| Message ID | 20190131071545.31869-5-oohall@gmail.com |
|---|---|
| State | Superseded |
| Headers | show |
| Series | [v2,1/5] phb4: Rework BDFN filtering in phb4_set_pe() | expand |
| Context | Check | Description |
|---|---|---|
| snowpatch_ozlabs/apply_patch | success | master/apply_patch Successfully applied |
| snowpatch_ozlabs/snowpatch_job_snowpatch-skiboot | success | Test snowpatch/job/snowpatch-skiboot on branch master |
On 31/1/19 6:15 pm, Oliver O'Halloran wrote: > I now know what an IODA cache is and I'm not happy about it. With > the power of Comments™ you too can share the misery. > > Signed-off-by: Oliver O'Halloran <oohall@gmail.com> Sigh. Reviewed-by: Andrew Donnellan <andrew.donnellan@au1.ibm.com> > --- > hw/phb4.c | 31 ++++++++++++------------------- > 1 file changed, 12 insertions(+), 19 deletions(-) > > diff --git a/hw/phb4.c b/hw/phb4.c > index 9ece0f05cd41..216915da99b4 100644 > --- a/hw/phb4.c > +++ b/hw/phb4.c > @@ -898,30 +898,23 @@ static uint64_t phb4_default_mbt0(struct phb4 *p, unsigned int bar_idx) > return mbt0; > } > > -/* Clear IODA cache tables */ > +/* > + * Clear the saved (cached) IODA state. > + * > + * The caches here are used to save the configuration of the IODA tables > + * done by the OS. When the PHB is reset it loses all of its internal state > + * so we need to keep a copy to restore from. > + */ > static void phb4_init_ioda_cache(struct phb4 *p) > { > uint32_t i; > > /* > - * RTT and PELTV. RTE should be 0xFF's to indicate > - * invalid PE# for the corresponding RID. > - * > - * Note: Instead we set all RTE entries to 0x00 to > - * work around a problem where PE lookups might be > - * done before Linux has established valid PE's > - * (during PCI probing). We can revisit that once/if > - * Linux has been fixed to always setup valid PEs. > - * > - * The value 0x00 corresponds to the default PE# Linux > - * uses to check for config space freezes before it > - * has assigned PE# to busses. > - * > - * WARNING: Additionally, we need to be careful, there's > - * a HW issue, if we get an MSI on an RTT entry that is > - * FF, things will go bad. We need to ensure we don't > - * ever let a live FF RTT even temporarily when resetting > - * for EEH etc... (HW278969). > + * The RTT entries (RTE) are supposed to be initialised to > + * 0xFF which indicates an invalid PE# for that RTT index > + * (the bdfn). However, we set them to 0x00 since Linux > + * needs to find the devices first by scanning config space > + * and this occurs before PEs have been assigned. > */ > for (i = 0; i < RTT_TABLE_ENTRIES; i++) > p->tbl_rtt[i] = PHB4_RESERVED_PE_NUM(p); >
diff --git a/hw/phb4.c b/hw/phb4.c index 9ece0f05cd41..216915da99b4 100644 --- a/hw/phb4.c +++ b/hw/phb4.c @@ -898,30 +898,23 @@ static uint64_t phb4_default_mbt0(struct phb4 *p, unsigned int bar_idx) return mbt0; } -/* Clear IODA cache tables */ +/* + * Clear the saved (cached) IODA state. + * + * The caches here are used to save the configuration of the IODA tables + * done by the OS. When the PHB is reset it loses all of its internal state + * so we need to keep a copy to restore from. + */ static void phb4_init_ioda_cache(struct phb4 *p) { uint32_t i; /* - * RTT and PELTV. RTE should be 0xFF's to indicate - * invalid PE# for the corresponding RID. - * - * Note: Instead we set all RTE entries to 0x00 to - * work around a problem where PE lookups might be - * done before Linux has established valid PE's - * (during PCI probing). We can revisit that once/if - * Linux has been fixed to always setup valid PEs. - * - * The value 0x00 corresponds to the default PE# Linux - * uses to check for config space freezes before it - * has assigned PE# to busses. - * - * WARNING: Additionally, we need to be careful, there's - * a HW issue, if we get an MSI on an RTT entry that is - * FF, things will go bad. We need to ensure we don't - * ever let a live FF RTT even temporarily when resetting - * for EEH etc... (HW278969). + * The RTT entries (RTE) are supposed to be initialised to + * 0xFF which indicates an invalid PE# for that RTT index + * (the bdfn). However, we set them to 0x00 since Linux + * needs to find the devices first by scanning config space + * and this occurs before PEs have been assigned. */ for (i = 0; i < RTT_TABLE_ENTRIES; i++) p->tbl_rtt[i] = PHB4_RESERVED_PE_NUM(p);
I now know what an IODA cache is and I'm not happy about it. With the power of Comments™ you too can share the misery. Signed-off-by: Oliver O'Halloran <oohall@gmail.com> --- hw/phb4.c | 31 ++++++++++++------------------- 1 file changed, 12 insertions(+), 19 deletions(-)