| Message ID | 20170615224121.23949-1-tony.luck@intel.com (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Headers | show |
On Thu, Jun 15, 2017 at 03:41:21PM -0700, Luck, Tony wrote: > From: Tony Luck <tony.luck@intel.com> > > Quoting version 6.1 of the ACPI specification. Section 18.3.1 "Boot > Error Source" says: > > The Boot Error Region is a range of addressable memory OSPM can access > during initialization to determine if an unhandled error condition > occurred. System firmware must report this memory range as firmware > reserved. The format of the Boot Error Region follow that of an Error > Status Block, this is defined in Section 18.3.2.7. The format of the > error status block is described by Table 18-342. > > This clarifies some points that were obfuscated in earlier versions. > E.g. there is no longer a separate table to describe the format of the > "Boot Error Region" (which was identical to the "Error Status Block"). Hmm, ok, so do we know whether some fw writers "misunderstood" the previous version of the spec and actually added that separate table and now we need to handle the different layout? Probably not but I'm not going to be surprised someone did. At all. > Also saying "follow that of *an* Error Status Block" makes it clear that > there is just one block (which can still contain multiple "Generic Error > Data Entry structures"). > > The loop inside bert_print_all() is unnecessary (but probably harmless > as the "while (remain > sizeof(struct acpi_bert_region))" loop should > terminate after we skipped over the first entry. > > We can drop the "bert_print_all()" function and just move the four > relevant lines inline in "bert_init()". > > Also this driver clears the "block_status" status field of the error > status block in ACPI memory. I can see no justification for doing this > in the ACPI spec. It is also highly kernel-ego-centric. There are > user mode BERT harvesting tools that dig into /sys/firmware/acpi/tables/BERT > which might like to see this summary value. > > Drop the: > estatus->block_status = 0; This whole treatise of the clearing of block_status should most likely be a separate patch, though. As it is clearly a fix. Probably make it the first patch and the second one removing bert_print_all(). Something like that. Thanks.
From: Tony Luck <tony.luck@intel.com>
At Boris' suggestion I've split this into two parts for version 5.
First one stops the driver from clearing the block_status field as:
1) it is unnecessary
2) not mentioned in the spec
3) may upset user mode tools that extract information from the BERT table.
Second is the code simplification that you've seen before in ealier
versions of this patch.
Cc: Len Brown <lenb@kernel.org>
Cc: Huang Ying <ying.huang@intel.com>
Cc: Borislav Petkov <bp@suse.de>
Cc: Tomasz Nowicki <tomasz.nowicki@linaro.org>
Cc: Jonathan (Zhixiong) Zhang <zjzhang@codeaurora.org>
Cc: Tyler Baicar <tbaicar@codeaurora.org>
Cc: linux-acpi@vger.kernel.org
Tony Luck (2):
ACPI / APEI / BERT: Don't clear "block_status" in header
ACPI / APEI: Boot Error Record Table processing was needlessly
complicated
drivers/acpi/apei/bert.c | 59 ++++++++++--------------------------------------
1 file changed, 12 insertions(+), 47 deletions(-)
On Wed, Jun 21, 2017 at 10:20:40AM -0700, Luck, Tony wrote: > From: Tony Luck <tony.luck@intel.com> > > At Boris' suggestion I've split this into two parts for version 5. > First one stops the driver from clearing the block_status field as: > 1) it is unnecessary > 2) not mentioned in the spec > 3) may upset user mode tools that extract information from the BERT table. > > Second is the code simplification that you've seen before in ealier > versions of this patch. > > Cc: Len Brown <lenb@kernel.org> > Cc: Huang Ying <ying.huang@intel.com> > Cc: Borislav Petkov <bp@suse.de> > Cc: Tomasz Nowicki <tomasz.nowicki@linaro.org> > Cc: Jonathan (Zhixiong) Zhang <zjzhang@codeaurora.org> > Cc: Tyler Baicar <tbaicar@codeaurora.org> > Cc: linux-acpi@vger.kernel.org > > Tony Luck (2): > ACPI / APEI / BERT: Don't clear "block_status" in header > ACPI / APEI: Boot Error Record Table processing was needlessly > complicated Both: Reviewed-by: Borislav Petkov <bp@suse.de>
diff --git a/drivers/acpi/apei/bert.c b/drivers/acpi/apei/bert.c index 12771fcf0417..3b9b296dc84e 100644 --- a/drivers/acpi/apei/bert.c +++ b/drivers/acpi/apei/bert.c @@ -34,50 +34,6 @@ static int bert_disable; -static void __init bert_print_all(struct acpi_bert_region *region, - unsigned int region_len) -{ - struct acpi_hest_generic_status *estatus = - (struct acpi_hest_generic_status *)region; - int remain = region_len; - u32 estatus_len; - - if (!estatus->block_status) - return; - - while (remain > sizeof(struct acpi_bert_region)) { - if (cper_estatus_check(estatus)) { - pr_err(FW_BUG "Invalid error record.\n"); - return; - } - - estatus_len = cper_estatus_len(estatus); - if (remain < estatus_len) { - pr_err(FW_BUG "Truncated status block (length: %u).\n", - estatus_len); - return; - } - - pr_info_once("Error records from previous boot:\n"); - - cper_estatus_print(KERN_INFO HW_ERR, estatus); - - /* - * Because the boot error source is "one-time polled" type, - * clear Block Status of current Generic Error Status Block, - * once it's printed. - */ - estatus->block_status = 0; - - estatus = (void *)estatus + estatus_len; - /* No more error records. */ - if (!estatus->block_status) - return; - - remain -= estatus_len; - } -} - static int __init setup_bert_disable(char *str) { bert_disable = 1; @@ -89,7 +45,7 @@ __setup("bert_disable", setup_bert_disable); static int __init bert_check_table(struct acpi_table_bert *bert_tab) { if (bert_tab->header.length < sizeof(struct acpi_table_bert) || - bert_tab->region_length < sizeof(struct acpi_bert_region)) + bert_tab->region_length < sizeof(struct acpi_hest_generic_status)) return -EINVAL; return 0; @@ -98,7 +54,7 @@ static int __init bert_check_table(struct acpi_table_bert *bert_tab) static int __init bert_init(void) { struct apei_resources bert_resources; - struct acpi_bert_region *boot_error_region; + struct acpi_hest_generic_status *boot_error_region; struct acpi_table_bert *bert_tab; unsigned int region_len; acpi_status status; @@ -138,7 +94,16 @@ static int __init bert_init(void) goto out_fini; boot_error_region = ioremap_cache(bert_tab->address, region_len); if (boot_error_region) { - bert_print_all(boot_error_region, region_len); + if (boot_error_region->block_status) { + rc = cper_estatus_check(boot_error_region); + if (rc) { + pr_err(FW_BUG "Invalid error record.\n"); + iounmap(boot_error_region); + return rc; + } + pr_info("Error records from previous boot:\n"); + cper_estatus_print(KERN_INFO HW_ERR, boot_error_region); + } iounmap(boot_error_region); } else { rc = -ENOMEM;