| Message ID | 1440022048-6285-2-git-send-email-al.stone@linaro.org |
|---|---|
| State | New |
| Headers | show |
Hi Al, On 19/08/15 23:07, Al Stone wrote: I finally got a chance to try this series on Juno. Well it exposed a firmware bug in MADT table :) [..] > acpi_tbl_entry_handler handler, > @@ -245,6 +484,8 @@ acpi_parse_entries(char *id, unsigned long table_size, > table_end) { > if (entry->type == entry_id > && (!max_entries || count < max_entries)) { > + if (bad_madt_entry(table_header, entry)) > + return -EINVAL; Not sure if we can have the above check here unconditionally. Currently I can see there are 2 other users of acpi_parse_entries i.e. PCC and NUMA. So may be it can be made conditional or return success for non-MADT tables from bad_madt_entry ? Other than that, you can add for ARM64 specific parts: Reviewed-and-tested-by: Sudeep Holla <sudeep.holla@arm.com> Regards, Sudeep
On 09/07/2015 09:32 AM, Sudeep Holla wrote: > Hi Al, > > On 19/08/15 23:07, Al Stone wrote: > > I finally got a chance to try this series on Juno. Well it exposed a firmware > bug in MADT table :) What? The code did what it was supposed to do :-)? Very cool. Good to know. I talked to Graeme a bit, too, and he had some good suggestions for clean up. I'll post a v3 tomorrow. > [..] > >> acpi_tbl_entry_handler handler, >> @@ -245,6 +484,8 @@ acpi_parse_entries(char *id, unsigned long table_size, >> table_end) { >> if (entry->type == entry_id >> && (!max_entries || count < max_entries)) { >> + if (bad_madt_entry(table_header, entry)) >> + return -EINVAL; > > Not sure if we can have the above check here unconditionally. > Currently I can see there are 2 other users of acpi_parse_entries i.e. > PCC and NUMA. So may be it can be made conditional or return success for > non-MADT tables from bad_madt_entry ? I'll double check these uses. I thought I had before, and based on what I saw the check would be reasonable. It never hurts to check again, though. > Other than that, you can add for ARM64 specific parts: > Reviewed-and-tested-by: Sudeep Holla <sudeep.holla@arm.com> Thanks! > Regards, > Sudeep > -- > To unsubscribe from this list: send the line "unsubscribe linux-acpi" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html
On 09/07/2015 09:32 AM, Sudeep Holla wrote: > Hi Al, > > On 19/08/15 23:07, Al Stone wrote: > > I finally got a chance to try this series on Juno. Well it exposed a firmware > bug in MADT table :) > > [..] > >> acpi_tbl_entry_handler handler, >> @@ -245,6 +484,8 @@ acpi_parse_entries(char *id, unsigned long table_size, >> table_end) { >> if (entry->type == entry_id >> && (!max_entries || count < max_entries)) { >> + if (bad_madt_entry(table_header, entry)) >> + return -EINVAL; > > Not sure if we can have the above check here unconditionally. > Currently I can see there are 2 other users of acpi_parse_entries i.e. > PCC and NUMA. So may be it can be made conditional or return success for > non-MADT tables from bad_madt_entry ? So, I went back and double checked the other users and they're looking at the return value for acpi_parse_entries properly; adding in the check above unconditionally should not cause any behavior change. Further, despite the name, acpi_parse_entries is only used to examine MADT subtables. Granted, we should probably make the name clearer at some point (too ambiguous as to which entries are parsed right now). Nonetheless, current usage seems to be in order.
On 09/09/15 20:57, Al Stone wrote: > On 09/07/2015 09:32 AM, Sudeep Holla wrote: >> Hi Al, >> >> On 19/08/15 23:07, Al Stone wrote: >> >> I finally got a chance to try this series on Juno. Well it exposed a firmware >> bug in MADT table :) >> >> [..] >> >>> acpi_tbl_entry_handler handler, >>> @@ -245,6 +484,8 @@ acpi_parse_entries(char *id, unsigned long table_size, >>> table_end) { >>> if (entry->type == entry_id >>> && (!max_entries || count < max_entries)) { >>> + if (bad_madt_entry(table_header, entry)) >>> + return -EINVAL; >> >> Not sure if we can have the above check here unconditionally. >> Currently I can see there are 2 other users of acpi_parse_entries i.e. >> PCC and NUMA. So may be it can be made conditional or return success for >> non-MADT tables from bad_madt_entry ? > > So, I went back and double checked the other users and they're looking at > the return value for acpi_parse_entries properly; adding in the check above > unconditionally should not cause any behavior change. I disagree. I populated PCCT table on Juno to get this error for PCCT(PCCT header gets interpreted as MADT header): " ACPI: undefined version for either FADT 5.1 or MADT 1 Error parsing PCC subspaces from PCCT " And here the stacktrace: [<ffffffc000881e58>] bad_madt_entry+0x90/0x16c [<ffffffc000882030>] acpi_table_parse_entries+0xfc/0x180 [<ffffffc000895af8>] pcc_init+0x70/0x148 > Further, despite the name, acpi_parse_entries is only used to examine MADT > subtables. Granted, we should probably make the name clearer at some point > (too ambiguous as to which entries are parsed right now). Nonetheless, current > usage seems to be in order. > From the code inspection, I can see we have 3 users of acpi_parse_entries not just MADT but also PCC and NUMA/SRAT Something like this solves this issue: - if (bad_madt_entry(table_header, entry)) + if (!strncmp(id, ACPI_SIG_MADT, 4) && + bad_madt_entry(table_header, entry) Or am I still missing something ? Regards, Sudeep
On 09/10/2015 10:20 AM, Sudeep Holla wrote: > > > On 09/09/15 20:57, Al Stone wrote: >> On 09/07/2015 09:32 AM, Sudeep Holla wrote: >>> Hi Al, >>> >>> On 19/08/15 23:07, Al Stone wrote: >>> >>> I finally got a chance to try this series on Juno. Well it exposed a firmware >>> bug in MADT table :) >>> >>> [..] >>> >>>> acpi_tbl_entry_handler handler, >>>> @@ -245,6 +484,8 @@ acpi_parse_entries(char *id, unsigned long table_size, >>>> table_end) { >>>> if (entry->type == entry_id >>>> && (!max_entries || count < max_entries)) { >>>> + if (bad_madt_entry(table_header, entry)) >>>> + return -EINVAL; >>> >>> Not sure if we can have the above check here unconditionally. >>> Currently I can see there are 2 other users of acpi_parse_entries i.e. >>> PCC and NUMA. So may be it can be made conditional or return success for >>> non-MADT tables from bad_madt_entry ? >> >> So, I went back and double checked the other users and they're looking at >> the return value for acpi_parse_entries properly; adding in the check above >> unconditionally should not cause any behavior change. > > I disagree. I populated PCCT table on Juno to get this error for > PCCT(PCCT header gets interpreted as MADT header): > " > ACPI: undefined version for either FADT 5.1 or MADT 1 > Error parsing PCC subspaces from PCCT > " > And here the stacktrace: > [<ffffffc000881e58>] bad_madt_entry+0x90/0x16c > [<ffffffc000882030>] acpi_table_parse_entries+0xfc/0x180 > [<ffffffc000895af8>] pcc_init+0x70/0x148 > >> Further, despite the name, acpi_parse_entries is only used to examine MADT >> subtables. Granted, we should probably make the name clearer at some point >> (too ambiguous as to which entries are parsed right now). Nonetheless, current >> usage seems to be in order. >> > > From the code inspection, I can see we have 3 users of acpi_parse_entries not > just MADT but also PCC and NUMA/SRAT > > Something like this solves this issue: > - if (bad_madt_entry(table_header, entry)) > + if (!strncmp(id, ACPI_SIG_MADT, 4) && > + bad_madt_entry(table_header, entry) > > > Or am I still missing something ? > > Regards, > Sudeep Nope, I missed it. Your fix above will solve the problem; I misunderstood how acpi_parse_entries() was being used -- somehow I had it in my head that only MADT was in use, and just not seeing that it's being used for several other subtable traversals also. Sorry about that, Sudeep. My mistake. I'll add this fix for a v4, but I'll wait for a few days to see if I get any additional comments -- I haven't heard from any x86, ia64 or ACPI maintainers yet. OTOH, it's nice to know we've already found and fixed two sets of arm64 ACPI tables that are in error by using these patches, even with the flaws :).
On 10/09/15 21:43, Al Stone wrote: > On 09/10/2015 10:20 AM, Sudeep Holla wrote: >> [...] >> >> From the code inspection, I can see we have 3 users of acpi_parse_entries not >> just MADT but also PCC and NUMA/SRAT >> >> Something like this solves this issue: >> - if (bad_madt_entry(table_header, entry)) >> + if (!strncmp(id, ACPI_SIG_MADT, 4) && >> + bad_madt_entry(table_header, entry) >> >> >> Or am I still missing something ? > > Nope, I missed it. Your fix above will solve the problem; I misunderstood > how acpi_parse_entries() was being used -- somehow I had it in my head that > only MADT was in use, and just not seeing that it's being used for several > other subtable traversals also. Sorry about that, Sudeep. My mistake. > No worries. > I'll add this fix for a v4, but I'll wait for a few days to see if I get any > additional comments -- I haven't heard from any x86, ia64 or ACPI maintainers Makes sense. > yet. OTOH, it's nice to know we've already found and fixed two sets of arm64 > ACPI tables that are in error by using these patches, even with the flaws :). > Very much true indeed :) Regards, Sudeep
diff --git a/drivers/acpi/tables.c b/drivers/acpi/tables.c index 17a6fa0..d1c0efc 100644 --- a/drivers/acpi/tables.c +++ b/drivers/acpi/tables.c @@ -210,6 +210,245 @@ void acpi_table_print_madt_entry(struct acpi_subtable_header *header) } } +/* + * The Long, Sad, True Story of the MADT + * or + * What does bad_madt_entry() actually do? + * + * Once upon a time in ACPI 1.0, there was the MADT. It was a nice table, + * and it had two subtables all of its own. But, it was also a pretty + * busy table, too, so over time the MADT gathered up other nice little + * subtables. By the time ACPI 6.0 came around, the MADT had 16 of the + * little guys. + * + * Now, the MADT kept a little counter around for the subtables. In fact, + * it kept two counters: one was the revision level, which was supposed to + * change when new subtables came to be, or as the ones already around grew + * up. The second counter was a type number, because the MADT needed a unique + * type for each subtable so he could tell them apart. But, sometimes the + * MADT got so busy, he forgot to increment the revision level when he needed + * to. Fortunately, the type counter kept increasing since that's the only + * way the MADT could find each little subtable. It just wouldn't do to have + * every subtable called Number 6. + * + * In the next valley over, a castle full of wizards was watching the MADT + * and made a pact to keep their own counter. Every time the MADT found a + * new subtable, or a subtable grew up, the wizards promised they would + * increment their counter. Well, wizards being the forgetful sort, they + * didn't alway do that. And, since there quite a lot of them, they + * couldn't always remember who was supposed to keep track of the MADT, + * especially if dinner was coming up soon. Their counter was called the + * spec version. + * + * Every now and then, the MADT would gather up all its little subtables + * and take them in to the cobbler to get new boots. This was a very, very + * meticulous cobbler, so every time they came, he wrote down all the boot + * sizes for all of the little subtables. The cobbler would ask each subtable + * for its length, check that against his careful notes, and then go get the + * right boots. Sometimes, a little subtable would change a bit, and their + * length did not match what the cobbler had written down. If the wizards + * or the MADT had incremented their counters, the cobbler would breath a + * sigh of relief and write down the new length as the right one. But, if + * none of the counters had changed, this would make the cobbler very, very + * mad. He couldn't tell if he had the right size boots or not for the + * little subtable. He would have to *guess* and this really bugged him. + * + * Well, when the cobbler got mad like this, he would go into hiding. He + * would not make or sell any boots. He would not go out at all. Pretty + * soon, the coffee shop would have to close because the cobbler wasn't + * coming by twice a day any more. Then the grocery store would have to + * close because he wouldn't eat much. After a while, everyone would panic + * and have to move from the village and go live with all their relatives + * (usually the ones they didn't like very much). + * + * Eventually, the cobbler would work his way out of his bad mood, and + * open up his boot business again. Then, everyone else could move back + * to the village and restart their lives, too. + * + * Fortunately, we have been able to collect up all the cobbler's careful + * notes (and we wrote them down below). We'll have to keep checking these + * notes over time, too, just as the cobbler does. But, in the meantime, + * we can avoid the panic and the reboot since we can make sure that each + * subtable is doing okay. And that's what bad_madt_entry() does. + * + * + * FADT Major Version -> 1 3 4 4 5 5 6 + * FADT Minor Version -> x x x x x 1 0 + * MADT revision -> 1 1 2 3 3 3 3 + * Spec Version -> 1.0 2.0 3.0b 4.0a 5.0b 5.1a 6.0 + * Subtable Name Type Expected Length -> + * Processor Local APIC 0x0 8 8 8 8 8 8 8 + * IO APIC 0x1 12 12 12 12 12 12 12 + * Int Src Override 0x2 10 10 10 10 10 10 + * NMI Src 0x3 8 8 8 8 8 8 + * Local APIC NMI Struct 0x4 6 6 6 6 6 6 + * Local APIC Addr Ovrrd 0x5 16 12 12 12 12 12 + * IO SAPIC 0x6 20 16 16 16 16 16 + * Local SAPIC 0x7 8 >16 >16 >16 >16 >16 + * Platform Int Src 0x8 16 16 16 16 16 16 + * Proc Local x2APIC 0x9 16 16 16 16 + * Local x2APIC NMI 0xa 12 12 12 12 + * GICC CPU I/F 0xb 40 76 80 + * GICD 0xc 24 24 24 + * GICv2m MSI 0xd 24 24 + * GICR 0xe 16 16 + * GIC ITS 0xf 16 + * + * In the table, each length entry is what should be in the length + * field of the subtable, and -- in general -- it should match the + * size of the struct for the subtable. Any value that is not set + * (i.e., is zero) indicates that the subtable is not defined for + * that version of the ACPI spec. + * + */ +#define SUBTABLE_UNDEFINED 0x00 +#define SUBTABLE_VARIABLE 0xff +#define NUM_SUBTABLE_TYPES 16 + +struct acpi_madt_subtable_lengths { + unsigned short major_version; /* from revision in FADT header */ + unsigned short minor_version; /* FADT field starting with 5.1 */ + unsigned short madt_version; /* MADT revision */ + unsigned short num_types; /* types possible for this version */ + unsigned short lengths[NUM_SUBTABLE_TYPES]; + /* subtable lengths, indexed by type */ +}; + +static struct acpi_madt_subtable_lengths spec_info[] = { + { /* for ACPI 1.0 */ + .major_version = 1, + .minor_version = 0, + .madt_version = 1, + .num_types = 2, + .lengths = { 8, 12 } + }, + { /* for ACPI 2.0 */ + .major_version = 3, + .minor_version = 0, + .madt_version = 1, + .num_types = 9, + .lengths = { 8, 12, 10, 8, 6, 16, 20, 8, 16 } + }, + { /* for ACPI 3.0b */ + .major_version = 4, + .minor_version = 0, + .madt_version = 2, + .num_types = 9, + .lengths = { 8, 12, 10, 8, 6, 12, 16, SUBTABLE_VARIABLE, 16 } + }, + { /* for ACPI 4.0a */ + .major_version = 4, + .minor_version = 0, + .madt_version = 3, + .num_types = 11, + .lengths = { 8, 12, 10, 8, 6, 12, 16, SUBTABLE_VARIABLE, + 16, 16, 12 } + }, + { /* for ACPI 5.0b */ + .major_version = 5, + .minor_version = 0, + .madt_version = 3, + .num_types = 13, + .lengths = { 8, 12, 10, 8, 6, 12, 16, SUBTABLE_VARIABLE, + 16, 16, 12, 40, 24 } + }, + { /* for ACPI 5.1a */ + .major_version = 5, + .minor_version = 1, + .madt_version = 3, + .num_types = 15, + .lengths = { 8, 12, 10, 8, 6, 12, 16, SUBTABLE_VARIABLE, + 16, 16, 12, 76, 24, 24, 16 } + }, + { /* for ACPI 6.0 */ + .major_version = 6, + .minor_version = 0, + .madt_version = 3, + .num_types = 16, + .lengths = { 8, 12, 10, 8, 6, 12, 16, SUBTABLE_VARIABLE, + 16, 16, 12, 80, 24, 24, 16, 16 } + }, + { /* terminator */ + .major_version = 0, + .minor_version = 0, + .madt_version = 0, + .num_types = 0, + .lengths = { 0 } + } +}; + +int __init bad_madt_entry(struct acpi_table_header *table, + struct acpi_subtable_header *entry) +{ + struct acpi_madt_subtable_lengths *ms; + struct acpi_table_madt *madt; + unsigned short major; + unsigned short minor; + unsigned short len; + + /* simple sanity checking on MADT subtable entries */ + if (!entry || !table) + return 1; + + /* FADT minor numbers were not introduced until ACPI 5.1 */ + major = acpi_gbl_FADT.header.revision; + if (major >= 5 && acpi_gbl_FADT.header.length >= 268) + minor = acpi_gbl_FADT.minor_revision; + else + minor = 0; + + madt = (struct acpi_table_madt *)table; + ms = spec_info; + while (ms->num_types != 0) { + if (ms->major_version == major && + ms->minor_version == minor && + ms->madt_version == madt->header.revision) + break; + ms++; + } + if (!ms->num_types) { + pr_err("undefined FADT version: %d.%d\n", major, minor); + return 1; + } + + if (entry->type >= ms->num_types) { + pr_err("undefined MADT subtable type for FADT %d.%d: %d (length %d)\n", + major, minor, entry->type, entry->length); + return 1; + } + + /* verify that the table is allowed for this version of the spec */ + len = ms->lengths[entry->type]; + if (!len) { + pr_err("MADT subtable %d not defined for FADT %d.%d\n", + entry->type, major, minor); + return 1; + } + + /* verify that the length is what we expect */ + if (len == SUBTABLE_VARIABLE) { + if (entry->type == ACPI_MADT_TYPE_LOCAL_SAPIC) { + struct acpi_madt_local_sapic *lsapic = + (struct acpi_madt_local_sapic *)entry; + + if (sizeof(struct acpi_madt_local_sapic) + + strlen(lsapic->uid_string) + 1 != entry->length) { + pr_err("Variable length MADT subtable %d is wrong size: %d\n", + entry->type, entry->length); + return 1; + } + } + } else { + if (entry->length != len) { + pr_err("MADT subtable %d is wrong size: %d\n", + len, entry->type); + return 1; + } + } + + return 0; +} + int __init acpi_parse_entries(char *id, unsigned long table_size, acpi_tbl_entry_handler handler, @@ -245,6 +484,8 @@ acpi_parse_entries(char *id, unsigned long table_size, table_end) { if (entry->type == entry_id && (!max_entries || count < max_entries)) { + if (bad_madt_entry(table_header, entry)) + return -EINVAL; if (handler(entry, table_end)) return -EINVAL;
The existing BAD_MADT_ENTRY macro only checks that the size of the data structure for an MADT subtable matches the length entry in the subtable. This is, unfortunately, not reliable. Nor, as it turns out, does it have anything to do with what the length should be in any particular table. We introduce the bad_madt_entry() function that uses a data set to do some basic sanity checks on any given MADT subtable. Over time, as the spec changes, we should just be able to add entries to the data set to reflect the changes. What the data set captures is the allowed MADT subtable length for each type of subtable, for each revision of the specification. While there is a revision number in the MADT that we should be able to use to figure out the proper subtable length, it was not changed when subtables did. And, while there is a major and minor revision in the FADT that could also help, it was not always changed as the subtables changed either. So, the data set captures for each published version of the ACPI spec what the FADT revisions numbers should be, the corresponding MADT revision number, and the subtable types and lengths that were defined at that time. The sanity checks done are: -- is the length non-zero? -- is the subtable type defined/allowed for the revision of the FADT we're using? -- is the subtable type defined/allowed for the revision of the MADT we're using? -- is the length entry what it should be for this revision of the MADT and FADT? These checks are more thorough than the previous macro provided, and are now insulated from data structure size changes by ACPICA, which have been the source of other patches in the past. Now that the bad_madt_entry() function is available, we add code to also invoke it before any subtable handlers are called to use the info in the subtable. Subsequent patches will remove the use of the BAD_MADT_ENTRY macro which is now redundant as a result. Any ACPI functions that use acpi_parse_madt_entries() will always have all of the MADT subtables checked from now on. Signed-off-by: Al Stone <al.stone@linaro.org> Cc: Rafael J. Wysocki <rjw@rjwysocki.net> Cc: Len Brown <lenb@kernel.org> --- drivers/acpi/tables.c | 241 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 241 insertions(+)