On 7/9/19 7:10 PM, Dr. David Alan Gilbert wrote: > * Philippe Mathieu-Daudé (phi...@redhat.com) wrote: >> Hi David, >> >> On 7/9/19 12:30 PM, Dr. David Alan Gilbert wrote: >>> * Philippe Mathieu-Daudé (phi...@redhat.com) wrote: >>>> In the "Read Array Flowchart" the command has a value of 0xFF. >>>> >>>> In the document [*] the "Read Array Flowchart", the READ_ARRAY >>>> command has a value of 0xff. >>>> >>>> Use the correct value in the pflash model. >>>> >>>> There is no change of behavior in the guest, because: >>>> - when the guest were sending 0xFF, the reset_flash label >>>> was setting the command value as 0x00 >>>> - 0x00 was used internally for READ_ARRAY >>>> >>>> To keep migration behaving correctly, we have to increase >>>> the VMState version. When migrating from an older version, >>>> we use the correct command value. >>> >>> The problem is that incrementing the version will break backwards >>> compatibility; so you won't be able to migrate this back to an older >>> QEMU version; so for example a q35/uefi with this won't be able >>> to migrate backwards to a 4.0.0 or older qemu. >>> >>> So instead of bumping the version_id you probably need to wire >>> the behaviour to a machine type and then on your new type >>> wire a subsection containing a flag; the reception of that subsection >>> tells you to use the new/correct semantics. >> >> I'm starting to understand VMState subsections, but it might be overkill >> for this change... >> >> Subsections >> ----------- >> >> The most common structure change is adding new data, e.g. when adding >> a newer form of device, or adding that state that you previously >> forgot to migrate. This is best solved using a subsection. >> >> This is not the case here, the field is already present and migrated. >> >> It seems I can use a simple pre_save hook, always migrating the >> READ_ARRAY using the incorrect value: >> >> -- >8 -- >> --- a/hw/block/pflash_cfi01.c >> +++ b/hw/block/pflash_cfi01.c >> @@ -97,12 +97,29 @@ struct PFlashCFI01 { >> bool incorrect_read_array_command; >> }; >> >> +static int pflash_pre_save(void *opaque) >> +{ >> + PFlashCFI01 *s = opaque; >> + >> + /* >> + * Previous to QEMU v4.1 an incorrect value of 0x00 was used for the >> + * READ_ARRAY command. To preserve migrating to these older version, >> + * always migrate the READ_ARRAY command as 0x00. >> + */ >> + if (s->cmd == 0xff) { >> + s->cmd = 0x00; >> + } >> + >> + return 0; >> +} > > Be careful what happens if migration fails and you continue on the > source - is that OK - or are you going to have to flip that back somehow > (in a post_save).
Hmm OK... > > Another way to do the same is to have a dummy field; tmp_cmd, and the > tmp_cmd is the thing that's actually migrated and filled by pre_save > (or use VMSTATE_WITH_TMP ) > > >> static int pflash_post_load(void *opaque, int version_id); >> >> static const VMStateDescription vmstate_pflash = { >> .name = "pflash_cfi01", >> .version_id = 1, >> .minimum_version_id = 1, >> + .pre_save = pflash_pre_save, >> .post_load = pflash_post_load, >> .fields = (VMStateField[]) { >> VMSTATE_UINT8(wcycle, PFlashCFI01), >> @@ -1001,5 +1018,14 @@ static int pflash_post_load(void *opaque, int >> version_id) >> pfl->vmstate = qemu_add_vm_change_state_handler(postload_update_cb, >> pfl); >> } >> + >> + /* >> + * Previous to QEMU v4.1 an incorrect value of 0x00 was used for the >> + * READ_ARRAY command. >> + */ >> + if (pfl->cmd == 0x00) { >> + pfl->cmd = 0xff; >> + } >> + >> return 0; >> } >> --- >> >> Being simpler and less intrusive (no new property in hw/core/machine.c), >> is this acceptable? > > From the migration point of view yes; I don't know enough about pflash > to say if it makes sense; for example could there ever be a 00 command > really used and then you'd have to distinguish that somehow? Well, I think this change is simpler than it looks. Currently the code is (what will run on older guest): static uint32_t pflash_read(PFlashCFI01 *pfl, hwaddr offset, int width, int be) { switch (pfl->cmd) { default: /* This should never happen : reset state & treat it as a read*/ DPRINTF("%s: unknown command state: %x\n", __func__, pfl->cmd); pfl->wcycle = 0; pfl->cmd = 0; /* fall through to read code */ case 0x00: /* Flash area read */ ret = pflash_data_read(pfl, offset, width, be); break; and: static void pflash_write(PFlashCFI01 *pfl, hwaddr offset, uint32_t value, int width, int be) { switch (pfl->wcycle) { case 0: /* read mode */ switch (cmd) { case 0x00: /* ??? */ goto reset_flash; case 0xff: /* Read array mode */ DPRINTF("%s: Read array mode\n", __func__); goto reset_flash; default: goto error_flash; } So current (incorrect) 0x00 will be then 0xff, and will be backprocessed correctly. What I want is to get ride of this 0x00 processing that is confusing, the spec and the guests use 0xff for READ_ARRAY. So if I increase version I can not back-migrate, but luckily it seems I can simply update 0x00 -> 0xff without even increasing the version :) (I'm reluctant to skip this patch because I'd rather avoid Laszlo to re-run his tests). Regards, Phil.