptr<char> data
> + = target_fileio_read_stralloc (NULL, smaps_file.c_str ());
> +
> + if (data == nullptr)
> + return;
> +
> + std::vector<struct smaps_data> smaps;
> +
> + /* Parse the contents of smaps into a vector. */
> + smaps = parse_smaps_data (data.get (), smaps_file);
> +
> + for (const smaps_data &map : smaps)
> + {
> + /* Does this mapping have memory tagging enabled? If so, save the
> + memory tags to the core file note. */
> + if (map.vmflags.memory_tagging == 0)
> + continue;
> +
> + /* Ask the architecture to create (one or more) NT_MEMTAG notes for
> + this particular memory range, including the header.
> +
> + If the notes are too big, we may need to break up the transfer
> + into smaller chunks.
> +
> + If the architecture returns an empty vector, that means there are
> + no memory tag notes to write. */
> + std::vector<gdb::byte_vector> memory_tag_notes;
> + memory_tag_notes
> + = gdbarch_create_memtag_notes_from_range (gdbarch,
> + map.start_address,
> + map.end_address);
> + /* Write notes to the core file. */
> + for (gdb::byte_vector note : memory_tag_notes)
> + {
> + note_data.reset (elfcore_write_note (obfd, note_data.release (),
> + note_size, "CORE",
> + NT_MEMTAG, note.data (),
> + note.size ()));
> + }
> + }
> +}
> +
> /* List memory regions in the inferior for a corefile. */
>
> static int
> @@ -2051,6 +2183,9 @@ linux_make_corefile_notes (struct gdbarch *gdbarch, bfd *obfd, int *note_size)
> return NULL;
> }
>
> + /* Dump the memory tags, if any. */
> + linux_make_memtag_corefile_notes (gdbarch, obfd, note_data, note_size);
> +
> /* File mappings. */
> linux_make_mappings_corefile_notes (gdbarch, obfd, note_data, note_size);
>
> diff --git a/gdb/testsuite/gdb.arch/aarch64-mte-gcore.c b/gdb/testsuite/gdb.arch/aarch64-mte-gcore.c
> new file mode 100644
> index 00000000000..b20ebcff424
> --- /dev/null
> +++ b/gdb/testsuite/gdb.arch/aarch64-mte-gcore.c
> @@ -0,0 +1,93 @@
> +/* This test program is part of GDB, the GNU debugger.
> +
> + Copyright 2021 Free Software Foundation, Inc.
> +
> + This program is free software; you can redistribute it and/or modify
> + it under the terms of the GNU General Public License as published by
> + the Free Software Foundation; either version 3 of the License, or
> + (at your option) any later version.
> +
> + This program is distributed in the hope that it will be useful,
> + but WITHOUT ANY WARRANTY; without even the implied warranty of
> + MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> + GNU General Public License for more details.
> +
> + You should have received a copy of the GNU General Public License
> + along with this program. If not, see <http://www.gnu.org/licenses/>. */
> +
> +/* Exercise AArch64's Memory Tagging Extension with tagged pointers. */
> +
> +/* This test was based on the documentation for the AArch64 Memory Tagging
> + Extension from the Linux Kernel, found in the sources in
> + Documentation/arm64/memory-tagging-extension.rst. */
> +
> +#include <errno.h>
> +#include <stdio.h>
> +#include <stdlib.h>
> +#include <unistd.h>
> +#include <sys/auxv.h>
> +#include <sys/mman.h>
> +#include <sys/prctl.h>
> +
> +/* From arch/arm64/include/uapi/asm/hwcap.h */
> +#define HWCAP2_MTE (1 << 18)
> +
> +/* From arch/arm64/include/uapi/asm/mman.h */
> +#define PROT_MTE 0x20
> +
> +/* From include/uapi/linux/prctl.h */
> +#define PR_SET_TAGGED_ADDR_CTRL 55
> +#define PR_GET_TAGGED_ADDR_CTRL 56
> +#define PR_TAGGED_ADDR_ENABLE (1UL << 0)
> +#define PR_MTE_TCF_SHIFT 1
> +#define PR_MTE_TCF_SYNC (1UL << PR_MTE_TCF_SHIFT)
> +#define PR_MTE_TAG_SHIFT 3
> +
> +void
> +access_memory (unsigned char *tagged_ptr)
> +{
> + tagged_ptr[0] = 'a';
> +}
> +
> +int
> +main (int argc, char **argv)
> +{
> + unsigned char *tagged_ptr;
> + unsigned long page_sz = sysconf (_SC_PAGESIZE);
> + unsigned long hwcap2 = getauxval(AT_HWCAP2);
> +
> + /* Bail out if MTE is not supported. */
> + if (!(hwcap2 & HWCAP2_MTE))
> + return 1;
> +
> + /* Enable the tagged address ABI, synchronous MTE tag check faults and
> + allow all non-zero tags in the randomly generated set. */
> + if (prctl (PR_SET_TAGGED_ADDR_CTRL,
> + PR_TAGGED_ADDR_ENABLE | PR_MTE_TCF_SYNC
> + | (0xfffe << PR_MTE_TAG_SHIFT),
> + 0, 0, 0))
> + {
> + perror ("prctl () failed");
> + return 1;
> + }
> +
> + /* Create a mapping that will have PROT_MTE set. */
> + tagged_ptr = mmap (0, page_sz, PROT_READ | PROT_WRITE,
> + MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
> + if (tagged_ptr == MAP_FAILED)
> + {
> + perror ("mmap () failed");
> + return 1;
> + }
> +
> + /* Enable MTE on the above anonymous mmap. */
> + if (mprotect (tagged_ptr, page_sz, PROT_READ | PROT_WRITE | PROT_MTE))
> + {
> + perror ("mprotect () failed");
> + return 1;
> + }
> +
> + access_memory (tagged_ptr);
> +
> + return 0;
> +}
> diff --git a/gdb/testsuite/gdb.arch/aarch64-mte-gcore.exp b/gdb/testsuite/gdb.arch/aarch64-mte-gcore.exp
> new file mode 100644
> index 00000000000..bb529a8b369
> --- /dev/null
> +++ b/gdb/testsuite/gdb.arch/aarch64-mte-gcore.exp
> @@ -0,0 +1,115 @@
> +# Copyright (C) 2018-2021 Free Software Foundation, Inc.
2018?
> +#
> +# This program is free software; you can redistribute it and/or modify
> +# it under the terms of the GNU General Public License as published by
> +# the Free Software Foundation; either version 3 of the License, or
> +# (at your option) any later version.
> +#
> +# This program is distributed in the hope that it will be useful,
> +# but WITHOUT ANY WARRANTY; without even the implied warranty of
> +# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
> +# GNU General Public License for more details.
> +#
> +# You should have received a copy of the GNU General Public License
> +# along with this program. If not, see <http://www.gnu.org/licenses/>.
> +
> +# This file is part of the gdb testsuite.
> +
> +# Test generating and reading a core file with MTE memory tags.
> +
> +if {![is_aarch64_target]} {
> + verbose "Skipping ${gdb_test_file_name}."
> + return
> +}
> +
> +standard_testfile
> +if { [prepare_for_testing "failed to prepare" ${testfile} ${srcfile}] } {
> + return -1
> +}
> +
> +if ![runto_main] {
> + untested "could not run to main"
> + return -1
> +}
> +
> +# Targets that don't support memory tagging should not execute the
> +# runtime memory tagging tests.
> +if {![supports_memtag]} {
> + unsupported "memory tagging unsupported"
> + return -1
> +}
> +
> +gdb_breakpoint "access_memory"
> +
> +if [gdb_continue "access_memory"] {
> + return -1
> +}
> +
> +# Set each tag granule to a different tag value, from 0x0 to 0xf.
> +set atag_msg "Allocation tag\\(s\\) updated successfully\."
> +for {set i 15} {$i >= 0} {incr i -1} {
> + set index [expr [expr 15 - $i] * 16]
> + set tag [format "%02x" $i]
> + gdb_test "memory-tag set-allocation-tag &tagged_ptr\[$index\] 1 $tag" \
> + $atag_msg \
> + "set memory tag of &tagged_ptr\[$index\] to $tag"
> +}
> +
> +# Run until a crash and confirm GDB displays memory tag violation
> +# information.
> +gdb_test "continue" \
> + [multi_line \
> + "Program received signal SIGSEGV, Segmentation fault" \
> + "Memory tag violation while accessing address $hex" \
> + "Allocation tag $hex" \
> + "Logical tag $hex\." \
> + "$hex in access_memory \\(.*\\) at .*" \
> + ".*tagged_ptr\\\[0\\\] = 'a';"] \
> + "display tag violation information for live process"
> +
> +# Generate the core file.
> +set core_filename [standard_output_file "$testfile.core"]
> +set core_generated [gdb_gcore_cmd "$core_filename" "generate core file"]
> +
> +if { !$core_generated } {
> + return -1
> +}
> +
> +clean_restart
> +
> +# Load the program file.
> +set program_filename [standard_output_file $testfile]
> +set program_loaded [gdb_file_cmd $program_filename]
> +
> +if { $program_loaded } {
> + return -1
> +}
> +
> +# Load the core file and make sure we see the tag violation fault
> +# information.
> +gdb_test "core $core_filename" \
> + [multi_line \
> + "Core was generated by.*\." \
> + "Program terminated with signal SIGSEGV, Segmentation fault" \
> + "Memory tag violation while accessing address $hex" \
> + "Allocation tag 0xf" \
> + "Logical tag 0x0\." \
> + "#0.*$hex in access_memory \\(.*\\) at .*" \
> + ".*tagged_ptr\\\[0\\\] = 'a';"] \
> + "core file shows tag violation information"
> +
> +# Make sure we have the tag_ctl register.
> +gdb_test "info register tag_ctl" \
> + "tag_ctl.*$hex.*${::decimal}" \
> + "tag_ctl is available"
> +
> +# Check if the tag granules have the expected values. If they do, that
> +# means the core file saved the tags properly and GDB has read them
> +# correctly.
> +for {set i 15} {$i >= 0} {incr i -1} {
> + set index [expr [expr 15 - $i] * 16]
> + set tag [format "%x" $i]
> + gdb_test "memory-tag print-allocation-tag &tagged_ptr\[$index\]" \
> + "= 0x$tag" \
> + "memory tag of &tagged_ptr\[$index\] is correct"
> +}
> --
> 2.25.1
>
next prev parent reply other threads:[~2021-05-21 15:13 UTC|newest]
Thread overview: 33+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-05-18 20:20 Luis Machado
2021-05-19 10:01 ` David Spickett
2021-05-19 11:11 ` Luis Machado
2021-05-19 12:13 ` Eli Zaretskii
2021-05-21 15:12 ` Alan Hayward [this message]
2021-05-21 15:30 ` Luis Machado
2021-05-21 17:20 ` John Baldwin
2021-05-24 13:41 ` Luis Machado
2021-05-24 8:07 ` Alan Hayward
2021-05-24 12:45 ` Luis Machado
2021-05-26 14:08 ` [PATCH,v2] " Luis Machado
2021-05-29 3:14 ` Simon Marchi
2021-05-31 14:12 ` Luis Machado
2021-05-31 14:49 ` Simon Marchi
2021-05-31 14:56 ` Luis Machado
2021-05-31 14:15 ` [PATCH,v3][AArch64] " Luis Machado
2021-05-31 16:44 ` [PATCH,v4][AArch64] " Luis Machado
2021-06-01 17:45 ` [PATCH,v5][AArch64] " Luis Machado
2021-06-15 14:10 ` [Ping][PATCH,v5][AArch64] " Luis Machado
2021-06-24 14:00 ` [PATCH,v5][AArch64] " Alan Hayward
2021-06-24 14:37 ` Luis Machado
2021-06-24 15:18 ` Alan Hayward
2021-07-01 13:50 ` [PING][PATCH,v5][AArch64] " Luis Machado
2021-07-11 14:22 ` Joel Brobecker
2021-07-14 13:07 ` Catalin Marinas
2021-07-29 2:26 ` Simon Marchi
2021-07-29 16:03 ` John Baldwin
2021-07-29 18:10 ` Catalin Marinas
2021-07-29 18:20 ` Simon Marchi
2021-08-01 15:44 ` Joel Brobecker
2021-08-02 12:06 ` Luis Machado
2021-07-19 19:05 ` Luis Machado
2021-07-27 16:10 ` Luis Machado
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=8214C923-1DC2-4996-B070-FF248FE1669C@arm.com \
--to=alan.hayward@arm.com \
--cc=Catalin.Marinas@arm.com \
--cc=david.spickett@linaro.org \
--cc=gdb-patches@sourceware.org \
--cc=jhb@FreeBSD.org \
--cc=luis.machado@linaro.org \
--cc=nd@arm.com \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for read-only IMAP folder(s) and NNTP newsgroup(s).