Skip to contentSkip to conversation

fix handling of bitfields in btf_loader

PATCH In review Andrii Nakryiko 5 participants 23 messages Feb 20, 2019 – Feb 26, 2019 BPF

lore
Share & tools
lore
#0 lore raw

cover fix handling of bitfields in btf_loader

This patchset fixes the logic of determining bitfield_offsets and initial bit_offset when using BTF type information. It eliminates all the remaining discrepancies, when doing btfdiff on vmlinux image, module two instances of incorrectly reporting struct member type name when bitfield is the very first field in a struct, which is only happening when using DWARF data.

Patch #1 makes btfdiff script easier to use during local development.

Patch #2 fixes list of known base type names to handle clang-generated type descriptions better.

Patch #3 fixes bitfield handling logic in btf_loader.

Andrii Nakryiko (3):
btfdiff: support specifying custom pahole location
pahole: complete list of base type names
btf_loader: fix bitfield fixup code

 btf_loader.c | 21 ++++++++++++++-------
 btfdiff      |  5 +++--
 dwarves.c    |  3 +++
 3 files changed, 20 insertions(+), 9 deletions(-)
Signature · 1 line
2.17.1
#1 lore raw

PATCH 1/3 btfdiff: support specifying custom pahole location

During development it's convenient to be able to specify custom location of pahole binary, built locally.

E.g.:

$ PAHOLE=~/local/pahole/build/pahole ./btfdiff /tmp/vmlinux4
---
 btfdiff | 5 +++--
 1 file changed, 3 insertions(+), 2 deletions(-)
btfdiff +3 -2
index 2b7ae33..6289a42 100755
@@ -14,9 +14,10 @@ fi
 file=$1
 btf_output=$(mktemp /tmp/btfdiff.btf.XXXXXX)
 dwarf_output=$(mktemp /tmp/btfdiff.dwarf.XXXXXX)
+pahole_bin=${PAHOLE-"pahole"}
 
-pahole -F dwarf --flat_arrays --show_private_classes $file > $dwarf_output
-pahole -F btf $file > $btf_output
+${pahole_bin} -F dwarf --flat_arrays --show_private_classes $file > $dwarf_output
+${pahole_bin} -F btf $file > $btf_output
 
 diff -up $dwarf_output $btf_output
 
Signature · 1 line
2.17.1
#2 lore raw

PATCH 2/3 pahole: complete list of base type names

In reply to Andrii Nakryiko

Clang seems to generate base type with name "short", instead of "short in", but it also isn't inconceivable to imagine other compilers generating just "long" and/or "long long". This patch adds all those short forms to a list of base type names.

$ cat type_test.c
struct s {
    short x1;
    long x2;
    long long x3;
};
int main() {
    struct s s;
    return 0;
}
$ clang -g type_test.c -o type_test && ~/local/pahole/build/pahole -JV type_test
File type_test:
[1] INT int size=4 bit_offset=0 nr_bits=32 encoding=SIGNED
[2] STRUCT s kind_flag=0 size=24 vlen=3
        x1 type_id=3 bits_offset=0
        x2 type_id=4 bits_offset=64
        x3 type_id=5 bits_offset=128
[3] INT short size=2 bit_offset=0 nr_bits=16 encoding=SIGNED
[4] INT long int size=8 bit_offset=0 nr_bits=64 encoding=SIGNED
[5] INT long long int size=8 bit_offset=0 nr_bits=64 encoding=SIGNED

Before:

$ ~/local/pahole/build/pahole -F btf type_test
base_type__name_to_size: base_type short
class__fixup_btf_bitfields: unknown base type name "short"!
struct s {
        short                      x1;                   /*     0     0 */
        /* XXX 8 bytes hole, try to pack */

        long int                   x2;                   /*     8     8 */
        long long int              x3;                   /*    16     8 */

        /* size: 24, cachelines: 1, members: 3 */
        /* sum members: 16, holes: 1, sum holes: 8 */
        /* last cacheline: 24 bytes */
};

After:

$ ~/local/pahole/build/pahole -F btf type_test
struct s {
        short                      x1;                   /*     0     2 */
        /* XXX 6 bytes hole, try to pack */

        long int                   x2;                   /*     8     8 */
        long long int              x3;                   /*    16     8 */

        /* size: 24, cachelines: 1, members: 3 */
        /* sum members: 18, holes: 1, sum holes: 6 */
        /* last cacheline: 24 bytes */
};
---
 dwarves.c | 3 +++
 1 file changed, 3 insertions(+)
dwarves.c +3 -0
index 093de11..adc411a 100644
@@ -147,11 +147,13 @@ static struct base_type_name_to_size {
 	{ .name = "signed short",	    .size = 16, },
 	{ .name = "unsigned short",	    .size = 16, },
 	{ .name = "short int",		    .size = 16, },
+	{ .name = "short",		    .size = 16, },
 	{ .name = "char",		    .size =  8, },
 	{ .name = "signed char",	    .size =  8, },
 	{ .name = "unsigned char",	    .size =  8, },
 	{ .name = "signed long",	    .size =  0, },
 	{ .name = "long int",		    .size =  0, },
+	{ .name = "long",		    .size =  0, },
 	{ .name = "signed long",	    .size =  0, },
 	{ .name = "unsigned long",	    .size =  0, },
 	{ .name = "long unsigned int",	    .size =  0, },
@@ -159,6 +161,7 @@ static struct base_type_name_to_size {
 	{ .name = "_Bool",		    .size =  8, },
 	{ .name = "long long unsigned int", .size = 64, },
 	{ .name = "long long int",	    .size = 64, },
+	{ .name = "long long",		    .size = 64, },
 	{ .name = "signed long long",	    .size = 64, },
 	{ .name = "unsigned long long",	    .size = 64, },
 	{ .name = "double",		    .size = 64, },
Signature · 1 line
2.17.1
#3 lore raw

PATCH 3/3 btf_loader: fix bitfield fixup code

In reply to Andrii Nakryiko

Existing code assumes alignment of any integer type, which breaks for packed structs. This patch fixes all the current discrepanies between dwarf and btf loader, when compared using btfdiff. It preserves bit_offset of non-bitfield members, while for bitfield ones it re-calculates initial byte/bit offset using natural alignment of the underlying integer type, which seems to be always the case for bitfields.

I've tested this on toy examples for both x86-64 and arm targets, there were no differences reported by btfdiff.

Testing on vmlinux on x86-64 shows only these discrepancies, which are unrelated to bit offsets:

$ ./btfdiff /tmp/vmlinux4
--- /tmp/btfdiff.dwarf.GIVfpr   2019-02-20 12:18:29.138788970 -0800
+++ /tmp/btfdiff.btf.c3x2KY     2019-02-20 12:18:29.351786365 -0800
@@ -16884,7 +16884,7 @@ struct pebs_record_nhm {
 };
 union hsw_tsx_tuning {
        struct {
-               unsigned int       cycles_last_block:32; /*     0: 0  4 */
+               u32                cycles_last_block:32; /*     0: 0  4 */
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */
@@ -26154,7 +26154,7 @@ struct acpi_device_power {
        /* last cacheline: 40 bytes */
 };
 struct acpi_device_perf_flags {
-       unsigned char              reserved:8;           /*     0: 0  1 */
+       u8                         reserved:8;           /*     0: 0  1 */
        /* size: 1, cachelines: 1, members: 1 */
        /* last cacheline: 1 bytes */
---
 btf_loader.c | 21 ++++++++++++++-------
 1 file changed, 14 insertions(+), 7 deletions(-)
btf_loader.c +14 -7
index fc28884..62e7e30 100644
@@ -502,19 +502,26 @@ static int class__fixup_btf_bitfields(struct tag *tag, struct cu *cu, struct btf
 		 * such unlikely thing happens.
 		 */
 		pos->byte_size = integral_bit_size / 8;
+		pos->bit_size = type_bit_size;
 
 		if (integral_bit_size == 0) {
-			pos->bit_size = integral_bit_size;
+			pos->bit_size = 0;
 			continue;
 		}
 
-		pos->bitfield_offset = pos->bit_offset % integral_bit_size;
-		if (!btfe->is_big_endian)
-			pos->bitfield_offset = integral_bit_size - pos->bitfield_offset - pos->bitfield_size;
+		if (pos->bitfield_size) {
+			/* bitfields seem to be always aligned, no matter the packing */
+			pos->byte_offset = pos->bit_offset / integral_bit_size * integral_bit_size / 8;
+		} else {
+			pos->byte_offset = pos->bit_offset / 8;
+		}
+
 
-		pos->bit_size = type_bit_size;
-		pos->byte_offset = (((pos->bit_offset / integral_bit_size) *
-				     integral_bit_size) / 8);
+		if (pos->bitfield_size) {
+			pos->bitfield_offset = pos->bit_offset - pos->byte_offset * 8;
+			if (!btfe->is_big_endian)
+				pos->bitfield_offset = integral_bit_size - pos->bitfield_offset - pos->bitfield_size;
+		}
 	}
 
 	return 0;
Signature · 1 line
2.17.1
#4 lore raw
In reply to Andrii Nakryiko

Em Wed, Feb 20, 2019 at 12:57:29PM -0800, Andrii Nakryiko escreveu:

Quoted 7 lines

This patchset fixes the logic of determining bitfield_offsets and initial bit_offset when using BTF type information. It eliminates all the remaining discrepancies, when doing btfdiff on vmlinux image, module two instances of incorrectly reporting struct member type name when bitfield is the very first field in a struct, which is only happening when using DWARF data.

Quoted 6 lines

Patch #1 makes btfdiff script easier to use during local development.

Patch #2 fixes list of known base type names to handle clang-generated type descriptions better.

Patch #3 fixes bitfield handling logic in btf_loader.

Thanks a lot! Applied.

And here I see no differences in my vmlinux:

[acme@quaco pahole]$ btfdiff vmlinux
[acme@quaco pahole]$

[acme@quaco pahole]$ perf stat pahole -F dwarf --flat_arrays --show_private_classes vmlinux > /tmp/dwarf

Performance counter stats for 'pahole -F dwarf --flat_arrays --show_private_classes vmlinux':

         36,140.58 msec task-clock:u              #    0.998 CPUs utilized          
                 0      context-switches:u        #    0.000 K/sec                  
                 0      cpu-migrations:u          #    0.000 K/sec                  
           359,238      page-faults:u             #    0.010 M/sec                  
    83,741,861,962      cycles:u                  #    2.317 GHz                    
    99,851,434,338      instructions:u            #    1.19  insn per cycle         
    20,974,798,749      branches:u                #  580.367 M/sec                  
       288,410,045      branch-misses:u           #    1.38% of all branches        
      36.230455825 seconds time elapsed

      35.037146000 seconds user
       0.906593000 seconds sys

[acme@quaco pahole]$ perf stat pahole -F btf vmlinux > /tmp/btf

Performance counter stats for 'pahole -F btf vmlinux':

            215.21 msec task-clock:u              #    0.994 CPUs utilized          
                 0      context-switches:u        #    0.000 K/sec                  
                 0      cpu-migrations:u          #    0.000 K/sec                  
             3,195      page-faults:u             #    0.015 M/sec                  
       460,363,235      cycles:u                  #    2.139 GHz                    
       546,688,575      instructions:u            #    1.19  insn per cycle         
       104,488,443      branches:u                #  485.522 M/sec                  
         2,012,815      branch-misses:u           #    1.93% of all branches        
       0.216609739 seconds time elapsed

       0.200431000 seconds user
       0.014834000 seconds sys

[acme@quaco pahole]$

[acme@quaco pahole]$ wc -l /tmp/dwarf
106101 /tmp/dwarf
[acme@quaco pahole]$ wc -l /tmp/btf
106101 /tmp/btf
[acme@quaco pahole]$

[acme@quaco pahole]$ pahole -F btf -C sk_buff_head vmlinux struct sk_buff_head {

	struct sk_buff *           next;                 /*     0     8 */
	struct sk_buff *           prev;                 /*     8     8 */
	__u32                      qlen;                 /*    16     4 */
	spinlock_t                 lock;                 /*    20     4 */
	/* size: 24, cachelines: 1, members: 4 */
	/* last cacheline: 24 bytes */
};
[acme@quaco pahole]$ pahole --expand_types -F btf -C sk_buff_head vmlinux
struct sk_buff_head {
	struct sk_buff *           next;                                                 /*     0     8 */
	struct sk_buff *           prev;                                                 /*     8     8 */
	/* typedef __u32 */ unsigned int               qlen;                             /*    16     4 */
	/* typedef spinlock_t */ struct spinlock {
		union {
			struct raw_spinlock {
				/* typedef arch_spinlock_t */ struct qspinlock {
					union {
						/* typedef atomic_t */ struct {
							int                    counter;  /*    20     4 */
						} val; /*    20     4 */
						struct {
							/* typedef u8 -> __u8 */ unsigned char          locked; /*    20     1 */
							/* typedef u8 -> __u8 */ unsigned char          pending; /*    21     1 */
						};                                       /*    20     2 */
						struct {
							/* typedef u16 -> __u16 */ short unsigned int     locked_pending; /*    20     2 */
							/* typedef u16 -> __u16 */ short unsigned int     tail; /*    22     2 */
						};                                       /*    20     4 */
					};                                               /*    20     4 */
				} raw_lock; /*    20     4 */
			} rlock; /*    20     4 */
		};                                                                       /*    20     4 */
	} lock; /*    20     4 */

	/* size: 24, cachelines: 1, members: 4 */
	/* last cacheline: 24 bytes */
};

[acme@quaco pahole]$

The problem with the packed structs is gone:

[acme@quaco pahole]$ pahole -F dwarf -C screen_info vmlinux > /tmp/dwarf [acme@quaco pahole]$ pahole -F btf -C screen_info vmlinux > /tmp/btf [acme@quaco pahole]$ diff -u /tmp/btf /tmp/dwarf
[acme@quaco pahole]$ pahole --hex -F btf -C screen_info vmlinux

struct screen_info {
	__u8                       orig_x;               /*     0   0x1 */
	__u8                       orig_y;               /*   0x1   0x1 */
	__u16                      ext_mem_k;            /*   0x2   0x2 */
	__u16                      orig_video_page;      /*   0x4   0x2 */
	__u8                       orig_video_mode;      /*   0x6   0x1 */
	__u8                       orig_video_cols;      /*   0x7   0x1 */
	__u8                       flags;                /*   0x8   0x1 */
	__u8                       unused2;              /*   0x9   0x1 */
	__u16                      orig_video_ega_bx;    /*   0xa   0x2 */
	__u16                      unused3;              /*   0xc   0x2 */
	__u8                       orig_video_lines;     /*   0xe   0x1 */
	__u8                       orig_video_isVGA;     /*   0xf   0x1 */
	__u16                      orig_video_points;    /*  0x10   0x2 */
	__u16                      lfb_width;            /*  0x12   0x2 */
	__u16                      lfb_height;           /*  0x14   0x2 */
	__u16                      lfb_depth;            /*  0x16   0x2 */
	__u32                      lfb_base;             /*  0x18   0x4 */
	__u32                      lfb_size;             /*  0x1c   0x4 */
	__u16                      cl_magic;             /*  0x20   0x2 */
	__u16                      cl_offset;            /*  0x22   0x2 */
	__u16                      lfb_linelength;       /*  0x24   0x2 */
	__u8                       red_size;             /*  0x26   0x1 */
	__u8                       red_pos;              /*  0x27   0x1 */
	__u8                       green_size;           /*  0x28   0x1 */
	__u8                       green_pos;            /*  0x29   0x1 */
	__u8                       blue_size;            /*  0x2a   0x1 */
	__u8                       blue_pos;             /*  0x2b   0x1 */
	__u8                       rsvd_size;            /*  0x2c   0x1 */
	__u8                       rsvd_pos;             /*  0x2d   0x1 */
	__u16                      vesapm_seg;           /*  0x2e   0x2 */
	__u16                      vesapm_off;           /*  0x30   0x2 */
	__u16                      pages;                /*  0x32   0x2 */
	__u16                      vesa_attributes;      /*  0x34   0x2 */
	__u32                      capabilities;         /*  0x36   0x4 */
	__u32                      ext_lfb_base;         /*  0x3a   0x4 */
	__u8                       _reserved[2];         /*  0x3e   0x2 */

	/* size: 64, cachelines: 1, members: 36 */
};

[acme@quaco pahole]$

And the definition in include/uapi/linux/screen_info.h is:

/*
* These are set up by the setup-routine at boot-time:
*/

struct screen_info {
        __u8  orig_x;           /* 0x00 */
        __u8  orig_y;           /* 0x01 */
        __u16 ext_mem_k;        /* 0x02 */
        __u16 orig_video_page;  /* 0x04 */
        __u8  orig_video_mode;  /* 0x06 */
        __u8  orig_video_cols;  /* 0x07 */
        __u8  flags;            /* 0x08 */
        __u8  unused2;          /* 0x09 */
        __u16 orig_video_ega_bx;/* 0x0a */
        __u16 unused3;          /* 0x0c */
        __u8  orig_video_lines; /* 0x0e */
        __u8  orig_video_isVGA; /* 0x0f */
        __u16 orig_video_points;/* 0x10 */

        /* VESA graphic mode -- linear frame buffer */
        __u16 lfb_width;        /* 0x12 */
        __u16 lfb_height;       /* 0x14 */
        __u16 lfb_depth;        /* 0x16 */
        __u32 lfb_base;         /* 0x18 */
        __u32 lfb_size;         /* 0x1c */
        __u16 cl_magic, cl_offset; /* 0x20 */
        __u16 lfb_linelength;   /* 0x24 */
        __u8  red_size;         /* 0x26 */
        __u8  red_pos;          /* 0x27 */
        __u8  green_size;       /* 0x28 */
        __u8  green_pos;        /* 0x29 */
        __u8  blue_size;        /* 0x2a */
        __u8  blue_pos;         /* 0x2b */
        __u8  rsvd_size;        /* 0x2c */
        __u8  rsvd_pos;         /* 0x2d */
        __u16 vesapm_seg;       /* 0x2e */
        __u16 vesapm_off;       /* 0x30 */
        __u16 pages;            /* 0x32 */
        __u16 vesa_attributes;  /* 0x34 */
        __u32 capabilities;     /* 0x36 */
        __u32 ext_lfb_base;     /* 0x3a */
        __u8  _reserved[2];     /* 0x3e */
} __attribute__((packed));

Which ones where these: "module two instances of incorrectly reporting struct member type name when bitfield is the very first field in a struct, which is only happening when using DWARF data." ?

- Arnaldo

#5 lore raw
Quoted 20 lines · Arnaldo Carvalho de Melo

Em Wed, Feb 20, 2019 at 12:57:29PM -0800, Andrii Nakryiko escreveu:

Quoted 7 lines

This patchset fixes the logic of determining bitfield_offsets and initial bit_offset when using BTF type information. It eliminates all the remaining discrepancies, when doing btfdiff on vmlinux image, module two instances of incorrectly reporting struct member type name when bitfield is the very first field in a struct, which is only happening when using DWARF data.

Quoted 6 lines

Patch #1 makes btfdiff script easier to use during local development.

Patch #2 fixes list of known base type names to handle clang-generated type descriptions better.

Patch #3 fixes bitfield handling logic in btf_loader.

Thanks a lot! Applied.

And here I see no differences in my vmlinux:

<SNIP>

Quoted 4 lines

Which ones where these: "module two instances of incorrectly reporting struct member type name when bitfield is the very first field in a struct, which is only happening when using DWARF data." ?

$ ./btfdiff /tmp/vmlinux4
--- /tmp/btfdiff.dwarf.GIVfpr   2019-02-20 12:18:29.138788970 -0800
+++ /tmp/btfdiff.btf.c3x2KY     2019-02-20 12:18:29.351786365 -0800
@@ -16884,7 +16884,7 @@ struct pebs_record_nhm {
 };
 union hsw_tsx_tuning {
        struct {
-               unsigned int       cycles_last_block:32; /*     0: 0  4 */
+               u32                cycles_last_block:32; /*     0: 0  4 */
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */
@@ -26154,7 +26154,7 @@ struct acpi_device_power {
        /* last cacheline: 40 bytes */
 };
 struct acpi_device_perf_flags {
-       unsigned char              reserved:8;           /*     0: 0  1 */
+       u8                         reserved:8;           /*     0: 0  1 */
        /* size: 1, cachelines: 1, members: 1 */
        /* last cacheline: 1 bytes */

For hsw_tsx_tuning, it is defined in sources like this:

$ cat hsw_tsx_tuning.c
typedef unsigned long long u64;
typedef unsigned int u32;

union hsw_tsx_tuning {
struct {
u32 cycles_last_block : 32,
hle_abort : 1,
rtm_abort : 1,
instruction_abort : 1,
non_instruction_abort : 1,
retry : 1,
data_conflict : 1,
capacity_writes : 1,
capacity_reads : 1;
};
u64 value;
};

int main() {
    union hsw_tsx_tuning t;
    return 0;
}

Building same defition with gcc and clang reveals some more info.

$ cc -g -c hsw_tsx_tuning.c -o hsw_tsx_tuning-gcc &&
~/local/pahole/build/pahole -J hsw_tsx_tuning-gcc
$ clang -g -c hsw_tsx_tuning.c -o hsw_tsx_tuning-clang &&
~/local/pahole/build/pahole -J hsw_tsx_tuning-clang

GCC-generated DWARF/BTF exposes this bug:

$ PAHOLE=~/local/pahole/build/pahole ~/local/pahole/btfdiff hsw_tsx_tuning-gcc
--- /tmp/btfdiff.dwarf.khRiFg   2019-02-22 15:18:46.768858141 -0800
+++ /tmp/btfdiff.btf.jqdEsR     2019-02-22 15:18:46.770858140 -0800
@@ -1,6 +1,6 @@
 union hsw_tsx_tuning {
        struct {
-               unsigned int       cycles_last_block:32; /*     0: 0  4 */
+               u32                cycles_last_block:32; /*     0: 0  4 */
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */

But the one generated by clang doesn't:

$ PAHOLE=~/local/pahole/build/pahole ~/local/pahole/btfdiff hsw_tsx_tuning-clang

The only difference is that GCC correctly marks cycles_last_block as bitfield, while clang optimizes it to stand-alone u32.

$ ~/local/pahole/build/pahole -F btf hsw_tsx_tuning-gcc
union hsw_tsx_tuning {
        struct {
                u32                cycles_last_block:32; /*     0: 0  4 */
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */
                u32                non_instruction_abort:1; /*     4:28  4 */
                u32                retry:1;            /*     4:27  4 */
                u32                data_conflict:1;    /*     4:26  4 */
                u32                capacity_writes:1;  /*     4:25  4 */
                u32                capacity_reads:1;   /*     4:24  4 */
        };                                             /*     0     8 */
        u64                        value;              /*     0     8 */
};
$ ~/local/pahole/build/pahole -F btf hsw_tsx_tuning-clang
union hsw_tsx_tuning {
        struct {
                u32                cycles_last_block;  /*     0     4 */
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */
                u32                non_instruction_abort:1; /*     4:28  4 */
                u32                retry:1;            /*     4:27  4 */
                u32                data_conflict:1;    /*     4:26  4 */
                u32                capacity_writes:1;  /*     4:25  4 */
                u32                capacity_reads:1;   /*     4:24  4 */
        };                                             /*     0     8 */
        u64                        value;              /*     0     8 */
};

I've spent a bit of time debugging this, but didn't get far, as I'm pretty unfamiliar with overall flow of decoders, I was hoping this can give you some idea where to look, though.

Quoted 2 lines

- Arnaldo

#6 lore raw

Em Fri, Feb 22, 2019 at 03:23:52PM -0800, Andrii Nakryiko escreveu:

Quoted 34 lines
Quoted 20 lines · Arnaldo Carvalho de Melo

Em Wed, Feb 20, 2019 at 12:57:29PM -0800, Andrii Nakryiko escreveu:

Quoted 7 lines

This patchset fixes the logic of determining bitfield_offsets and initial bit_offset when using BTF type information. It eliminates all the remaining discrepancies, when doing btfdiff on vmlinux image, module two instances of incorrectly reporting struct member type name when bitfield is the very first field in a struct, which is only happening when using DWARF data.

Quoted 6 lines

Patch #1 makes btfdiff script easier to use during local development.

Patch #2 fixes list of known base type names to handle clang-generated type descriptions better.

Patch #3 fixes bitfield handling logic in btf_loader.

Thanks a lot! Applied.

And here I see no differences in my vmlinux:

<SNIP>

Quoted 4 lines

Which ones where these: "module two instances of incorrectly reporting struct member type name when bitfield is the very first field in a struct, which is only happening when using DWARF data." ?

$ ./btfdiff /tmp/vmlinux4

tmp/btfdiff.btf.c3x2KY 2019-02-20 12:18:29.351786365 -0800 struct pebs_record_nhm { · line 16887

Quoted 5 lines
tmp/btfdiff.btf.c3x2KY 2019-02-20 12:18:29.351786365 -0800 renamed from tmp/btfdiff.dwarf.GIVfpr 2019-02-20 12:18:29.138788970 -0800 +1 -1
@@ -16884,7 +16884,7 @@ struct pebs_record_nhm {
 };
 union hsw_tsx_tuning {
        struct {
-               unsigned int       cycles_last_block:32; /*     0: 0  4 */
+               u32                cycles_last_block:32; /*     0: 0  4 */
$ vim hsw_tsx_tuning.c
$ gcc -g -c hsw_tsx_tuning.c -o hsw_tsx_tuning-gcc
$ pahole -J hsw_tsx_tuning
pahole: hsw_tsx_tuning: No such file or directory
$ pahole -J hsw_tsx_tuning-gcc
$
$ btfdiff hsw_tsx_tuning-gcc
--- /tmp/btfdiff.dwarf.ErXffk	2019-02-25 10:26:56.863625697 -0300
+++ /tmp/btfdiff.btf.Y5EDdM	2019-02-25 10:26:56.864625706 -0300
@@ -1,6 +1,6 @@
 union hsw_tsx_tuning {
 	struct {
-		unsigned int       cycles_last_block:32; /*     0: 0  4 */
+		u32                cycles_last_block:32; /*     0: 0  4 */
 		u32                hle_abort:1;        /*     4:31  4 */
 		u32                rtm_abort:1;        /*     4:30  4 */
 		u32                instruction_abort:1; /*     4:29  4 */
$

Reproduced.

<2><5c>: Abbrev Number: 5 (DW_TAG_member)

    <5d>   DW_AT_name        : (indirect string, offset: 0x23): cycles_last_block
    <61>   DW_AT_decl_file   : 1
    <62>   DW_AT_decl_line   : 8
    <63>   DW_AT_decl_column : 13
    <64>   DW_AT_type        : <0x40>
    <68>   DW_AT_byte_size   : 4
    <69>   DW_AT_bit_size    : 32
    <6a>   DW_AT_bit_offset  : 0
    <6b>   DW_AT_data_member_location: 0
 <1><40>: Abbrev Number: 2 (DW_TAG_typedef)
    <41>   DW_AT_name        : u32
    <45>   DW_AT_decl_file   : 1
    <46>   DW_AT_decl_line   : 4
    <47>   DW_AT_decl_column : 22
    <48>   DW_AT_type        : <0x4c>
 <1><4c>: Abbrev Number: 3 (DW_TAG_base_type)
    <4d>   DW_AT_byte_size   : 4
    <4e>   DW_AT_encoding    : 7        (unsigned)
    <4f>   DW_AT_name        : (indirect string, offset: 0x16): unsigned int

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Quoted 105 lines
tmp/btfdiff.btf.c3x2KY 2019-02-20 12:18:29.351786365 -0800 renamed from tmp/btfdiff.dwarf.GIVfpr 2019-02-20 12:18:29.138788970 -0800 +1 -1
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */
@@ -26154,7 +26154,7 @@ struct acpi_device_power {
        /* last cacheline: 40 bytes */
 };
 struct acpi_device_perf_flags {
-       unsigned char              reserved:8;           /*     0: 0  1 */
+       u8                         reserved:8;           /*     0: 0  1 */
 
        /* size: 1, cachelines: 1, members: 1 */
        /* last cacheline: 1 bytes */

For hsw_tsx_tuning, it is defined in sources like this:

$ cat hsw_tsx_tuning.c
typedef unsigned long long u64;
typedef unsigned int u32;

union hsw_tsx_tuning {
struct {
u32 cycles_last_block : 32,
hle_abort : 1,
rtm_abort : 1,
instruction_abort : 1,
non_instruction_abort : 1,
retry : 1,
data_conflict : 1,
capacity_writes : 1,
capacity_reads : 1;
};
u64 value;
};

int main() {
    union hsw_tsx_tuning t;
    return 0;
}

Building same defition with gcc and clang reveals some more info.

$ cc -g -c hsw_tsx_tuning.c -o hsw_tsx_tuning-gcc &&
~/local/pahole/build/pahole -J hsw_tsx_tuning-gcc
$ clang -g -c hsw_tsx_tuning.c -o hsw_tsx_tuning-clang &&
~/local/pahole/build/pahole -J hsw_tsx_tuning-clang

GCC-generated DWARF/BTF exposes this bug:

$ PAHOLE=~/local/pahole/build/pahole ~/local/pahole/btfdiff hsw_tsx_tuning-gcc
tmp/btfdiff.btf.jqdEsR 2019-02-22 15:18:46.770858140 -0800 renamed from tmp/btfdiff.dwarf.khRiFg 2019-02-22 15:18:46.768858141 -0800 +1 -1
@@ -1,6 +1,6 @@
 union hsw_tsx_tuning {
        struct {
-               unsigned int       cycles_last_block:32; /*     0: 0  4 */
+               u32                cycles_last_block:32; /*     0: 0  4 */
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */

But the one generated by clang doesn't:

$ PAHOLE=~/local/pahole/build/pahole ~/local/pahole/btfdiff hsw_tsx_tuning-clang

The only difference is that GCC correctly marks cycles_last_block as bitfield, while clang optimizes it to stand-alone u32.

$ ~/local/pahole/build/pahole -F btf hsw_tsx_tuning-gcc
union hsw_tsx_tuning {
        struct {
                u32                cycles_last_block:32; /*     0: 0  4 */
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */
                u32                non_instruction_abort:1; /*     4:28  4 */
                u32                retry:1;            /*     4:27  4 */
                u32                data_conflict:1;    /*     4:26  4 */
                u32                capacity_writes:1;  /*     4:25  4 */
                u32                capacity_reads:1;   /*     4:24  4 */
        };                                             /*     0     8 */
        u64                        value;              /*     0     8 */
};
$ ~/local/pahole/build/pahole -F btf hsw_tsx_tuning-clang
union hsw_tsx_tuning {
        struct {
                u32                cycles_last_block;  /*     0     4 */
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */
                u32                non_instruction_abort:1; /*     4:28  4 */
                u32                retry:1;            /*     4:27  4 */
                u32                data_conflict:1;    /*     4:26  4 */
                u32                capacity_writes:1;  /*     4:25  4 */
                u32                capacity_reads:1;   /*     4:24  4 */
        };                                             /*     0     8 */
        u64                        value;              /*     0     8 */
};

I've spent a bit of time debugging this, but didn't get far, as I'm pretty unfamiliar with overall flow of decoders, I was hoping this can give you some idea where to look, though.

Quoted 2 lines

- Arnaldo

Signature · 1 line
- Arnaldo
#7 lore raw

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

[acme@quaco pahole]$ btfdiff vmlinux
[acme@quaco pahole]$ btfdiff examples/andrii/hsw_tsx_tuning-gcc [acme@quaco pahole]$ btfdiff examples/andrii/hsw_tsx_tuning-clang [acme@quaco pahole]$ pahole examples/andrii/hsw_tsx_tuning-clang

union hsw_tsx_tuning {
	struct {
		u32                cycles_last_block;  /*     0     4 */
		u32                hle_abort:1;        /*     4:31  4 */
		u32                rtm_abort:1;        /*     4:30  4 */
		u32                instruction_abort:1; /*     4:29  4 */
		u32                non_instruction_abort:1; /*     4:28  4 */
		u32                retry:1;            /*     4:27  4 */
		u32                data_conflict:1;    /*     4:26  4 */
		u32                capacity_writes:1;  /*     4:25  4 */
		u32                capacity_reads:1;   /*     4:24  4 */
	};                                             /*     0     8 */
	u64                        value;              /*     0     8 */
};
[acme@quaco pahole]$ pahole -F btf examples/andrii/hsw_tsx_tuning-clang
union hsw_tsx_tuning {
	struct {
		u32                cycles_last_block;  /*     0     4 */
		u32                hle_abort:1;        /*     4:31  4 */
		u32                rtm_abort:1;        /*     4:30  4 */
		u32                instruction_abort:1; /*     4:29  4 */
		u32                non_instruction_abort:1; /*     4:28  4 */
		u32                retry:1;            /*     4:27  4 */
		u32                data_conflict:1;    /*     4:26  4 */
		u32                capacity_writes:1;  /*     4:25  4 */
		u32                capacity_reads:1;   /*     4:24  4 */
	};                                             /*     0     8 */
	u64                        value;              /*     0     8 */
};

[acme@quaco pahole]$

dwarf_loader.c +7 -8
index adef437109f6..37477fb23be3 100644
@@ -607,7 +607,8 @@ static struct variable *variable__new(Dwarf_Die *die, struct cu *cu)
 
 int tag__recode_dwarf_bitfield(struct tag *tag, struct cu *cu, uint16_t bit_size)
 {
-	uint16_t id;
+	int id;
+	uint16_t short_id;
 	struct tag *recoded;
 	/* in all the cases the name is at the same offset */
 	strings_t name = tag__namespace(tag)->name;
@@ -620,7 +621,7 @@ int tag__recode_dwarf_bitfield(struct tag *tag, struct cu *cu, uint16_t bit_size
 		struct tag *type = dtype->tag;
 
 		id = tag__recode_dwarf_bitfield(type, cu, bit_size);
-		if (id == tag->type)
+		if (id < 0)
 			return id;
 
 		struct type *new_typedef = obstack_zalloc(&cu->obstack,
@@ -660,10 +661,9 @@ int tag__recode_dwarf_bitfield(struct tag *tag, struct cu *cu, uint16_t bit_size
 		 * the dwarf_cu as in dwarf there are no such things
 		 * as base_types of less than 8 bits, etc.
 		 */
-		recoded = cu__find_base_type_by_sname_and_size(cu, name, bit_size, &id);
+		recoded = cu__find_base_type_by_sname_and_size(cu, name, bit_size, &short_id);
 		if (recoded != NULL)
-			return id;
-
+			return short_id;
 
 		struct base_type *new_bt = obstack_zalloc(&cu->obstack,
 							  sizeof(*new_bt));
@@ -683,10 +683,9 @@ int tag__recode_dwarf_bitfield(struct tag *tag, struct cu *cu, uint16_t bit_size
 		 * the dwarf_cu as in dwarf there are no such things
 		 * as enumeration_types of less than 8 bits, etc.
 		 */
-		recoded = cu__find_enumeration_by_sname_and_size(cu, name,
-								 bit_size, &id);
+		recoded = cu__find_enumeration_by_sname_and_size(cu, name, bit_size, &short_id);
 		if (recoded != NULL)
-			return id;
+			return short_id;
 
 		struct type *alias = tag__type(tag);
 		struct type *new_enum = obstack_zalloc(&cu->obstack, sizeof(*new_enum));
Reviewed-by
#8 lore raw
Quoted 9 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Quoted 85 lines

[acme@quaco pahole]$ btfdiff vmlinux
[acme@quaco pahole]$ btfdiff examples/andrii/hsw_tsx_tuning-gcc [acme@quaco pahole]$ btfdiff examples/andrii/hsw_tsx_tuning-clang [acme@quaco pahole]$ pahole examples/andrii/hsw_tsx_tuning-clang

union hsw_tsx_tuning {
        struct {
                u32                cycles_last_block;  /*     0     4 */
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */
                u32                non_instruction_abort:1; /*     4:28  4 */
                u32                retry:1;            /*     4:27  4 */
                u32                data_conflict:1;    /*     4:26  4 */
                u32                capacity_writes:1;  /*     4:25  4 */
                u32                capacity_reads:1;   /*     4:24  4 */
        };                                             /*     0     8 */
        u64                        value;              /*     0     8 */
};
[acme@quaco pahole]$ pahole -F btf examples/andrii/hsw_tsx_tuning-clang
union hsw_tsx_tuning {
        struct {
                u32                cycles_last_block;  /*     0     4 */
                u32                hle_abort:1;        /*     4:31  4 */
                u32                rtm_abort:1;        /*     4:30  4 */
                u32                instruction_abort:1; /*     4:29  4 */
                u32                non_instruction_abort:1; /*     4:28  4 */
                u32                retry:1;            /*     4:27  4 */
                u32                data_conflict:1;    /*     4:26  4 */
                u32                capacity_writes:1;  /*     4:25  4 */
                u32                capacity_reads:1;   /*     4:24  4 */
        };                                             /*     0     8 */
        u64                        value;              /*     0     8 */
};

[acme@quaco pahole]$

dwarf_loader.c +7 -8
index adef437109f6..37477fb23be3 100644
@@ -607,7 +607,8 @@ static struct variable *variable__new(Dwarf_Die *die, struct cu *cu)
 
 int tag__recode_dwarf_bitfield(struct tag *tag, struct cu *cu, uint16_t bit_size)
 {
-       uint16_t id;
+       int id;
+       uint16_t short_id;
        struct tag *recoded;
        /* in all the cases the name is at the same offset */
        strings_t name = tag__namespace(tag)->name;
@@ -620,7 +621,7 @@ int tag__recode_dwarf_bitfield(struct tag *tag, struct cu *cu, uint16_t bit_size
                struct tag *type = dtype->tag;
 
                id = tag__recode_dwarf_bitfield(type, cu, bit_size);
-               if (id == tag->type)
+               if (id < 0)
                        return id;
 
                struct type *new_typedef = obstack_zalloc(&cu->obstack,
@@ -660,10 +661,9 @@ int tag__recode_dwarf_bitfield(struct tag *tag, struct cu *cu, uint16_t bit_size
                 * the dwarf_cu as in dwarf there are no such things
                 * as base_types of less than 8 bits, etc.
                 */
-               recoded = cu__find_base_type_by_sname_and_size(cu, name, bit_size, &id);
+               recoded = cu__find_base_type_by_sname_and_size(cu, name, bit_size, &short_id);
                if (recoded != NULL)
-                       return id;
-
+                       return short_id;
 
                struct base_type *new_bt = obstack_zalloc(&cu->obstack,
                                                          sizeof(*new_bt));
@@ -683,10 +683,9 @@ int tag__recode_dwarf_bitfield(struct tag *tag, struct cu *cu, uint16_t bit_size
                 * the dwarf_cu as in dwarf there are no such things
                 * as enumeration_types of less than 8 bits, etc.
                 */
-               recoded = cu__find_enumeration_by_sname_and_size(cu, name,
-                                                                bit_size, &id);
+               recoded = cu__find_enumeration_by_sname_and_size(cu, name, bit_size, &short_id);
                if (recoded != NULL)
-                       return id;
+                       return short_id;
 
                struct type *alias = tag__type(tag);
                struct type *new_enum = obstack_zalloc(&cu->obstack, sizeof(*new_enum));
#9 lore raw

encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>
[386033] STRUCT dl_scope_free_list kind_flag=0 size=408 vlen=2

	count type_id=385556 bits_offset=0
	list type_id=386034 bits_offset=64
[386034] ARRAY (anon) type_id=385511 index_type_id=13 nr_elems=50
[386035] STRUCT rtld_global kind_flag=0 size=3992 vlen=31
	_dl_ns type_id=386036 bits_offset=0
	_dl_nns type_id=385556 bits_offset=18432
	_dl_load_lock type_id=385917 bits_offset=18496
	_dl_load_write_lock type_id=385917 bits_offset=18816
	_dl_load_adds type_id=385567 bits_offset=19136
	_dl_initfirst type_id=385922 bits_offset=19200
	_dl_cpuclock_offset type_id=385662 bits_offset=19264
	_dl_profile_map type_id=385922 bits_offset=19328
	_dl_num_relocations type_id=385496 bits_offset=19392
	_dl_num_cache_relocations type_id=385496 bits_offset=19456
	_dl_all_dirs type_id=385926 bits_offset=19520
	_dl_rtld_map type_id=385919 bits_offset=19584
	audit_data type_id=386037 bits_offset=28736
	_dl_rtld_lock_recursive type_id=385680 bits_offset=30784
	_dl_rtld_unlock_recursive type_id=385680 bits_offset=30848
	_dl_x86_feature_1 type_id=385571 bits_offset=30912
	_dl_x86_legacy_bitmap type_id=385653 bits_offset=30976
	_dl_make_stack_executable_hook type_id=386039 bits_offset=31104
	_dl_stack_flags type_id=385525 bits_offset=31168
	_dl_tls_dtv_gaps type_id=385825 bits_offset=31200
	_dl_tls_max_dtv_idx type_id=385556 bits_offset=31232
	_dl_tls_dtv_slotinfo_list type_id=386031 bits_offset=31296
	_dl_tls_static_nelem type_id=385556 bits_offset=31360
	_dl_tls_static_size type_id=385556 bits_offset=31424
	_dl_tls_static_used type_id=385556 bits_offset=31488
	_dl_tls_static_align type_id=385556 bits_offset=31552
	_dl_initial_dtv type_id=385511 bits_offset=31616
	_dl_tls_generation type_id=385556 bits_offset=31680
	_dl_init_static_tls type_id=386041 bits_offset=31744
	_dl_wait_lookup_done type_id=385665 bits_offset=31808
	_dl_scope_free_list type_id=386042 bits_offset=31872

[386036] ARRAY (anon) type_id=386028 index_type_id=13 nr_elems=16 [386037] ARRAY (anon) type_id=385938 index_type_id=13 nr_elems=16 [386038] FUNC_PROTO (anon) return=385501 args=(385970 (anon)) [386039] PTR (anon) type_id=386038
[386040] FUNC_PROTO (anon) return=0 args=(385922 (anon)) [386041] PTR (anon) type_id=386040
[386042] PTR (anon) type_id=386033
[386043] STRUCT rtld_global_ro kind_flag=0 size=432 vlen=40

	_dl_debug_mask type_id=385501 bits_offset=0
	_dl_osversion type_id=385495 bits_offset=32
	_dl_platform type_id=385595 bits_offset=64
	_dl_platformlen type_id=385556 bits_offset=128
	_dl_pagesize type_id=385556 bits_offset=192
	_dl_inhibit_cache type_id=385501 bits_offset=256
	_dl_initial_searchlist type_id=385918 bits_offset=320
	_dl_clktck type_id=385501 bits_offset=448
	_dl_verbose type_id=385501 bits_offset=480
	_dl_debug_fd type_id=385501 bits_offset=512
	_dl_lazy type_id=385501 bits_offset=544
	_dl_bind_not type_id=385501 bits_offset=576
	_dl_dynamic_weak type_id=385501 bits_offset=608
	_dl_fpu_control type_id=385977 bits_offset=640
	_dl_correct_cache_id type_id=385501 bits_offset=672
	_dl_hwcap type_id=385520 bits_offset=704
	_dl_auxv type_id=386045 bits_offset=768
	_dl_x86_cpu_features type_id=385599 bits_offset=832
	_dl_x86_hwcap_flags type_id=386047 bits_offset=1664
	_dl_x86_platforms type_id=386049 bits_offset=1880
	_dl_inhibit_rpath type_id=385595 bits_offset=2176
	_dl_origin_path type_id=385595 bits_offset=2240
	_dl_use_load_bias type_id=385529 bits_offset=2304
	_dl_profile type_id=385595 bits_offset=2368
	_dl_profile_output type_id=385595 bits_offset=2432
	_dl_trace_prelink type_id=385595 bits_offset=2496
	_dl_trace_prelink_map type_id=385922 bits_offset=2560
	_dl_init_all_dirs type_id=385926 bits_offset=2624
	_dl_sysinfo_dso type_id=386050 bits_offset=2688
	_dl_sysinfo_map type_id=385922 bits_offset=2752
	_dl_hwcap2 type_id=385520 bits_offset=2816
	_dl_debug_printf type_id=386052 bits_offset=2880
	_dl_mcount type_id=386054 bits_offset=2944
	_dl_lookup_symbol_x type_id=386058 bits_offset=3008
	_dl_open type_id=386060 bits_offset=3072
	_dl_close type_id=385680 bits_offset=3136
	_dl_tls_get_addr_soft type_id=386062 bits_offset=3200
	_dl_discover_osversion type_id=386064 bits_offset=3264
	_dl_audit type_id=386023 bits_offset=3328
	_dl_naudit type_id=385495 bits_offset=3392

[386044] CONST (anon) type_id=386043
[386045] PTR (anon) type_id=385552
[386046] ARRAY (anon) type_id=385515 index_type_id=13 nr_elems=27
[386047] CONST (anon) type_id=386046
[386048] ARRAY (anon) type_id=385515 index_type_id=13 nr_elems=36
[386049] CONST (anon) type_id=386048
[386050] PTR (anon) type_id=385538
[386051] FUNC_PROTO (anon) return=0 args=(385595 (anon), vararg)
[386052] PTR (anon) type_id=386051
[386053] FUNC_PROTO (anon) return=0 args=(385529 (anon), 385529 (anon))
[386054] PTR (anon) type_id=386053
[386055] FUNC_PROTO (anon) return=385978 args=(385595 (anon), 385922 (anon), 386056 (anon), 385950 (anon), 386057 (anon), 385501 (anon), 385501 (anon), 385922 (anon)) [386056] PTR (anon) type_id=385937
[386057] PTR (anon) type_id=385943
[386058] PTR (anon) type_id=386055
[386059] FUNC_PROTO (anon) return=385511 args=(385595 (anon), 385501 (anon), 385640 (anon), 385602 (anon), 385501 (anon), 385954 (anon), 385954 (anon)) [386060] PTR (anon) type_id=386059
[386061] FUNC_PROTO (anon) return=385511 args=(385922 (anon))
[386062] PTR (anon) type_id=386061
[386063] FUNC_PROTO (anon) return=385501 args=(void)
[386064] PTR (anon) type_id=386063
[386065] ARRAY (anon) type_id=386069 index_type_id=13 nr_elems=28
[386066] CONST (anon) type_id=386065
[386067] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none)
[386068] INT char size=1 bit_offset=0 nr_bits=8 encoding=(none)
[386069] CONST (anon) type_id=386068
[386070] TYPEDEF ui32 type_id=386072
[386071] CONST (anon) type_id=386070
[386072] INT unsigned int size=4 bit_offset=0 nr_bits=32 encoding=(none)
[386073] ARRAY (anon) type_id=386071 index_type_id=13 nr_elems=1
[386074] CONST (anon) type_id=386073
[386075] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none)
Cannot open libc-2.28.so.debug
Failed to encode BTF
[acme@quaco pahole]$

#10 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

Quoted 3 lines

[386075] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none) Cannot open libc-2.28.so.debug
Failed to encode BTF

I've tried the same on libc-2.26.so.debug and it worked.

It failes after BTF encoding and deduping is done, exactly when it tries to re-open libc-2.28.so.debug to write out new contents. Not sure why it would fail, as it seems to be just plain open() call, but the fact that we both can repro this means there is something there.

It also seems that libc-2.28.so.debug is corrupted:

$ readelf -n /home/andriin/local/btf/libc-2.26.so.debug | grep
'Unknown note type' | wc -l
0
$ readelf -n /home/andriin/local/btf/libc-2.28.so.debug | grep
'Unknown note type' | wc -l
17953
Quoted 1 line

[acme@quaco pahole]$

#11 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

Quoted 39 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

Quoted 3 lines

[386075] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none) Cannot open libc-2.28.so.debug
Failed to encode BTF

I've tried the same on libc-2.26.so.debug and it worked.

It failes after BTF encoding and deduping is done, exactly when it tries to re-open libc-2.28.so.debug to write out new contents. Not sure why it would fail, as it seems to be just plain open() call, but the fact that we both can repro this means there is something there.

Quoted 9 lines

It also seems that libc-2.28.so.debug is corrupted:

$ readelf -n /home/andriin/local/btf/libc-2.26.so.debug | grep
'Unknown note type' | wc -l
0
$ readelf -n /home/andriin/local/btf/libc-2.28.so.debug | grep
'Unknown note type' | wc -l
17953

Ignore everything I've said (though 'Unknown note type' issue is strange), for me it was just permissions issues.

$ chmod +rw /home/andriin/local/btf/libc-2.28.so.debug
$ ~/local/pahole/build/pahole -J ~/local/btf/libc-2.28.so.debug
$ PAHOLE=~/local/pahole/build/pahole ./btfdiff ~/local/btf/libc-2.28.so.debug
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
--- /tmp/btfdiff.dwarf.aV4wSL   2019-02-25 13:31:54.787923673 -0800
+++ /tmp/btfdiff.btf.22NQmJ     2019-02-25 13:31:54.802923668 -0800
@@ -461,9 +461,15 @@ struct La_x86_64_retval {
        uint64_t                   lrv_rdx;              /*     8     8 */
        La_x86_64_xmm              lrv_xmm0;             /*    16    16 */
        La_x86_64_xmm              lrv_xmm1;             /*    32    16 */
-       long double                lrv_st0;              /*    48    16 */
+       long double                lrv_st0;              /*    48     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
        /* --- cacheline 1 boundary (64 bytes) --- */
-       long double                lrv_st1;              /*    64    16 */
+       long double                lrv_st1;              /*    64     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
        La_x86_64_vector           lrv_vector0;          /*    80    64 */
        /* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
        La_x86_64_vector           lrv_vector1;          /*   144    64 */
@@ -472,6 +478,7 @@ struct La_x86_64_retval {
        __int128                   lrv_bnd1;             /*   224    16 */
        /* size: 240, cachelines: 4, members: 10 */
+       /* sum members: 224, holes: 2, sum holes: 16 */
        /* last cacheline: 48 bytes */
 };
 struct r_debug {
@@ -2044,7 +2051,7 @@ union ieee754_float {
        } ieee_nan;                                    /*     0     4 */
 };
 union ieee854_long_double {
-       long double                d;                  /*     0    16 */
+       long double                d;                  /*     0     8 */
        struct {
                unsigned int       mantissa1:32;       /*     0: 0  4 */
                unsigned int       mantissa0:32;       /*     4: 0  4 */
@@ -2141,7 +2148,7 @@ struct ucontext_t {
        /* last cacheline: 8 bytes */
 };
 union ieee854_float128 {
-       _Float128                  d;                  /*     0    16 */
+       _Float128                  d;                  /*     0     0 */

struct {
unsigned int mantissa3:32; /* 0: 0 4 */
unsigned int mantissa2:32; /* 4: 0 4 */
@@ -2219,7 +2226,7 @@ union printf_arg {

        long unsigned int          pa_u_long_int;      /*     0     8 */
        long long unsigned int     pa_u_long_long_int; /*     0     8 */
        double                     pa_double;          /*     0     8 */
-       long double                pa_long_double;     /*     0    16 */
+       long double                pa_long_double;     /*     0     8 */
        const char  *              pa_string;          /*     0     8 */
        const wchar_t  *           pa_wstring;         /*     0     8 */
        void *                     pa_pointer;         /*     0     8 */

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

#12 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

One more datapoint. I've compiled latest redis sources and tested pahole/BTF on it:

$ ~/local/pahole/build/pahole -J ~/local/btf/redis-server.debug
$ PAHOLE=~/local/pahole/build/pahole ./btfdiff ~/local/btf/redis-server.debug
$ readelf -S ~/local/btf/redis-server.debug | rg -A1 'BTF|debug_str|debug_info'
readelf: /home/andriin/local/btf/redis-server.debug: Warning: the
.dynamic section is not contained within the dynamic segment
  [30] .debug_info       PROGBITS         0000000000000000  0019738d
       00000000002529cf  0000000000000000           0     0     1
--
  [33] .debug_str        PROGBITS         0000000000000000  004574a0
       000000000004fba1  0000000000000001  MS       0     0     1
--
  [40] .BTF              PROGBITS         0000000000000000  007dcc92
       000000000001553c  0000000000000000           0     0     1
$ printf "dwarf data: %d, dwarf strs: %d, btf: %d (ratio: %.3lf)\n"
0x2529cf 0x4fba1 0x1553c '(0x2529cf+0x4fba1+1.0)/0x1553c'
dwarf data: 2435535, dwarf strs: 326561, btf: 87356 (ratio: 31.619)

Even for relatively small DWARF in redis (2.6MB, including strings), BTF deduplication gives 31.5x compression.

-- Andrii

Quoted 135 lines · Andrii Nakryiko
Quoted 39 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

Quoted 3 lines

[386075] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none) Cannot open libc-2.28.so.debug
Failed to encode BTF

I've tried the same on libc-2.26.so.debug and it worked.

It failes after BTF encoding and deduping is done, exactly when it tries to re-open libc-2.28.so.debug to write out new contents. Not sure why it would fail, as it seems to be just plain open() call, but the fact that we both can repro this means there is something there.

Quoted 9 lines

It also seems that libc-2.28.so.debug is corrupted:

$ readelf -n /home/andriin/local/btf/libc-2.26.so.debug | grep
'Unknown note type' | wc -l
0
$ readelf -n /home/andriin/local/btf/libc-2.28.so.debug | grep
'Unknown note type' | wc -l
17953

Ignore everything I've said (though 'Unknown note type' issue is strange), for me it was just permissions issues.

$ chmod +rw /home/andriin/local/btf/libc-2.28.so.debug
$ ~/local/pahole/build/pahole -J ~/local/btf/libc-2.28.so.debug
$ PAHOLE=~/local/pahole/build/pahole ./btfdiff ~/local/btf/libc-2.28.so.debug
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
tmp/btfdiff.btf.22NQmJ 2019-02-25 13:31:54.802923668 -0800 renamed from tmp/btfdiff.dwarf.aV4wSL 2019-02-25 13:31:54.787923673 -0800 +12 -5
@@ -461,9 +461,15 @@ struct La_x86_64_retval {
        uint64_t                   lrv_rdx;              /*     8     8 */
        La_x86_64_xmm              lrv_xmm0;             /*    16    16 */
        La_x86_64_xmm              lrv_xmm1;             /*    32    16 */
-       long double                lrv_st0;              /*    48    16 */
+       long double                lrv_st0;              /*    48     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
        /* --- cacheline 1 boundary (64 bytes) --- */
-       long double                lrv_st1;              /*    64    16 */
+       long double                lrv_st1;              /*    64     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
        La_x86_64_vector           lrv_vector0;          /*    80    64 */
        /* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
        La_x86_64_vector           lrv_vector1;          /*   144    64 */
@@ -472,6 +478,7 @@ struct La_x86_64_retval {
        __int128                   lrv_bnd1;             /*   224    16 */
 
        /* size: 240, cachelines: 4, members: 10 */
+       /* sum members: 224, holes: 2, sum holes: 16 */
        /* last cacheline: 48 bytes */
 };
 struct r_debug {
@@ -2044,7 +2051,7 @@ union ieee754_float {
        } ieee_nan;                                    /*     0     4 */
 };
 union ieee854_long_double {
-       long double                d;                  /*     0    16 */
+       long double                d;                  /*     0     8 */
        struct {
                unsigned int       mantissa1:32;       /*     0: 0  4 */
                unsigned int       mantissa0:32;       /*     4: 0  4 */
@@ -2141,7 +2148,7 @@ struct ucontext_t {
        /* last cacheline: 8 bytes */
 };
 union ieee854_float128 {
-       _Float128                  d;                  /*     0    16 */
+       _Float128                  d;                  /*     0     0 */
        struct {
                unsigned int       mantissa3:32;       /*     0: 0  4 */
                unsigned int       mantissa2:32;       /*     4: 0  4 */
@@ -2219,7 +2226,7 @@ union printf_arg {
        long unsigned int          pa_u_long_int;      /*     0     8 */
        long long unsigned int     pa_u_long_long_int; /*     0     8 */
        double                     pa_double;          /*     0     8 */
-       long double                pa_long_double;     /*     0    16 */
+       long double                pa_long_double;     /*     0     8 */
        const char  *              pa_string;          /*     0     8 */
        const wchar_t  *           pa_wstring;         /*     0     8 */
        void *                     pa_pointer;         /*     0     8 */

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

#13 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

In reply to Andrii Nakryiko
Quoted 127 lines · Andrii Nakryiko
Quoted 39 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

Quoted 3 lines

[386075] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none) Cannot open libc-2.28.so.debug
Failed to encode BTF

I've tried the same on libc-2.26.so.debug and it worked.

It failes after BTF encoding and deduping is done, exactly when it tries to re-open libc-2.28.so.debug to write out new contents. Not sure why it would fail, as it seems to be just plain open() call, but the fact that we both can repro this means there is something there.

Quoted 9 lines

It also seems that libc-2.28.so.debug is corrupted:

$ readelf -n /home/andriin/local/btf/libc-2.26.so.debug | grep
'Unknown note type' | wc -l
0
$ readelf -n /home/andriin/local/btf/libc-2.28.so.debug | grep
'Unknown note type' | wc -l
17953

Ignore everything I've said (though 'Unknown note type' issue is strange), for me it was just permissions issues.

$ chmod +rw /home/andriin/local/btf/libc-2.28.so.debug
$ ~/local/pahole/build/pahole -J ~/local/btf/libc-2.28.so.debug
$ PAHOLE=~/local/pahole/build/pahole ./btfdiff ~/local/btf/libc-2.28.so.debug
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
tmp/btfdiff.btf.22NQmJ 2019-02-25 13:31:54.802923668 -0800 renamed from tmp/btfdiff.dwarf.aV4wSL 2019-02-25 13:31:54.787923673 -0800 +12 -5
@@ -461,9 +461,15 @@ struct La_x86_64_retval {
         uint64_t                   lrv_rdx;              /*     8     8 */
         La_x86_64_xmm              lrv_xmm0;             /*    16    16 */
         La_x86_64_xmm              lrv_xmm1;             /*    32    16 */
-       long double                lrv_st0;              /*    48    16 */
+       long double                lrv_st0;              /*    48     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
         /* --- cacheline 1 boundary (64 bytes) --- */
-       long double                lrv_st1;              /*    64    16 */
+       long double                lrv_st1;              /*    64     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
         La_x86_64_vector           lrv_vector0;          /*    80    64 */
         /* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
         La_x86_64_vector           lrv_vector1;          /*   144    64 */
@@ -472,6 +478,7 @@ struct La_x86_64_retval {
         __int128                   lrv_bnd1;             /*   224    16 */
 
         /* size: 240, cachelines: 4, members: 10 */
+       /* sum members: 224, holes: 2, sum holes: 16 */
         /* last cacheline: 48 bytes */
  };
  struct r_debug {
@@ -2044,7 +2051,7 @@ union ieee754_float {
         } ieee_nan;                                    /*     0     4 */
  };
  union ieee854_long_double {
-       long double                d;                  /*     0    16 */
+       long double                d;                  /*     0     8 */
         struct {
                 unsigned int       mantissa1:32;       /*     0: 0  4 */
                 unsigned int       mantissa0:32;       /*     4: 0  4 */
@@ -2141,7 +2148,7 @@ struct ucontext_t {
         /* last cacheline: 8 bytes */
  };
  union ieee854_float128 {
-       _Float128                  d;                  /*     0    16 */
+       _Float128                  d;                  /*     0     0 */
         struct {
                 unsigned int       mantissa3:32;       /*     0: 0  4 */
                 unsigned int       mantissa2:32;       /*     4: 0  4 */
@@ -2219,7 +2226,7 @@ union printf_arg {
         long unsigned int          pa_u_long_int;      /*     0     8 */
         long long unsigned int     pa_u_long_long_int; /*     0     8 */
         double                     pa_double;          /*     0     8 */
-       long double                pa_long_double;     /*     0    16 */
+       long double                pa_long_double;     /*     0     8 */
         const char  *              pa_string;          /*     0     8 */
         const wchar_t  *           pa_wstring;         /*     0     8 */
         void *                     pa_pointer;         /*     0     8 */
Quoted 2 lines

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

BPF does not support floating point instructions. As the result, BTF design does not contain floating point types.

Looks like kernel also uses float types (in certain drivers and arch's), but types referred by bpf program mostly will not contain float types.

Currently, both llvm and pahole will ignore floating types, so from type comparison point of view, we will be fine. The issue may come up with people using vmlinux BTF inside the kernel for different purpose where a float type needs to be specified, e.g., for a floating point function argument in ftrace. But this is rare too.

I am not 100% sure whether pahole dwarf/btf parity is considered as a strong reason to add floating point type support in BTF or not.

Quoted 5 lines

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

#14 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

Quoted 2 lines · Yonghong Song

I am not 100% sure whether pahole dwarf/btf parity is considered as a strong reason to add floating point type support in BTF or not.

I wouldn't add floating point to BTF yet. BTF is not trying to replace dwarf.
Let's make it working well for BPF use cases first.

#15 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

In reply to Yonghong Song
Quoted 147 lines · Yonghong Song
Quoted 127 lines · Andrii Nakryiko
Quoted 39 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

Quoted 3 lines

[386075] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none) Cannot open libc-2.28.so.debug
Failed to encode BTF

I've tried the same on libc-2.26.so.debug and it worked.

It failes after BTF encoding and deduping is done, exactly when it tries to re-open libc-2.28.so.debug to write out new contents. Not sure why it would fail, as it seems to be just plain open() call, but the fact that we both can repro this means there is something there.

Quoted 9 lines

It also seems that libc-2.28.so.debug is corrupted:

$ readelf -n /home/andriin/local/btf/libc-2.26.so.debug | grep
'Unknown note type' | wc -l
0
$ readelf -n /home/andriin/local/btf/libc-2.28.so.debug | grep
'Unknown note type' | wc -l
17953

Ignore everything I've said (though 'Unknown note type' issue is strange), for me it was just permissions issues.

$ chmod +rw /home/andriin/local/btf/libc-2.28.so.debug
$ ~/local/pahole/build/pahole -J ~/local/btf/libc-2.28.so.debug
$ PAHOLE=~/local/pahole/build/pahole ./btfdiff ~/local/btf/libc-2.28.so.debug
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
tmp/btfdiff.btf.22NQmJ 2019-02-25 13:31:54.802923668 -0800 renamed from tmp/btfdiff.dwarf.aV4wSL 2019-02-25 13:31:54.787923673 -0800 +12 -5
@@ -461,9 +461,15 @@ struct La_x86_64_retval {
         uint64_t                   lrv_rdx;              /*     8     8 */
         La_x86_64_xmm              lrv_xmm0;             /*    16    16 */
         La_x86_64_xmm              lrv_xmm1;             /*    32    16 */
-       long double                lrv_st0;              /*    48    16 */
+       long double                lrv_st0;              /*    48     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
         /* --- cacheline 1 boundary (64 bytes) --- */
-       long double                lrv_st1;              /*    64    16 */
+       long double                lrv_st1;              /*    64     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
         La_x86_64_vector           lrv_vector0;          /*    80    64 */
         /* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
         La_x86_64_vector           lrv_vector1;          /*   144    64 */
@@ -472,6 +478,7 @@ struct La_x86_64_retval {
         __int128                   lrv_bnd1;             /*   224    16 */
 
         /* size: 240, cachelines: 4, members: 10 */
+       /* sum members: 224, holes: 2, sum holes: 16 */
         /* last cacheline: 48 bytes */
  };
  struct r_debug {
@@ -2044,7 +2051,7 @@ union ieee754_float {
         } ieee_nan;                                    /*     0     4 */
  };
  union ieee854_long_double {
-       long double                d;                  /*     0    16 */
+       long double                d;                  /*     0     8 */
         struct {
                 unsigned int       mantissa1:32;       /*     0: 0  4 */
                 unsigned int       mantissa0:32;       /*     4: 0  4 */
@@ -2141,7 +2148,7 @@ struct ucontext_t {
         /* last cacheline: 8 bytes */
  };
  union ieee854_float128 {
-       _Float128                  d;                  /*     0    16 */
+       _Float128                  d;                  /*     0     0 */
         struct {
                 unsigned int       mantissa3:32;       /*     0: 0  4 */
                 unsigned int       mantissa2:32;       /*     4: 0  4 */
@@ -2219,7 +2226,7 @@ union printf_arg {
         long unsigned int          pa_u_long_int;      /*     0     8 */
         long long unsigned int     pa_u_long_long_int; /*     0     8 */
         double                     pa_double;          /*     0     8 */
-       long double                pa_long_double;     /*     0    16 */
+       long double                pa_long_double;     /*     0     8 */
         const char  *              pa_string;          /*     0     8 */
         const wchar_t  *           pa_wstring;         /*     0     8 */
         void *                     pa_pointer;         /*     0     8 */
Quoted 2 lines

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

BPF does not support floating point instructions. As the result, BTF design does not contain floating point types.

Looks like kernel also uses float types (in certain drivers and arch's), but types referred by bpf program mostly will not contain float types.

Currently, both llvm and pahole will ignore floating types, so from type comparison point of view, we will be fine. The issue may come up with people using vmlinux BTF inside the kernel for different purpose where a float type needs to be specified, e.g., for a floating point function argument in ftrace. But this is rare too.

Quoted 2 lines

I am not 100% sure whether pahole dwarf/btf parity is considered as a strong reason to add floating point type support in BTF or not.

Isn't float and double just another base type with corresponding name? Fundamentally, how is it different from int or long? It's still just a collection of bits in 32-bit, 64-bit, etc number, except interpreted differently. Do you mean that for BTF to properly support floats, we'll need to extends set of possible encodings in INT type (in addition to today's CHAR, SIGNED and UNSIGNED)?

Quoted 6 lines
Quoted 6 lines

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

#16 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

Quoted 151 lines · Andrii Nakryiko
Quoted 147 lines · Yonghong Song
Quoted 127 lines · Andrii Nakryiko
Quoted 39 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

Quoted 3 lines

[386075] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none) Cannot open libc-2.28.so.debug
Failed to encode BTF

I've tried the same on libc-2.26.so.debug and it worked.

It failes after BTF encoding and deduping is done, exactly when it tries to re-open libc-2.28.so.debug to write out new contents. Not sure why it would fail, as it seems to be just plain open() call, but the fact that we both can repro this means there is something there.

Quoted 9 lines

It also seems that libc-2.28.so.debug is corrupted:

$ readelf -n /home/andriin/local/btf/libc-2.26.so.debug | grep
'Unknown note type' | wc -l
0
$ readelf -n /home/andriin/local/btf/libc-2.28.so.debug | grep
'Unknown note type' | wc -l
17953

Ignore everything I've said (though 'Unknown note type' issue is strange), for me it was just permissions issues.

$ chmod +rw /home/andriin/local/btf/libc-2.28.so.debug
$ ~/local/pahole/build/pahole -J ~/local/btf/libc-2.28.so.debug
$ PAHOLE=~/local/pahole/build/pahole ./btfdiff ~/local/btf/libc-2.28.so.debug
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
tmp/btfdiff.btf.22NQmJ 2019-02-25 13:31:54.802923668 -0800 renamed from tmp/btfdiff.dwarf.aV4wSL 2019-02-25 13:31:54.787923673 -0800 +12 -5
@@ -461,9 +461,15 @@ struct La_x86_64_retval {
          uint64_t                   lrv_rdx;              /*     8     8 */
          La_x86_64_xmm              lrv_xmm0;             /*    16    16 */
          La_x86_64_xmm              lrv_xmm1;             /*    32    16 */
-       long double                lrv_st0;              /*    48    16 */
+       long double                lrv_st0;              /*    48     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
          /* --- cacheline 1 boundary (64 bytes) --- */
-       long double                lrv_st1;              /*    64    16 */
+       long double                lrv_st1;              /*    64     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
          La_x86_64_vector           lrv_vector0;          /*    80    64 */
          /* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
          La_x86_64_vector           lrv_vector1;          /*   144    64 */
@@ -472,6 +478,7 @@ struct La_x86_64_retval {
          __int128                   lrv_bnd1;             /*   224    16 */
 
          /* size: 240, cachelines: 4, members: 10 */
+       /* sum members: 224, holes: 2, sum holes: 16 */
          /* last cacheline: 48 bytes */
   };
   struct r_debug {
@@ -2044,7 +2051,7 @@ union ieee754_float {
          } ieee_nan;                                    /*     0     4 */
   };
   union ieee854_long_double {
-       long double                d;                  /*     0    16 */
+       long double                d;                  /*     0     8 */
          struct {
                  unsigned int       mantissa1:32;       /*     0: 0  4 */
                  unsigned int       mantissa0:32;       /*     4: 0  4 */
@@ -2141,7 +2148,7 @@ struct ucontext_t {
          /* last cacheline: 8 bytes */
   };
   union ieee854_float128 {
-       _Float128                  d;                  /*     0    16 */
+       _Float128                  d;                  /*     0     0 */
          struct {
                  unsigned int       mantissa3:32;       /*     0: 0  4 */
                  unsigned int       mantissa2:32;       /*     4: 0  4 */
@@ -2219,7 +2226,7 @@ union printf_arg {
          long unsigned int          pa_u_long_int;      /*     0     8 */
          long long unsigned int     pa_u_long_long_int; /*     0     8 */
          double                     pa_double;          /*     0     8 */
-       long double                pa_long_double;     /*     0    16 */
+       long double                pa_long_double;     /*     0     8 */
          const char  *              pa_string;          /*     0     8 */
          const wchar_t  *           pa_wstring;         /*     0     8 */
          void *                     pa_pointer;         /*     0     8 */
Quoted 2 lines

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

BPF does not support floating point instructions. As the result, BTF design does not contain floating point types.

Looks like kernel also uses float types (in certain drivers and arch's), but types referred by bpf program mostly will not contain float types.

Currently, both llvm and pahole will ignore floating types, so from type comparison point of view, we will be fine. The issue may come up with people using vmlinux BTF inside the kernel for different purpose where a float type needs to be specified, e.g., for a floating point function argument in ftrace. But this is rare too.

Quoted 2 lines

I am not 100% sure whether pahole dwarf/btf parity is considered as a strong reason to add floating point type support in BTF or not.

Quoted 6 lines

Isn't float and double just another base type with corresponding name? Fundamentally, how is it different from int or long? It's still just a collection of bits in 32-bit, 64-bit, etc number, except interpreted differently. Do you mean that for BTF to properly support floats, we'll need to extends set of possible encodings in INT type (in addition to today's CHAR, SIGNED and UNSIGNED)?

Probably another kind like BTF_KIND_FLOAT with additional u32 (in addition to btf_type) indicating the size (4 bytes, 8 bytes and 16 bytes).

Quoted 7 lines
Quoted 7 lines
Quoted 7 lines

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

#17 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

Quoted 39 lines · Yonghong Song
Quoted 35 lines · Andrii Nakryiko
Quoted 34 lines · Yonghong Song
Quoted 30 lines · Andrii Nakryiko
Quoted 28 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

<snip>

Quoted 27 lines
Quoted 20 lines
Quoted 17 lines
Quoted 3 lines

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

BPF does not support floating point instructions. As the result, BTF design does not contain floating point types.

Looks like kernel also uses float types (in certain drivers and arch's), but types referred by bpf program mostly will not contain float types.

Currently, both llvm and pahole will ignore floating types, so from type comparison point of view, we will be fine. The issue may come up with people using vmlinux BTF inside the kernel for different purpose where a float type needs to be specified, e.g., for a floating point function argument in ftrace. But this is rare too.

Quoted 2 lines

I am not 100% sure whether pahole dwarf/btf parity is considered as a strong reason to add floating point type support in BTF or not.

Quoted 6 lines

Isn't float and double just another base type with corresponding name? Fundamentally, how is it different from int or long? It's still just a collection of bits in 32-bit, 64-bit, etc number, except interpreted differently. Do you mean that for BTF to properly support floats, we'll need to extends set of possible encodings in INT type (in addition to today's CHAR, SIGNED and UNSIGNED)?

Quoted 2 lines

Probably another kind like BTF_KIND_FLOAT with additional u32 (in addition to btf_type) indicating the size (4 bytes, 8 bytes and 16 bytes).

This sounds exactly like current format of BTF_KIND_INT. What BTF_KIND_FLOAT will have that can't be represented by BTF_KIND_INT? Wouldn't it be equivalent to just adding BTF_INT_FLOAT in addition to existing BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL to BTF_INT_ENCODING? Btw, while BTF_INT_SIGNED seems like it deserves separate bit, because it's mostly orthogonal to CHAR vs INT vs FLOAT vs (not so much) BOOL, BTF_INT_CHAR and BTF_INT_BOOL take two independent bits seems excessive. You can't have BOOL and CHAR, right? So why don't we just repurpose bits 1 and 2 to be a small 4-value enum:

enum btf_int_encoding {
    INT = 0,
    CHAR = 1,
    BOOL = 2,
    FLOAT = 3,
};

This is backwards compatible and covers all floats (long double vs double vs float is just a question of bitness).

Am I missing something here?

Quoted 8 lines
Quoted 8 lines
Quoted 8 lines
Quoted 8 lines

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

#18 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

Quoted 91 lines · Andrii Nakryiko
Quoted 39 lines · Yonghong Song
Quoted 35 lines · Andrii Nakryiko
Quoted 34 lines · Yonghong Song
Quoted 30 lines · Andrii Nakryiko
Quoted 28 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

<snip>

Quoted 27 lines
Quoted 20 lines
Quoted 17 lines
Quoted 3 lines

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

BPF does not support floating point instructions. As the result, BTF design does not contain floating point types.

Looks like kernel also uses float types (in certain drivers and arch's), but types referred by bpf program mostly will not contain float types.

Currently, both llvm and pahole will ignore floating types, so from type comparison point of view, we will be fine. The issue may come up with people using vmlinux BTF inside the kernel for different purpose where a float type needs to be specified, e.g., for a floating point function argument in ftrace. But this is rare too.

Quoted 2 lines

I am not 100% sure whether pahole dwarf/btf parity is considered as a strong reason to add floating point type support in BTF or not.

Quoted 6 lines

Isn't float and double just another base type with corresponding name? Fundamentally, how is it different from int or long? It's still just a collection of bits in 32-bit, 64-bit, etc number, except interpreted differently. Do you mean that for BTF to properly support floats, we'll need to extends set of possible encodings in INT type (in addition to today's CHAR, SIGNED and UNSIGNED)?

Quoted 2 lines

Probably another kind like BTF_KIND_FLOAT with additional u32 (in addition to btf_type) indicating the size (4 bytes, 8 bytes and 16 bytes).

This sounds exactly like current format of BTF_KIND_INT. What BTF_KIND_FLOAT will have that can't be represented by BTF_KIND_INT? Wouldn't it be equivalent to just adding BTF_INT_FLOAT in addition to existing BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL to BTF_INT_ENCODING? Btw, while BTF_INT_SIGNED seems like it deserves separate bit, because it's mostly orthogonal to CHAR vs INT vs FLOAT vs (not so much) BOOL, BTF_INT_CHAR and BTF_INT_BOOL take two independent bits seems excessive. You can't have BOOL and CHAR, right? So why don't we just repurpose bits 1 and 2 to be a small 4-value enum:

enum btf_int_encoding {
     INT = 0,
     CHAR = 1,
     BOOL = 2,
     FLOAT = 3,
};
Quoted 4 lines

This is backwards compatible and covers all floats (long double vs double vs float is just a question of bitness).

Am I missing something here?

float is not an "int". Encoding "float" in INT_KIND does not sound right in my opinion.

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 9 lines

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

#19 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

Quoted 95 lines · Yonghong Song
Quoted 91 lines · Andrii Nakryiko
Quoted 39 lines · Yonghong Song
Quoted 35 lines · Andrii Nakryiko
Quoted 34 lines · Yonghong Song
Quoted 30 lines · Andrii Nakryiko
Quoted 28 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

<snip>

Quoted 27 lines
Quoted 20 lines
Quoted 17 lines
Quoted 3 lines

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

BPF does not support floating point instructions. As the result, BTF design does not contain floating point types.

Looks like kernel also uses float types (in certain drivers and arch's), but types referred by bpf program mostly will not contain float types.

Currently, both llvm and pahole will ignore floating types, so from type comparison point of view, we will be fine. The issue may come up with people using vmlinux BTF inside the kernel for different purpose where a float type needs to be specified, e.g., for a floating point function argument in ftrace. But this is rare too.

Quoted 2 lines

I am not 100% sure whether pahole dwarf/btf parity is considered as a strong reason to add floating point type support in BTF or not.

Quoted 6 lines

Isn't float and double just another base type with corresponding name? Fundamentally, how is it different from int or long? It's still just a collection of bits in 32-bit, 64-bit, etc number, except interpreted differently. Do you mean that for BTF to properly support floats, we'll need to extends set of possible encodings in INT type (in addition to today's CHAR, SIGNED and UNSIGNED)?

Quoted 2 lines

Probably another kind like BTF_KIND_FLOAT with additional u32 (in addition to btf_type) indicating the size (4 bytes, 8 bytes and 16 bytes).

This sounds exactly like current format of BTF_KIND_INT. What BTF_KIND_FLOAT will have that can't be represented by BTF_KIND_INT? Wouldn't it be equivalent to just adding BTF_INT_FLOAT in addition to existing BTF_INT_SIGNED, BTF_INT_CHAR, BTF_INT_BOOL to BTF_INT_ENCODING? Btw, while BTF_INT_SIGNED seems like it deserves separate bit, because it's mostly orthogonal to CHAR vs INT vs FLOAT vs (not so much) BOOL, BTF_INT_CHAR and BTF_INT_BOOL take two independent bits seems excessive. You can't have BOOL and CHAR, right? So why don't we just repurpose bits 1 and 2 to be a small 4-value enum:

enum btf_int_encoding {
     INT = 0,
     CHAR = 1,
     BOOL = 2,
     FLOAT = 3,
};
Quoted 7 lines
Quoted 4 lines

This is backwards compatible and covers all floats (long double vs double vs float is just a question of bitness).

Am I missing something here?

float is not an "int". Encoding "float" in INT_KIND does not sound right in my opinion.

So isn't "bool". But in the end, it's just a collection of bits that are interpreted according to some encoding based on convention. From application perspective, none of float, int or bool has any more structure, those are "opaque" values. So maybe name BTF_KIND_INT is a bit unfortunate, it should be more like BTF_KIND_PRIMITIVE or something (or in pahole's lingo: base type).

Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

#20 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

In reply to Andrii Nakryiko

Em Mon, Feb 25, 2019 at 01:36:57PM -0800, Andrii Nakryiko escreveu:

Quoted 51 lines
Quoted 39 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

Quoted 3 lines

[386075] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none) Cannot open libc-2.28.so.debug
Failed to encode BTF

I've tried the same on libc-2.26.so.debug and it worked.

It failes after BTF encoding and deduping is done, exactly when it tries to re-open libc-2.28.so.debug to write out new contents. Not sure why it would fail, as it seems to be just plain open() call, but the fact that we both can repro this means there is something there.

Quoted 9 lines

It also seems that libc-2.28.so.debug is corrupted:

$ readelf -n /home/andriin/local/btf/libc-2.26.so.debug | grep
'Unknown note type' | wc -l
0
$ readelf -n /home/andriin/local/btf/libc-2.28.so.debug | grep
'Unknown note type' | wc -l
17953
Quoted 2 lines

Ignore everything I've said (though 'Unknown note type' issue is strange), for me it was just permissions issues.

yeah, we need to fix up these messages, I did the chmod here and got further.

[acme@quaco pahole]$ ls -la libc-2.28.so.debug -r--r--r--. 1 acme acme 17189368 Feb 25 17:04 libc-2.28.so.debug [acme@quaco pahole]$ pahole -J libc-2.28.so.debug Cannot open libc-2.28.so.debug
Failed to encode BTF
[acme@quaco pahole]$ chmod +w libc-2.28.so.debug [acme@quaco pahole]$ ls -la libc-2.28.so.debug -rw-rw-r--. 1 acme acme 17189368 Feb 25 17:04 libc-2.28.so.debug [acme@quaco pahole]$ pahole -J libc-2.28.so.debug [acme@quaco pahole]$

And I get the same warnings.

base_type__name_to_size: base_type _Float128

If we just add _Float128 to the base_type__name_to_size arrays I think its good enough for pretty printing, i.e. we won't know from encoding that this is a float or double, as CTF and DWARF can, but for printing purposes, even regenerating the type for consumption by other tools, it should just work, no?

So, two bugs, the bit_size for "long double" is 128, not 64, and if we add an entry for _Float128, then btfdiff is clean for libc, without any further work on properly supporting float types as in CTF or DWARF:

[acme@quaco pahole]$ btfdiff libc-2.28.so.debug
[acme@quaco pahole]$ git diff

dwarves.c +3 -2
index adc411a8279d..48a018a9c354 100644
@@ -168,11 +168,12 @@ static struct base_type_name_to_size {
        { .name = "double double",          .size = 64, },
        { .name = "single float",           .size = 32, },
        { .name = "float",                  .size = 32, },
-       { .name = "long double",            .size = 64, },
-       { .name = "long double long double", .size = 64, },
+       { .name = "long double",            .size = sizeof(long double) * 8, },
+       { .name = "long double long double", .size = sizeof(long double) * 8, },
        { .name = "__int128",               .size = 128, },
        { .name = "unsigned __int128",      .size = 128, },
        { .name = "__int128 unsigned",      .size = 128, },
+       { .name = "_Float128",              .size = 128, },
        { .name = NULL },
 };
 

[acme@quaco pahole]$

I'll add a series of patches fixing up the types that don't have encoded in its name the size in bits, so that it works on all arches, one more for _Float128, should be enough for now, right?

- Arnaldo

Quoted 80 lines
$ chmod +rw /home/andriin/local/btf/libc-2.28.so.debug
$ ~/local/pahole/build/pahole -J ~/local/btf/libc-2.28.so.debug
$ PAHOLE=~/local/pahole/build/pahole ./btfdiff ~/local/btf/libc-2.28.so.debug
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
tmp/btfdiff.btf.22NQmJ 2019-02-25 13:31:54.802923668 -0800 renamed from tmp/btfdiff.dwarf.aV4wSL 2019-02-25 13:31:54.787923673 -0800 +12 -5
@@ -461,9 +461,15 @@ struct La_x86_64_retval {
        uint64_t                   lrv_rdx;              /*     8     8 */
        La_x86_64_xmm              lrv_xmm0;             /*    16    16 */
        La_x86_64_xmm              lrv_xmm1;             /*    32    16 */
-       long double                lrv_st0;              /*    48    16 */
+       long double                lrv_st0;              /*    48     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
        /* --- cacheline 1 boundary (64 bytes) --- */
-       long double                lrv_st1;              /*    64    16 */
+       long double                lrv_st1;              /*    64     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
        La_x86_64_vector           lrv_vector0;          /*    80    64 */
        /* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
        La_x86_64_vector           lrv_vector1;          /*   144    64 */
@@ -472,6 +478,7 @@ struct La_x86_64_retval {
        __int128                   lrv_bnd1;             /*   224    16 */
 
        /* size: 240, cachelines: 4, members: 10 */
+       /* sum members: 224, holes: 2, sum holes: 16 */
        /* last cacheline: 48 bytes */
 };
 struct r_debug {
@@ -2044,7 +2051,7 @@ union ieee754_float {
        } ieee_nan;                                    /*     0     4 */
 };
 union ieee854_long_double {
-       long double                d;                  /*     0    16 */
+       long double                d;                  /*     0     8 */
        struct {
                unsigned int       mantissa1:32;       /*     0: 0  4 */
                unsigned int       mantissa0:32;       /*     4: 0  4 */
@@ -2141,7 +2148,7 @@ struct ucontext_t {
        /* last cacheline: 8 bytes */
 };
 union ieee854_float128 {
-       _Float128                  d;                  /*     0    16 */
+       _Float128                  d;                  /*     0     0 */
        struct {
                unsigned int       mantissa3:32;       /*     0: 0  4 */
                unsigned int       mantissa2:32;       /*     4: 0  4 */
@@ -2219,7 +2226,7 @@ union printf_arg {
        long unsigned int          pa_u_long_int;      /*     0     8 */
        long long unsigned int     pa_u_long_long_int; /*     0     8 */
        double                     pa_double;          /*     0     8 */
-       long double                pa_long_double;     /*     0    16 */
+       long double                pa_long_double;     /*     0     8 */
        const char  *              pa_string;          /*     0     8 */
        const wchar_t  *           pa_wstring;         /*     0     8 */
        void *                     pa_pointer;         /*     0     8 */

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

Signature · 1 line
- Arnaldo
#21 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

Quoted 72 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 01:36:57PM -0800, Andrii Nakryiko escreveu:

Quoted 51 lines
Quoted 39 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

Quoted 3 lines

[386075] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none) Cannot open libc-2.28.so.debug
Failed to encode BTF

I've tried the same on libc-2.26.so.debug and it worked.

It failes after BTF encoding and deduping is done, exactly when it tries to re-open libc-2.28.so.debug to write out new contents. Not sure why it would fail, as it seems to be just plain open() call, but the fact that we both can repro this means there is something there.

Quoted 9 lines

It also seems that libc-2.28.so.debug is corrupted:

$ readelf -n /home/andriin/local/btf/libc-2.26.so.debug | grep
'Unknown note type' | wc -l
0
$ readelf -n /home/andriin/local/btf/libc-2.28.so.debug | grep
'Unknown note type' | wc -l
17953
Quoted 2 lines

Ignore everything I've said (though 'Unknown note type' issue is strange), for me it was just permissions issues.

yeah, we need to fix up these messages, I did the chmod here and got further.

[acme@quaco pahole]$ ls -la libc-2.28.so.debug -r--r--r--. 1 acme acme 17189368 Feb 25 17:04 libc-2.28.so.debug [acme@quaco pahole]$ pahole -J libc-2.28.so.debug Cannot open libc-2.28.so.debug
Failed to encode BTF
[acme@quaco pahole]$ chmod +w libc-2.28.so.debug [acme@quaco pahole]$ ls -la libc-2.28.so.debug -rw-rw-r--. 1 acme acme 17189368 Feb 25 17:04 libc-2.28.so.debug [acme@quaco pahole]$ pahole -J libc-2.28.so.debug [acme@quaco pahole]$

And I get the same warnings.

Quoted 7 lines

base_type__name_to_size: base_type _Float128

If we just add _Float128 to the base_type__name_to_size arrays I think its good enough for pretty printing, i.e. we won't know from encoding that this is a float or double, as CTF and DWARF can, but for printing purposes, even regenerating the type for consumption by other tools, it should just work, no?

With "btf_loader: Simplify fixup code by relying on BTF data more" patch, btf_loader doesn't rely on this mapping at all. So if there is any other base type we didn't foresee, BTF output should still be correct.

Quoted 4 lines

So, two bugs, the bit_size for "long double" is 128, not 64, and if we add an entry for _Float128, then btfdiff is clean for libc, without any further work on properly supporting float types as in CTF or DWARF:

My worry with this table (and one of the reasons for that patch to not use it at all) was that those sizeof() sizes might be correct for the architecture on which pahole was built, but not necessarily on the target arch for which ELF file was built. E.g., I don't know if it's true that long double is always 128-bit on all architectures, so relying on that table might still break output for CTF and DWARF (but not BTF, because it takes bitness from BTF data itself).

But otherwise it's still good thing to complete that table.

Quoted 114 lines

[acme@quaco pahole]$ btfdiff libc-2.28.so.debug
[acme@quaco pahole]$ git diff

dwarves.c +3 -2
index adc411a8279d..48a018a9c354 100644
@@ -168,11 +168,12 @@ static struct base_type_name_to_size {
        { .name = "double double",          .size = 64, },
        { .name = "single float",           .size = 32, },
        { .name = "float",                  .size = 32, },
-       { .name = "long double",            .size = 64, },
-       { .name = "long double long double", .size = 64, },
+       { .name = "long double",            .size = sizeof(long double) * 8, },
+       { .name = "long double long double", .size = sizeof(long double) * 8, },
        { .name = "__int128",               .size = 128, },
        { .name = "unsigned __int128",      .size = 128, },
        { .name = "__int128 unsigned",      .size = 128, },
+       { .name = "_Float128",              .size = 128, },
        { .name = NULL },
 };
 

[acme@quaco pahole]$

I'll add a series of patches fixing up the types that don't have encoded in its name the size in bits, so that it works on all arches, one more for _Float128, should be enough for now, right?

- Arnaldo

Quoted 73 lines
$ chmod +rw /home/andriin/local/btf/libc-2.28.so.debug
$ ~/local/pahole/build/pahole -J ~/local/btf/libc-2.28.so.debug
$ PAHOLE=~/local/pahole/build/pahole ./btfdiff ~/local/btf/libc-2.28.so.debug
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
tmp/btfdiff.btf.22NQmJ 2019-02-25 13:31:54.802923668 -0800 renamed from tmp/btfdiff.dwarf.aV4wSL 2019-02-25 13:31:54.787923673 -0800 +12 -5
@@ -461,9 +461,15 @@ struct La_x86_64_retval {
        uint64_t                   lrv_rdx;              /*     8     8 */
        La_x86_64_xmm              lrv_xmm0;             /*    16    16 */
        La_x86_64_xmm              lrv_xmm1;             /*    32    16 */
-       long double                lrv_st0;              /*    48    16 */
+       long double                lrv_st0;              /*    48     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
        /* --- cacheline 1 boundary (64 bytes) --- */
-       long double                lrv_st1;              /*    64    16 */
+       long double                lrv_st1;              /*    64     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
        La_x86_64_vector           lrv_vector0;          /*    80    64 */
        /* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
        La_x86_64_vector           lrv_vector1;          /*   144    64 */
@@ -472,6 +478,7 @@ struct La_x86_64_retval {
        __int128                   lrv_bnd1;             /*   224    16 */
 
        /* size: 240, cachelines: 4, members: 10 */
+       /* sum members: 224, holes: 2, sum holes: 16 */
        /* last cacheline: 48 bytes */
 };
 struct r_debug {
@@ -2044,7 +2051,7 @@ union ieee754_float {
        } ieee_nan;                                    /*     0     4 */
 };
 union ieee854_long_double {
-       long double                d;                  /*     0    16 */
+       long double                d;                  /*     0     8 */
        struct {
                unsigned int       mantissa1:32;       /*     0: 0  4 */
                unsigned int       mantissa0:32;       /*     4: 0  4 */
@@ -2141,7 +2148,7 @@ struct ucontext_t {
        /* last cacheline: 8 bytes */
 };
 union ieee854_float128 {
-       _Float128                  d;                  /*     0    16 */
+       _Float128                  d;                  /*     0     0 */
        struct {
                unsigned int       mantissa3:32;       /*     0: 0  4 */
                unsigned int       mantissa2:32;       /*     4: 0  4 */
@@ -2219,7 +2226,7 @@ union printf_arg {
        long unsigned int          pa_u_long_int;      /*     0     8 */
        long long unsigned int     pa_u_long_long_int; /*     0     8 */
        double                     pa_double;          /*     0     8 */
-       long double                pa_long_double;     /*     0    16 */
+       long double                pa_long_double;     /*     0     8 */
        const char  *              pa_string;          /*     0     8 */
        const wchar_t  *           pa_wstring;         /*     0     8 */
        void *                     pa_pointer;         /*     0     8 */
Quoted 7 lines

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

--

- Arnaldo

#22 lore raw

Re: encoding BTF on glibc was: Re: [PATCH pahole 0/3] fix handling of bitfields in btf_loader

Em Tue, Feb 26, 2019 at 09:14:57AM -0800, Andrii Nakryiko escreveu:

Quoted 86 lines
Quoted 72 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 01:36:57PM -0800, Andrii Nakryiko escreveu:

Quoted 51 lines
Quoted 39 lines · Andrii Nakryiko
Quoted 22 lines · Arnaldo Carvalho de Melo

Em Mon, Feb 25, 2019 at 11:42:29AM -0800, Andrii Nakryiko escreveu:

Quoted 11 lines · On Mon, Feb 25, 2019 at 11:00 AM Arnaldo Carvalho de Melo>…
Quoted 9 lines

Em Mon, Feb 25, 2019 at 01:02:39PM -0300, Arnaldo Carvalho de Melo escreveu:

Quoted 2 lines

So yeah, the BTF encoder/decoder is working just fine, the problem is in pahole's DWARF code, lemme see...

Please try the patch below, for me btfdiff continues to show no diff for all types in my vmlinux and now it also produces the same output for when the first element of a bitfield has its bit_size equal to its byte_size * 8:

Quoted 4 lines

Yes, this fixes all the issues I've seen. btfdiff output is now empty for my kernel image. Thanks for quick fix!

Thanks, just for some extra testing I tried encoding BTF for glibc 2.28:

[acme@quaco pahole]$ pahole -JV libc-2.28.so.debug <BIG snip>

<snip>

Quoted 3 lines

[386075] INT long unsigned int size=8 bit_offset=0 nr_bits=64 encoding=(none) Cannot open libc-2.28.so.debug
Failed to encode BTF

I've tried the same on libc-2.26.so.debug and it worked.

It failes after BTF encoding and deduping is done, exactly when it tries to re-open libc-2.28.so.debug to write out new contents. Not sure why it would fail, as it seems to be just plain open() call, but the fact that we both can repro this means there is something there.

Quoted 9 lines

It also seems that libc-2.28.so.debug is corrupted:

$ readelf -n /home/andriin/local/btf/libc-2.26.so.debug | grep
'Unknown note type' | wc -l
0
$ readelf -n /home/andriin/local/btf/libc-2.28.so.debug | grep
'Unknown note type' | wc -l
17953
Quoted 2 lines

Ignore everything I've said (though 'Unknown note type' issue is strange), for me it was just permissions issues.

yeah, we need to fix up these messages, I did the chmod here and got further.

[acme@quaco pahole]$ ls -la libc-2.28.so.debug -r--r--r--. 1 acme acme 17189368 Feb 25 17:04 libc-2.28.so.debug [acme@quaco pahole]$ pahole -J libc-2.28.so.debug Cannot open libc-2.28.so.debug
Failed to encode BTF
[acme@quaco pahole]$ chmod +w libc-2.28.so.debug [acme@quaco pahole]$ ls -la libc-2.28.so.debug -rw-rw-r--. 1 acme acme 17189368 Feb 25 17:04 libc-2.28.so.debug [acme@quaco pahole]$ pahole -J libc-2.28.so.debug [acme@quaco pahole]$

And I get the same warnings.

Quoted 7 lines

base_type__name_to_size: base_type _Float128

If we just add _Float128 to the base_type__name_to_size arrays I think its good enough for pretty printing, i.e. we won't know from encoding that this is a float or double, as CTF and DWARF can, but for printing purposes, even regenerating the type for consumption by other tools, it should just work, no?

With "btf_loader: Simplify fixup code by relying on BTF data more" patch, btf_loader doesn't rely on this mapping at all. So if there is any other base type we didn't foresee, BTF output should still be correct.

Quoted 8 lines
Quoted 3 lines

So, two bugs, the bit_size for "long double" is 128, not 64, and if we add an entry for _Float128, then btfdiff is clean for libc, without any further work on properly supporting float types as in CTF or DWARF:

My worry with this table (and one of the reasons for that patch to not use it at all) was that those sizeof() sizes might be correct for the architecture on which pahole was built, but not necessarily on the target arch for which ELF file was built. E.g., I don't know if it's true that

Right, that is why there are some entries with size zero, meaning they need to be calculated based on the size of a pointer in the arch where it was produced, not the one where it is beind processed.

But yeah, I even made a note about that in the cset I added fixing the bit_size for "long double".

As you said, I have to look and see if we can use the same logic in DWARF that you used for BTF and in the end just ditch this table completely.

- Arnaldo

Quoted 120 lines

long double is always 128-bit on all architectures, so relying on that table might still break output for CTF and DWARF (but not BTF, because it takes bitness from BTF data itself).

But otherwise it's still good thing to complete that table.

Quoted 114 lines

[acme@quaco pahole]$ btfdiff libc-2.28.so.debug
[acme@quaco pahole]$ git diff

dwarves.c +3 -2
index adc411a8279d..48a018a9c354 100644
@@ -168,11 +168,12 @@ static struct base_type_name_to_size {
        { .name = "double double",          .size = 64, },
        { .name = "single float",           .size = 32, },
        { .name = "float",                  .size = 32, },
-       { .name = "long double",            .size = 64, },
-       { .name = "long double long double", .size = 64, },
+       { .name = "long double",            .size = sizeof(long double) * 8, },
+       { .name = "long double long double", .size = sizeof(long double) * 8, },
        { .name = "__int128",               .size = 128, },
        { .name = "unsigned __int128",      .size = 128, },
        { .name = "__int128 unsigned",      .size = 128, },
+       { .name = "_Float128",              .size = 128, },
        { .name = NULL },
 };
 

[acme@quaco pahole]$

I'll add a series of patches fixing up the types that don't have encoded in its name the size in bits, so that it works on all arches, one more for _Float128, should be enough for now, right?

- Arnaldo

Quoted 73 lines
$ chmod +rw /home/andriin/local/btf/libc-2.28.so.debug
$ ~/local/pahole/build/pahole -J ~/local/btf/libc-2.28.so.debug
$ PAHOLE=~/local/pahole/build/pahole ./btfdiff ~/local/btf/libc-2.28.so.debug
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
base_type__name_to_size: base_type _Float128
class__fixup_btf_bitfields: unknown base type name "_Float128"!
tmp/btfdiff.btf.22NQmJ 2019-02-25 13:31:54.802923668 -0800 renamed from tmp/btfdiff.dwarf.aV4wSL 2019-02-25 13:31:54.787923673 -0800 +12 -5
@@ -461,9 +461,15 @@ struct La_x86_64_retval {
        uint64_t                   lrv_rdx;              /*     8     8 */
        La_x86_64_xmm              lrv_xmm0;             /*    16    16 */
        La_x86_64_xmm              lrv_xmm1;             /*    32    16 */
-       long double                lrv_st0;              /*    48    16 */
+       long double                lrv_st0;              /*    48     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
        /* --- cacheline 1 boundary (64 bytes) --- */
-       long double                lrv_st1;              /*    64    16 */
+       long double                lrv_st1;              /*    64     8 */
+
+       /* XXX 8 bytes hole, try to pack */
+
        La_x86_64_vector           lrv_vector0;          /*    80    64 */
        /* --- cacheline 2 boundary (128 bytes) was 16 bytes ago --- */
        La_x86_64_vector           lrv_vector1;          /*   144    64 */
@@ -472,6 +478,7 @@ struct La_x86_64_retval {
        __int128                   lrv_bnd1;             /*   224    16 */
 
        /* size: 240, cachelines: 4, members: 10 */
+       /* sum members: 224, holes: 2, sum holes: 16 */
        /* last cacheline: 48 bytes */
 };
 struct r_debug {
@@ -2044,7 +2051,7 @@ union ieee754_float {
        } ieee_nan;                                    /*     0     4 */
 };
 union ieee854_long_double {
-       long double                d;                  /*     0    16 */
+       long double                d;                  /*     0     8 */
        struct {
                unsigned int       mantissa1:32;       /*     0: 0  4 */
                unsigned int       mantissa0:32;       /*     4: 0  4 */
@@ -2141,7 +2148,7 @@ struct ucontext_t {
        /* last cacheline: 8 bytes */
 };
 union ieee854_float128 {
-       _Float128                  d;                  /*     0    16 */
+       _Float128                  d;                  /*     0     0 */
        struct {
                unsigned int       mantissa3:32;       /*     0: 0  4 */
                unsigned int       mantissa2:32;       /*     4: 0  4 */
@@ -2219,7 +2226,7 @@ union printf_arg {
        long unsigned int          pa_u_long_int;      /*     0     8 */
        long long unsigned int     pa_u_long_long_int; /*     0     8 */
        double                     pa_double;          /*     0     8 */
-       long double                pa_long_double;     /*     0    16 */
+       long double                pa_long_double;     /*     0     8 */
        const char  *              pa_string;          /*     0     8 */
        const wchar_t  *           pa_wstring;         /*     0     8 */
        void *                     pa_pointer;         /*     0     8 */
Quoted 7 lines

Seems like "long double" size is invalid for BTF. And we should probably add "_Float128" as another "well known" base type name?

Though for the latter, it probably would be better to resolve base type size using BTF data itself.

Quoted 1 line
Quoted 1 line

[acme@quaco pahole]$

--

- Arnaldo

Signature · 1 line
- Arnaldo