1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
|
user-level: 0.534
TCG: 0.514
VMM: 0.466
peripherals: 0.451
hypervisor: 0.412
arm: 0.409
ppc: 0.407
x86: 0.403
mistranslation: 0.390
virtual: 0.383
i386: 0.378
architecture: 0.376
permissions: 0.374
KVM: 0.371
PID: 0.363
device: 0.344
vnc: 0.343
performance: 0.338
socket: 0.336
register: 0.334
network: 0.331
semantic: 0.329
graphic: 0.327
kernel: 0.323
files: 0.321
risc-v: 0.319
debug: 0.296
assembly: 0.285
boot: 0.285
hw/scsi/scsi-disk.c line 2554 allocation overflow
When compiling qemu from git master (at commit 03bf012e523ecdf047ac56b2057950247256064d ) on Linux amd64, with gcc-9 9.2.1 , and using `-march=native -flto`, during linking of most target binaries, compiler does detect an issue with allocation in scsi_disk_new_request_dump and aborts compilation.
make[1]: Entering directory '/home/user/qemu/slirp'
make[1]: Nothing to be done for 'all'.
make[1]: Leaving directory '/home/user/qemu/slirp'
nm: stats64.o: no symbols
LINK aarch64-softmmu/qemu-system-aarch64
In function ‘scsi_disk_new_request_dump’,
inlined from ‘scsi_new_request’ at hw/scsi/scsi-disk.c:2580:9,
inlined from ‘scsi_new_request’ at hw/scsi/scsi-disk.c:2564:21:
hw/scsi/scsi-disk.c:2554:19: error: argument 1 value ‘18446744073709551612’ exceeds maximum object size 9223372036854775807 [-Werror=alloc-size-larger-than=]
hw/scsi/scsi-disk.c: In function ‘scsi_new_request’:
/usr/include/glib-2.0/glib/gmem.h:78:10: note: in a call to allocation function ‘g_malloc’ declared here
78 | gpointer g_malloc (gsize n_bytes) G_GNUC_MALLOC G_GNUC_ALLOC_SIZE(1);
| ^
lto1: all warnings being treated as errors
lto-wrapper: fatal error: c++ returned 1 exit status
compilation terminated.
/usr/bin/ld: error: lto-wrapper failed
collect2: error: ld returned 1 exit status
same happens for most other targets: alpha-softmmu/qemu-system-alpha arm-softmmu/qemu-system-arm hppa-softmmu/qemu-system-hppa i386-softmmu/qemu-system-i386 lm32-softmmu/qemu-system-lm32 mips-softmmu/qemu-system-mips mips64-softmmu/qemu-system-mips64 mips64el-softmmu/qemu-system-mips64el mipsel-softmmu/qemu-system-mipsel ppc-softmmu/qemu-system-ppc ppc64-softmmu/qemu-system-ppc64 riscv32-softmmu/qemu-system-riscv32 riscv64-softmmu/qemu-system-riscv64 s390x-softmmu/qemu-system-s390x sh4-softmmu/qemu-system-sh4 sh4eb-softmmu/qemu-system-sh4eb sparc-softmmu/qemu-system-sparc sparc64-softmmu/qemu-system-sparc64 x86_64-softmmu/qemu-system-x86_64 xtensa-softmmu/qemu-system-xtensa xtensaeb-softmmu/qemu-system-xtensaeb
Notice -softmmu being a common factor here.
The size of the allocation for the temporary buffer for dumping using snprintf is determined based on the size of the buffer via call to scsi_cdb_length. I believe the heavy inlining and constant propagation makes scsi_cdb_length return -1, so len = -1. Then allocation size is 5*len + 1, or -4. Which overflows to 2^64 - 4 or so.
The case of len==-1 from scsi_cdb_length happens if the (buf[0] >> 5) is not 0, 1, 2, 4 or 5.
However, I can't find out how gcc figures out that buf[0] is not one of these variables. To me looking at this function, compiler should not know anything about buf[0].
I tried following the chain of calls back, including devirtualize alloc_req, and I found scsi_device_alloc_req calling these alloc_req callbacks, but it is itself called from scsi_req_new, which is called in get_scsi_requests , just after buf is filled from QEMUFile using qemu_get_buffer, which ultimately goes even further into read paths, which there might be many AFAIK.
glib2 version 2.62.1-1
FYI. Adding if (len <= 0) return; in the scsi_disk_new_request_dump solved the compilation issue for me.
So indeed gcc thinks len == -1
I am pretty sure the build qemu is functional, as this path is only taken if the trace_event_get_state_backends(TRACE_SCSI_DISK_NEW_REQUEST) is true, which by default it is not.
BTW. Also, aarch64-softmmu/qemu-system-aarch64 takes very long time to link compared to other targets, so I recommend using -flto=16 to increase parallelism, and reduce lto link time to about 4 minutes. (But 64GB of memory recommended).
I also tested with --disable-slirp configure flag. Still same issue.
The QEMU project is currently considering to move its bug tracking to
another system. For this we need to know which bugs are still valid
and which could be closed already. Thus we are setting older bugs to
"Incomplete" now.
If you still think this bug report here is valid, then please switch
the state back to "New" within the next 60 days, otherwise this report
will be marked as "Expired". Or please mark it as "Fix Released" if
the problem has been solved with a newer version of QEMU already.
Thank you and sorry for the inconvenience.
Likely not happening anymore since commit e91bae8e98a ("scsi: Silence gcc warning").
[Expired for QEMU because there has been no activity for 60 days.]
|