On Sun, Feb 14, 2021 at 11:00:48AM -0800, Linus Torvalds wrote: > On Sun, Feb 14, 2021 at 10:42 AM Ramsay Jones > <ramsay@xxxxxxxxxxxxxxxxxxxx> wrote: > > > > > > > > I looked around but didn't find any hints how to fix this. Any pointers > > > I missed (added the sparse list to cc:)? > > > > This is a limitation of sparse; when using the 'stringize' pre-processor > > operator #, the maximum size of the resulting string is about 8k (if I > > remember correctly). > > Well, yes and no. > > The C89 standard actually says that a string literal can be at most > 509 characters to be portable. C99 increased it to 4095 characters. > > Sparse makes the limit higher, and the limit could easily be expanded > way past 8kB - but the point is that large string literals are > actually not guaranteed to be valid C. > > So honestly, it really sounds like that TRACE_EVENT() thing is doing > something it shouldn't be doing. In itself, it's OKish but it does a lot of macro expansions and most arguments are macros of macros of ... but the problem seems to be limited to TP_printk(). In the current case, the offender is the string 'print_fmt_tps6598x_status' which is just under 26K long especially because it expand TPS6598X_STATUS_FLAGS_MASK but also because the arguments use FIELD_GET() and thus __BF_FIELD_CHECK(). > > I don't think there's any fundamental limit why sparse does 8kB as a > limit (just a few random buffers). Making sparse accept larger ones > should be as simple as just increasing MAX_STRING, but I really don't > think the kernel should encourage that kind of excessive string sizes. Like you noted, there are just a few cases in the kernel and IIRC there is or was one case in it too. I would tend to increase MAX_STRING to something like 32 or 64K, in order to keep it reasonable but let sparse to continue its processing, but add a warning when the string/token is bigger than the current 8K. -- Luc
