Bug 1668703 - gcc9 failure with inline asm on s390
Summary: gcc9 failure with inline asm on s390
Keywords:
Status: CLOSED NOTABUG
Alias: None
Product: Fedora
Classification: Fedora
Component: gcc
Version: rawhide
Hardware: Unspecified
OS: Unspecified
unspecified
unspecified
Target Milestone: ---
Assignee: Jakub Jelinek
QA Contact: Fedora Extras Quality Assurance
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2019-01-23 11:31 UTC by Laura Abbott
Modified: 2019-01-23 12:46 UTC (History)
10 users (show)

Fixed In Version:
Doc Type: If docs needed, set a value
Doc Text:
Clone Of:
Environment:
Last Closed: 2019-01-23 12:19:20 UTC
Type: Bug
Embargoed:


Attachments (Terms of Use)
original file (28.38 KB, text/x-csrc)
2019-01-23 11:31 UTC, Laura Abbott
no flags Details
preprocessed file ('make init/main.i') (2.37 MB, text/plain)
2019-01-23 11:32 UTC, Laura Abbott
no flags Details
generated assembly ('make init/main.s') (1.27 MB, text/plain)
2019-01-23 11:33 UTC, Laura Abbott
no flags Details

Description Laura Abbott 2019-01-23 11:31:55 UTC
Created attachment 1522612 [details]
original file

gcc9 fails on this inline asm snippet in the kernel:

static inline bool arch_static_branch(struct static_key *key, bool branch)
{
        asm_volatile_goto("0:   brcl    0,"__stringify(JUMP_LABEL_NOP_OFFSET)"\n"
                          ".pushsection __jump_table,\"aw\"\n"
                          ".balign      8\n"
                          ".long        0b-.,%l[label]-.\n"
                          ".quad        %c0-.\n"
                          ".popsection\n"
                          : : "X" (&((char *)key)[branch]) : : label);
        return false;
label:
        return true;
}


BUILDSTDERR: ./arch/s390/include/asm/jump_label.h: Assembler messages:
BUILDSTDERR: ./arch/s390/include/asm/jump_label.h:23: Error: bad expression
BUILDSTDERR: ./arch/s390/include/asm/jump_label.h:23: Error: junk at end of line, first unrecognized character is `r'
BUILDSTDERR: make[1]: *** [scripts/Makefile.build:277: init/main.o] Error 1


This seems to be from this block:

static int __init initcall_blacklist(char *str)
{
        char *str_entry;
        struct blacklist_entry *entry;

        /* str argument is a comma-separated list of functions */
        do {
                str_entry = strsep(&str, ",");
                if (str_entry) {
                        pr_debug("blacklisting initcall %s\n", str_entry);
                        entry = memblock_alloc(sizeof(*entry),
                                               SMP_CACHE_BYTES);
                        entry->buf = memblock_alloc(strlen(str_entry) + 1,
                                                    SMP_CACHE_BYTES);
                        strcpy(entry->buf, str_entry);
                        list_add(&entry->next, &blacklisted_initcalls);
                }
        } while (str_entry);

        return 0;
}

specifically, the strsep. 

# 19 "./arch/s390/include/asm/jump_label.h" 1
        0:      brcl    0,2
.pushsection __jump_table,"aw"
.balign 8
.long   0b-.,.L249-.    #
.quad   %r10-.  # tmp98
.popsection


This looks like the bad assembly, it's generating a referenece to a register.

Comment 1 Laura Abbott 2019-01-23 11:32:39 UTC
Created attachment 1522614 [details]
preprocessed file ('make init/main.i')

Comment 2 Laura Abbott 2019-01-23 11:33:13 UTC
Created attachment 1522615 [details]
generated assembly ('make init/main.s')

Comment 3 Florian Weimer 2019-01-23 12:11:50 UTC
The "X" constraint looks wrong here because it includes registers.  "i" might work instead.

Comment 4 Jakub Jelinek 2019-01-23 12:19:20 UTC
Yeah, "X" means anything, it should be used only if you don't reference the operand in the asm pattern at all.  The usual use case for "X" is if you need e.g. a scratch register in one alternative, but not in another one, so you use e.g. =r when you need it and =X if you don't.
"i" indeed covers non-symbolic as well as symbolic constants, might not work with e.g. -fpic if a PIC base needs to be added, but for -fpic this technique will not really work anyway.

Comment 5 Laura Abbott 2019-01-23 12:46:08 UTC
Thanks for the explanation, I'll talk to people upstream about getting this fixed.


Note You need to log in before you can comment on or make changes to this bug.