On 13.01.20 16:24, Claudio Imbrenda wrote: > On Mon, 13 Jan 2020 15:43:28 +0100 > David Hildenbrand <david@xxxxxxxxxx> wrote: > >> On 13.01.20 15:05, Claudio Imbrenda wrote: >>> On Mon, 13 Jan 2020 14:10:10 +0100 >>> David Hildenbrand <david@xxxxxxxxxx> wrote: >>> >>> [...] >>> >>>> :) I'm confused by the fact that you "expect_pgm_int()" but >>>> actually don't expect one ... >>>> >>>> Please enlighten me why this isn't >>>> >>>> + sclp_mark_busy(); >>>> + h->length = 8; >>>> + sclp_setup_int(); >>>> + >>>> + asm volatile( >>>> + " .insn rre,0xb2204200,%1,%2\n" /* servc >>>> %1,%2 */ >>>> + " ipm %0\n" >>>> + " srl %0,28" >>>> + : "=&d" (cc) : "d" (valid_code), "a" >>>> (__pa(pagebuf)) >>>> + : "cc", "memory"); >>>> + if (!cc) >>>> + sclp_wait_busy(); >>>> + report(cc == 0, "Instruction format ignored bits"); >>>> >>>> I feel like I am missing something important. >>> >>> because if we take an unexpected pgm interrupt: >>> * the interrupt handler will write stuff on the console using SCLP >>> * it will wait for the busy flag to be cleared before doing so >>> * thus it will hang. >>> >>> this would be solved by adding special logic to the pgm interrupt >>> handler (as we have discussed in your previous email) >>> >> >> I see, so the issue should hold for all SCLP checks where we don't >> expect an exception ... hmmm > > which is why my wrapper in the unit test is so complicated :) > so .... if we would implement my suggestion (if we get an exception on a servc instruction, clear sclp_busy) that code would get simplified as well? :) It would be really beneficial if we could rely on expect_pgm_int() ... check_pgm_int_code(whatever) if we *expect an interrupt and none of that if we don't. So if it's a matter of clearing sclp_busy on PGM exceptions, that shouldn't be too hard ... but maybe I am missing something (haven't looked again at the whole patch ...). -- Thanks, David / dhildenb
