| Message ID | 20161119182228.GA22775@obsidianresearch.com |
|---|---|
| State | New |
| Headers | show |
On Sat, Nov 19, 2016 at 11:22:28AM -0700, Jason Gunthorpe wrote: > On Fri, Nov 18, 2016 at 07:52:49AM -0800, Jarkko Sakkinen wrote: > > On Thu, Nov 17, 2016 at 07:30:04PM -0500, Stefan Berger wrote: > > > tpm_bios_log_setup() may return -ENODEV in case no log was > > > found. In this case we do not need to fail the device. > > > > > > Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com> > > > drivers/char/tpm/tpm-chip.c | 2 +- > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c > > > index 3f27753..2d6530b 100644 > > > +++ b/drivers/char/tpm/tpm-chip.c > > > @@ -346,7 +346,7 @@ int tpm_chip_register(struct tpm_chip *chip) > > > tpm_sysfs_add_device(chip); > > > > > > rc = tpm_bios_log_setup(chip); > > > - if (rc == -ENODEV) > > > + if (rc != -ENODEV) > > > return rc; > > > > > > tpm_add_ppi(chip); > > > > CC to linux-security-module > > > > LGTM > > > > Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com> > > Erm, what about rc == 0? And all the other problems? Sorry my bad. I was not thinking clearly. This whole situation looks like a mess. I gave a lot of thought on this during my plane trips. > Here, use this (untested) should take care of everything on this > topic.. > > The two things I haven't seen explained are the sysfs unregister crash > and the acpi iounmap crash :/ Yup. The reason I'm not weighting that yet so much is that I do not know the environment. > > From 8768bcb8cd2a5a17cc4d811a9298b20c3a2c0884 Mon Sep 17 00:00:00 2001 > From: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > Date: Sat, 19 Nov 2016 11:18:28 -0700 > Subject: [PATCH] tpm: Fix handling of missing event log > > The event log is an optional firmware feature, if the firmware > does not support it then the securityfs files should not be created > and no other notification given. > > - Uniformly return -ENODEV from the tpm_bios_log_setup cone if > no event log is detected. > - Check in ACPI if this node was discovered via ACPI. > - Improve the check in OF to make sure there is a parent and to > fail detection if the two log properties are not declared > - Pass through all other error codes instead of filtering just some > > Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > --- > drivers/char/tpm/tpm-chip.c | 2 +- > drivers/char/tpm/tpm_acpi.c | 8 +++++++- > drivers/char/tpm/tpm_eventlog.c | 26 +++++++++++++------------- > drivers/char/tpm/tpm_of.c | 11 +++++------ > 4 files changed, 26 insertions(+), 21 deletions(-) > > diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c > index 3f27753d96aab5..7a4869151d3b90 100644 > --- a/drivers/char/tpm/tpm-chip.c > +++ b/drivers/char/tpm/tpm-chip.c > @@ -346,7 +346,7 @@ int tpm_chip_register(struct tpm_chip *chip) > tpm_sysfs_add_device(chip); > > rc = tpm_bios_log_setup(chip); > - if (rc == -ENODEV) > + if (rc != 0 && rc != -ENODEV) > return rc; > > tpm_add_ppi(chip); > diff --git a/drivers/char/tpm/tpm_acpi.c b/drivers/char/tpm/tpm_acpi.c > index 0cb43ef5f79a6e..99366bf64f3359 100644 > --- a/drivers/char/tpm/tpm_acpi.c > +++ b/drivers/char/tpm/tpm_acpi.c > @@ -56,12 +56,18 @@ int tpm_read_log_acpi(struct tpm_chip *chip) > > log = &chip->log; > > + /* Unfortuntely ACPI does not associate the event log with a specific > + * TPM, like PPI. Thus all ACPI TPMs will read the same log. > + */ > + if (!chip->acpi_dev_handle) > + return -ENODEV; > + > /* Find TCPA entry in RSDT (ACPI_LOGICAL_ADDRESSING) */ > status = acpi_get_table(ACPI_SIG_TCPA, 1, > (struct acpi_table_header **)&buff); > > if (ACPI_FAILURE(status)) > - return -EIO; > + return -ENODEV; > > switch(buff->platform_class) { > case BIOS_SERVER: > diff --git a/drivers/char/tpm/tpm_eventlog.c b/drivers/char/tpm/tpm_eventlog.c > index fb603a74cbd29e..2a15b866ac257a 100644 > --- a/drivers/char/tpm/tpm_eventlog.c > +++ b/drivers/char/tpm/tpm_eventlog.c > @@ -377,14 +377,21 @@ static int tpm_read_log(struct tpm_chip *chip) > } > > rc = tpm_read_log_acpi(chip); > - if ((rc == 0) || (rc == -ENOMEM)) > + if (rc != -ENODEV) > return rc; > > - rc = tpm_read_log_of(chip); > - > - return rc; > + return tpm_read_log_of(chip); > } > > +/* > + * tpm_bios_log_setup() - Read the event log from the firmware > + * @chip: TPM chip to use. > + * > + * If an event log is found then the securityfs files are setup to > + * export it to userspace, otherwise nothing is done. > + * > + * Returns -ENODEV if the firmware has no event log. > + */ > int tpm_bios_log_setup(struct tpm_chip *chip) > { > const char *name = dev_name(&chip->dev); > @@ -395,15 +402,8 @@ int tpm_bios_log_setup(struct tpm_chip *chip) > return 0; > > rc = tpm_read_log(chip); > - /* > - * read_log failure means event log is not supported except for ENOMEM. > - */ > - if (rc < 0) { > - if (rc == -ENOMEM) > - return -ENODEV; > - else > - return rc; > - } WTF. I really have to be much more focused when I looked this. That is more than wrong... Too much multitasking last couple of weeks. That's my excuse... I can consider putting the patch set to the next release but I really would want yet another version with change log what fixes were done and why. /Jarkko ------------------------------------------------------------------------------
On Sun, Nov 20, 2016 at 09:46:25AM +0000, Jarkko Sakkinen wrote: > On Sat, Nov 19, 2016 at 11:22:28AM -0700, Jason Gunthorpe wrote: > > On Fri, Nov 18, 2016 at 07:52:49AM -0800, Jarkko Sakkinen wrote: > > > On Thu, Nov 17, 2016 at 07:30:04PM -0500, Stefan Berger wrote: > > > > tpm_bios_log_setup() may return -ENODEV in case no log was > > > > found. In this case we do not need to fail the device. > > > > > > > > Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com> > > > > drivers/char/tpm/tpm-chip.c | 2 +- > > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > > > diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c > > > > index 3f27753..2d6530b 100644 > > > > +++ b/drivers/char/tpm/tpm-chip.c > > > > @@ -346,7 +346,7 @@ int tpm_chip_register(struct tpm_chip *chip) > > > > tpm_sysfs_add_device(chip); > > > > > > > > rc = tpm_bios_log_setup(chip); > > > > - if (rc == -ENODEV) > > > > + if (rc != -ENODEV) > > > > return rc; > > > > > > > > tpm_add_ppi(chip); > > > > > > CC to linux-security-module > > > > > > LGTM > > > > > > Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com> > > > > Erm, what about rc == 0? And all the other problems? > > Sorry my bad. I was not thinking clearly. > > This whole situation looks like a mess. I gave a lot of thought on this > during my plane trips. > > > Here, use this (untested) should take care of everything on this > > topic.. > > > > The two things I haven't seen explained are the sysfs unregister crash > > and the acpi iounmap crash :/ > > Yup. The reason I'm not weighting that yet so much is that I do not know > the environment. > > > > > From 8768bcb8cd2a5a17cc4d811a9298b20c3a2c0884 Mon Sep 17 00:00:00 2001 > > From: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > > Date: Sat, 19 Nov 2016 11:18:28 -0700 > > Subject: [PATCH] tpm: Fix handling of missing event log > > > > The event log is an optional firmware feature, if the firmware > > does not support it then the securityfs files should not be created > > and no other notification given. > > > > - Uniformly return -ENODEV from the tpm_bios_log_setup cone if > > no event log is detected. > > - Check in ACPI if this node was discovered via ACPI. > > - Improve the check in OF to make sure there is a parent and to > > fail detection if the two log properties are not declared > > - Pass through all other error codes instead of filtering just some > > > > Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > > --- > > drivers/char/tpm/tpm-chip.c | 2 +- > > drivers/char/tpm/tpm_acpi.c | 8 +++++++- > > drivers/char/tpm/tpm_eventlog.c | 26 +++++++++++++------------- > > drivers/char/tpm/tpm_of.c | 11 +++++------ > > 4 files changed, 26 insertions(+), 21 deletions(-) > > > > diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c > > index 3f27753d96aab5..7a4869151d3b90 100644 > > --- a/drivers/char/tpm/tpm-chip.c > > +++ b/drivers/char/tpm/tpm-chip.c > > @@ -346,7 +346,7 @@ int tpm_chip_register(struct tpm_chip *chip) > > tpm_sysfs_add_device(chip); > > > > rc = tpm_bios_log_setup(chip); > > - if (rc == -ENODEV) > > + if (rc != 0 && rc != -ENODEV) > > return rc; > > > > tpm_add_ppi(chip); > > diff --git a/drivers/char/tpm/tpm_acpi.c b/drivers/char/tpm/tpm_acpi.c > > index 0cb43ef5f79a6e..99366bf64f3359 100644 > > --- a/drivers/char/tpm/tpm_acpi.c > > +++ b/drivers/char/tpm/tpm_acpi.c > > @@ -56,12 +56,18 @@ int tpm_read_log_acpi(struct tpm_chip *chip) > > > > log = &chip->log; > > > > + /* Unfortuntely ACPI does not associate the event log with a specific > > + * TPM, like PPI. Thus all ACPI TPMs will read the same log. > > + */ > > + if (!chip->acpi_dev_handle) > > + return -ENODEV; > > + > > /* Find TCPA entry in RSDT (ACPI_LOGICAL_ADDRESSING) */ > > status = acpi_get_table(ACPI_SIG_TCPA, 1, > > (struct acpi_table_header **)&buff); > > > > if (ACPI_FAILURE(status)) > > - return -EIO; > > + return -ENODEV; > > > > switch(buff->platform_class) { > > case BIOS_SERVER: > > diff --git a/drivers/char/tpm/tpm_eventlog.c b/drivers/char/tpm/tpm_eventlog.c > > index fb603a74cbd29e..2a15b866ac257a 100644 > > --- a/drivers/char/tpm/tpm_eventlog.c > > +++ b/drivers/char/tpm/tpm_eventlog.c > > @@ -377,14 +377,21 @@ static int tpm_read_log(struct tpm_chip *chip) > > } > > > > rc = tpm_read_log_acpi(chip); > > - if ((rc == 0) || (rc == -ENOMEM)) > > + if (rc != -ENODEV) > > return rc; > > > > - rc = tpm_read_log_of(chip); > > - > > - return rc; > > + return tpm_read_log_of(chip); > > } > > > > +/* > > + * tpm_bios_log_setup() - Read the event log from the firmware > > + * @chip: TPM chip to use. > > + * > > + * If an event log is found then the securityfs files are setup to > > + * export it to userspace, otherwise nothing is done. > > + * > > + * Returns -ENODEV if the firmware has no event log. > > + */ > > int tpm_bios_log_setup(struct tpm_chip *chip) > > { > > const char *name = dev_name(&chip->dev); > > @@ -395,15 +402,8 @@ int tpm_bios_log_setup(struct tpm_chip *chip) > > return 0; > > > > rc = tpm_read_log(chip); > > - /* > > - * read_log failure means event log is not supported except for ENOMEM. > > - */ > > - if (rc < 0) { > > - if (rc == -ENOMEM) > > - return -ENODEV; > > - else > > - return rc; > > - } > > WTF. I really have to be much more focused when I looked this. That > is more than wrong... Too much multitasking last couple of weeks. That's > my excuse... > > I can consider putting the patch set to the next release but I really > would want yet another version with change log what fixes were done and > why. I mean one more version of the patch set. Otherwise, I probably just postpone the whole patch set to 4.10. /Jarkko ------------------------------------------------------------------------------
I was able to apply the patch after manually editing the mbox file. Not nice. It'd be better to send these things with git-send-email to simplify the mgmt (like getting them into patchwork for starters) especially in a situation that is bit chaotic to begin with :-/ It's there. I'm not yet sure what to do with these changes. I'll probably move my pull request towards late next week. Lets see if things stabilize. Too early to decide. In any case I won't squash the fixes to existing series even if I include them to the pull request. I'll add new tested-by's only to the fixes. /Jarkko On Sat, Nov 19, 2016 at 11:22:28AM -0700, Jason Gunthorpe wrote: > On Fri, Nov 18, 2016 at 07:52:49AM -0800, Jarkko Sakkinen wrote: > > On Thu, Nov 17, 2016 at 07:30:04PM -0500, Stefan Berger wrote: > > > tpm_bios_log_setup() may return -ENODEV in case no log was > > > found. In this case we do not need to fail the device. > > > > > > Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com> > > > drivers/char/tpm/tpm-chip.c | 2 +- > > > 1 file changed, 1 insertion(+), 1 deletion(-) > > > > > > diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c > > > index 3f27753..2d6530b 100644 > > > +++ b/drivers/char/tpm/tpm-chip.c > > > @@ -346,7 +346,7 @@ int tpm_chip_register(struct tpm_chip *chip) > > > tpm_sysfs_add_device(chip); > > > > > > rc = tpm_bios_log_setup(chip); > > > - if (rc == -ENODEV) > > > + if (rc != -ENODEV) > > > return rc; > > > > > > tpm_add_ppi(chip); > > > > CC to linux-security-module > > > > LGTM > > > > Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com> > > Erm, what about rc == 0? And all the other problems? > > Here, use this (untested) should take care of everything on this > topic.. > > The two things I haven't seen explained are the sysfs unregister crash > and the acpi iounmap crash :/ > > From 8768bcb8cd2a5a17cc4d811a9298b20c3a2c0884 Mon Sep 17 00:00:00 2001 > From: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > Date: Sat, 19 Nov 2016 11:18:28 -0700 > Subject: [PATCH] tpm: Fix handling of missing event log > > The event log is an optional firmware feature, if the firmware > does not support it then the securityfs files should not be created > and no other notification given. > > - Uniformly return -ENODEV from the tpm_bios_log_setup cone if > no event log is detected. > - Check in ACPI if this node was discovered via ACPI. > - Improve the check in OF to make sure there is a parent and to > fail detection if the two log properties are not declared > - Pass through all other error codes instead of filtering just some > > Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > --- > drivers/char/tpm/tpm-chip.c | 2 +- > drivers/char/tpm/tpm_acpi.c | 8 +++++++- > drivers/char/tpm/tpm_eventlog.c | 26 +++++++++++++------------- > drivers/char/tpm/tpm_of.c | 11 +++++------ > 4 files changed, 26 insertions(+), 21 deletions(-) > > diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c > index 3f27753d96aab5..7a4869151d3b90 100644 > --- a/drivers/char/tpm/tpm-chip.c > +++ b/drivers/char/tpm/tpm-chip.c > @@ -346,7 +346,7 @@ int tpm_chip_register(struct tpm_chip *chip) > tpm_sysfs_add_device(chip); > > rc = tpm_bios_log_setup(chip); > - if (rc == -ENODEV) > + if (rc != 0 && rc != -ENODEV) > return rc; > > tpm_add_ppi(chip); > diff --git a/drivers/char/tpm/tpm_acpi.c b/drivers/char/tpm/tpm_acpi.c > index 0cb43ef5f79a6e..99366bf64f3359 100644 > --- a/drivers/char/tpm/tpm_acpi.c > +++ b/drivers/char/tpm/tpm_acpi.c > @@ -56,12 +56,18 @@ int tpm_read_log_acpi(struct tpm_chip *chip) > > log = &chip->log; > > + /* Unfortuntely ACPI does not associate the event log with a specific > + * TPM, like PPI. Thus all ACPI TPMs will read the same log. > + */ > + if (!chip->acpi_dev_handle) > + return -ENODEV; > + > /* Find TCPA entry in RSDT (ACPI_LOGICAL_ADDRESSING) */ > status = acpi_get_table(ACPI_SIG_TCPA, 1, > (struct acpi_table_header **)&buff); > > if (ACPI_FAILURE(status)) > - return -EIO; > + return -ENODEV; > > switch(buff->platform_class) { > case BIOS_SERVER: > diff --git a/drivers/char/tpm/tpm_eventlog.c b/drivers/char/tpm/tpm_eventlog.c > index fb603a74cbd29e..2a15b866ac257a 100644 > --- a/drivers/char/tpm/tpm_eventlog.c > +++ b/drivers/char/tpm/tpm_eventlog.c > @@ -377,14 +377,21 @@ static int tpm_read_log(struct tpm_chip *chip) > } > > rc = tpm_read_log_acpi(chip); > - if ((rc == 0) || (rc == -ENOMEM)) > + if (rc != -ENODEV) > return rc; > > - rc = tpm_read_log_of(chip); > - > - return rc; > + return tpm_read_log_of(chip); > } > > +/* > + * tpm_bios_log_setup() - Read the event log from the firmware > + * @chip: TPM chip to use. > + * > + * If an event log is found then the securityfs files are setup to > + * export it to userspace, otherwise nothing is done. > + * > + * Returns -ENODEV if the firmware has no event log. > + */ > int tpm_bios_log_setup(struct tpm_chip *chip) > { > const char *name = dev_name(&chip->dev); > @@ -395,15 +402,8 @@ int tpm_bios_log_setup(struct tpm_chip *chip) > return 0; > > rc = tpm_read_log(chip); > - /* > - * read_log failure means event log is not supported except for ENOMEM. > - */ > - if (rc < 0) { > - if (rc == -ENOMEM) > - return -ENODEV; > - else > - return rc; > - } > + if (rc) > + return rc; > > cnt = 0; > chip->bios_dir[cnt] = securityfs_create_dir(name, NULL); > diff --git a/drivers/char/tpm/tpm_of.c b/drivers/char/tpm/tpm_of.c > index 36df9df4c472b9..7dee42d7b5e05c 100644 > --- a/drivers/char/tpm/tpm_of.c > +++ b/drivers/char/tpm/tpm_of.c > @@ -29,13 +29,16 @@ int tpm_read_log_of(struct tpm_chip *chip) > struct tpm_bios_log *log; > > log = &chip->log; > - if (chip->dev.parent->of_node) > + if (chip->dev.parent && chip->dev.parent->of_node) > np = chip->dev.parent->of_node; > else > return -ENODEV; > > sizep = of_get_property(np, "linux,sml-size", NULL); > - if (sizep == NULL) > + basep = of_get_property(np, "linux,sml-base", NULL); > + if (sizep == NULL && basep == NULL) > + return -ENODEV; > + if (sizep == NULL || basep == NULL) > return -EIO; > > if (*sizep == 0) { > @@ -43,10 +46,6 @@ int tpm_read_log_of(struct tpm_chip *chip) > return -EIO; > } > > - basep = of_get_property(np, "linux,sml-base", NULL); > - if (basep == NULL) > - return -EIO; > - > log->bios_event_log = kmalloc(*sizep, GFP_KERNEL); > if (!log->bios_event_log) > return -ENOMEM; > -- > 2.7.4 > ------------------------------------------------------------------------------
On 11/19/2016 11:52 PM, Jason Gunthorpe wrote: > On Fri, Nov 18, 2016 at 07:52:49AM -0800, Jarkko Sakkinen wrote: >> On Thu, Nov 17, 2016 at 07:30:04PM -0500, Stefan Berger wrote: >>> tpm_bios_log_setup() may return -ENODEV in case no log was >>> found. In this case we do not need to fail the device. >>> >>> Signed-off-by: Stefan Berger <stefanb@linux.vnet.ibm.com> >>> drivers/char/tpm/tpm-chip.c | 2 +- >>> 1 file changed, 1 insertion(+), 1 deletion(-) >>> >>> diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c >>> index 3f27753..2d6530b 100644 >>> +++ b/drivers/char/tpm/tpm-chip.c >>> @@ -346,7 +346,7 @@ int tpm_chip_register(struct tpm_chip *chip) >>> tpm_sysfs_add_device(chip); >>> >>> rc = tpm_bios_log_setup(chip); >>> - if (rc == -ENODEV) >>> + if (rc != -ENODEV) >>> return rc; >>> >>> tpm_add_ppi(chip); >> >> CC to linux-security-module >> >> LGTM >> >> Reviewed-by: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com> > > Erm, what about rc == 0? And all the other problems? > > Here, use this (untested) should take care of everything on this > topic.. > > The two things I haven't seen explained are the sysfs unregister crash > and the acpi iounmap crash :/ > > From 8768bcb8cd2a5a17cc4d811a9298b20c3a2c0884 Mon Sep 17 00:00:00 2001 > From: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > Date: Sat, 19 Nov 2016 11:18:28 -0700 > Subject: [PATCH] tpm: Fix handling of missing event log > > The event log is an optional firmware feature, if the firmware > does not support it then the securityfs files should not be created > and no other notification given. > > - Uniformly return -ENODEV from the tpm_bios_log_setup cone if > no event log is detected. > - Check in ACPI if this node was discovered via ACPI. > - Improve the check in OF to make sure there is a parent and to > fail detection if the two log properties are not declared > - Pass through all other error codes instead of filtering just some > > Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > --- > drivers/char/tpm/tpm-chip.c | 2 +- > drivers/char/tpm/tpm_acpi.c | 8 +++++++- > drivers/char/tpm/tpm_eventlog.c | 26 +++++++++++++------------- > drivers/char/tpm/tpm_of.c | 11 +++++------ > 4 files changed, 26 insertions(+), 21 deletions(-) > > diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c > index 3f27753d96aab5..7a4869151d3b90 100644 > --- a/drivers/char/tpm/tpm-chip.c > +++ b/drivers/char/tpm/tpm-chip.c > @@ -346,7 +346,7 @@ int tpm_chip_register(struct tpm_chip *chip) > tpm_sysfs_add_device(chip); > > rc = tpm_bios_log_setup(chip); > - if (rc == -ENODEV) > + if (rc != 0 && rc != -ENODEV) > return rc; This will return in case of -EFAULT as well, where the check is that log is already initialized. Do we want to fail the probe here as well ? -EFAULT is returned from tpm_read_log() as below: tpm_read_log() has if (chip->log.bios_event_log != NULL) { dev_dbg(&chip->dev, "%s: ERROR - event log already initialized\n", __func__); return -EFAULT; } > > tpm_add_ppi(chip); > diff --git a/drivers/char/tpm/tpm_acpi.c b/drivers/char/tpm/tpm_acpi.c > index 0cb43ef5f79a6e..99366bf64f3359 100644 > --- a/drivers/char/tpm/tpm_acpi.c > +++ b/drivers/char/tpm/tpm_acpi.c > @@ -56,12 +56,18 @@ int tpm_read_log_acpi(struct tpm_chip *chip) > > log = &chip->log; > > + /* Unfortuntely ACPI does not associate the event log with a specific > + * TPM, like PPI. Thus all ACPI TPMs will read the same log. > + */ > + if (!chip->acpi_dev_handle) > + return -ENODEV; > + > /* Find TCPA entry in RSDT (ACPI_LOGICAL_ADDRESSING) */ > status = acpi_get_table(ACPI_SIG_TCPA, 1, > (struct acpi_table_header **)&buff); > > if (ACPI_FAILURE(status)) > - return -EIO; > + return -ENODEV; > > switch(buff->platform_class) { > case BIOS_SERVER: > diff --git a/drivers/char/tpm/tpm_eventlog.c b/drivers/char/tpm/tpm_eventlog.c > index fb603a74cbd29e..2a15b866ac257a 100644 > --- a/drivers/char/tpm/tpm_eventlog.c > +++ b/drivers/char/tpm/tpm_eventlog.c > @@ -377,14 +377,21 @@ static int tpm_read_log(struct tpm_chip *chip) > } > > rc = tpm_read_log_acpi(chip); This is to understand.. It can return -ENOMEM error here, contd below... > - if ((rc == 0) || (rc == -ENOMEM)) > + if (rc != -ENODEV) > return rc; > > - rc = tpm_read_log_of(chip); > - > - return rc; > + return tpm_read_log_of(chip); So, in ACPI if -ENOMEM error is returned, it will continue to tpm_read_log_of(chip), which will return -ENODEV. So, -ENOMEM error is now masked with -ENODEV error. Next, in tpm_chip_register(), this will be considered as -ENODEV > + if (rc != 0 && rc != -ENODEV) > return rc; and so will not fail the probe. It doesn't create securityfs files, but also does not fail the probe for memory error. Is it the expected behavior ? > } > > +/* > + * tpm_bios_log_setup() - Read the event log from the firmware > + * @chip: TPM chip to use. > + * > + * If an event log is found then the securityfs files are setup to > + * export it to userspace, otherwise nothing is done. > + * > + * Returns -ENODEV if the firmware has no event log. > + */ > int tpm_bios_log_setup(struct tpm_chip *chip) > { > const char *name = dev_name(&chip->dev); > @@ -395,15 +402,8 @@ int tpm_bios_log_setup(struct tpm_chip *chip) > return 0; > > rc = tpm_read_log(chip); > - /* > - * read_log failure means event log is not supported except for ENOMEM. > - */ > - if (rc < 0) { > - if (rc == -ENOMEM) > - return -ENODEV; > - else > - return rc; > - } > + if (rc) > + return rc; > > cnt = 0; > chip->bios_dir[cnt] = securityfs_create_dir(name, NULL); > diff --git a/drivers/char/tpm/tpm_of.c b/drivers/char/tpm/tpm_of.c > index 36df9df4c472b9..7dee42d7b5e05c 100644 > --- a/drivers/char/tpm/tpm_of.c > +++ b/drivers/char/tpm/tpm_of.c > @@ -29,13 +29,16 @@ int tpm_read_log_of(struct tpm_chip *chip) > struct tpm_bios_log *log; > > log = &chip->log; > - if (chip->dev.parent->of_node) > + if (chip->dev.parent && chip->dev.parent->of_node) > np = chip->dev.parent->of_node; > else > return -ENODEV; > > sizep = of_get_property(np, "linux,sml-size", NULL); > - if (sizep == NULL) > + basep = of_get_property(np, "linux,sml-base", NULL); > + if (sizep == NULL && basep == NULL) > + return -ENODEV; > + if (sizep == NULL || basep == NULL) > return -EIO; To confirm my understanding, For -ENODEV, it means that both properties are not supported, so event log is not supported. For -EIO , it means that event log is supported but there is some failure in getting one of them, so should fail the probe. Is my understanding right ? Thanks & Regards, - Nayna > > if (*sizep == 0) { > @@ -43,10 +46,6 @@ int tpm_read_log_of(struct tpm_chip *chip) > return -EIO; > } > > - basep = of_get_property(np, "linux,sml-base", NULL); > - if (basep == NULL) > - return -EIO; > - > log->bios_event_log = kmalloc(*sizep, GFP_KERNEL); > if (!log->bios_event_log) > return -ENOMEM; > ------------------------------------------------------------------------------
On Mon, Nov 21, 2016 at 12:06:20AM +0530, Nayna wrote: > > rc = tpm_bios_log_setup(chip); > >- if (rc == -ENODEV) > >+ if (rc != 0 && rc != -ENODEV) > > return rc; > > This will return in case of -EFAULT as well, where the check is that log is > already initialized. Do we want to fail the probe here as well ? > > -EFAULT is returned from tpm_read_log() as below: That is fine, we should never read the log twice. > >index fb603a74cbd29e..2a15b866ac257a 100644 > >+++ b/drivers/char/tpm/tpm_eventlog.c > >@@ -377,14 +377,21 @@ static int tpm_read_log(struct tpm_chip *chip) > > } > > > > rc = tpm_read_log_acpi(chip); > > This is to understand.. > It can return -ENOMEM error here, contd below... > > >- if ((rc == 0) || (rc == -ENOMEM)) > >+ if (rc != -ENODEV) > > return rc; > > > >- rc = tpm_read_log_of(chip); > >- > >- return rc; > >+ return tpm_read_log_of(chip); > > So, in ACPI if -ENOMEM error is returned, it will continue to > tpm_read_log_of(chip), which will return -ENODEV. So, -ENOMEM error is now > masked with -ENODEV error. No, if acpi is -ENOMEM then 'if (rc != -ENODEV)' is true and it returns -ENOMEM. > > sizep = of_get_property(np, "linux,sml-size", NULL); > >- if (sizep == NULL) > >+ basep = of_get_property(np, "linux,sml-base", NULL); > >+ if (sizep == NULL && basep == NULL) > >+ return -ENODEV; > >+ if (sizep == NULL || basep == NULL) > > return -EIO; > > To confirm my understanding, > > For -ENODEV, it means that both properties are not supported, so event log > is not supported. Yes > For -EIO , it means that event log is supported but there is some failure in > getting one of them, so should fail the probe. > Is my understanding right ? Yes Jason ------------------------------------------------------------------------------
On Sun, Nov 20, 2016 at 12:14:51PM +0000, Jarkko Sakkinen wrote: > I was able to apply the patch after manually editing the mbox file. > Not nice. It'd be better to send these things with git-send-email > to simplify the mgmt (like getting them into patchwork for starters) > especially in a situation that is bit chaotic to begin with :-/ It is in patchwork just fine: https://patchwork.kernel.org/patch/9438323/ git am is unfortunately much more limited than patchwork, but you can apply right out of patchwork. You forgot to add your signed-off-by though. And this one fixes the sysfs problem https://patchwork.kernel.org/patch/9438331/ Here is a commit message: tpm: Do not call tpm_chip_unregister if register has not succeeded This is no longer allowed, vtpm was relying on the check inside the tpm core which was removed. Fixes an oops during error unwind. Fixes: a8ed92219a04 ("tpm: Get rid of TPM_CHIP_FLAG_REGISTERED") Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> Tested-by: Stefan Berger <stefanb@linux.vnet.ibm.com> That just leaves the ACPI thing.. Jason ------------------------------------------------------------------------------
On Mon, Nov 21, 2016 at 10:15:30AM -0700, Jason Gunthorpe wrote: > On Mon, Nov 21, 2016 at 12:06:20AM +0530, Nayna wrote: > > > rc = tpm_bios_log_setup(chip); > > >- if (rc == -ENODEV) > > >+ if (rc != 0 && rc != -ENODEV) > > > return rc; > > > > This will return in case of -EFAULT as well, where the check is that log is > > already initialized. Do we want to fail the probe here as well ? > > > > -EFAULT is returned from tpm_read_log() as below: > > That is fine, we should never read the log twice. I just went through tpm_bios_log_setup(), tpm_read_log_of() and tpm_read_log_acpi(). The error handling is sound now but the condition should be in tpm_bios_log_setup(). Not in tpm_chip_register(). It is applied but if you don't mind I would like make a small commit that moves the condition to that function. I think tpm_chip_register() should only care when something went wrong. /Jarkko ------------------------------------------------------------------------------
On Mon, Nov 21, 2016 at 10:55:20AM -0700, Jason Gunthorpe wrote: > On Sun, Nov 20, 2016 at 12:14:51PM +0000, Jarkko Sakkinen wrote: > > I was able to apply the patch after manually editing the mbox file. > > Not nice. It'd be better to send these things with git-send-email > > to simplify the mgmt (like getting them into patchwork for starters) > > especially in a situation that is bit chaotic to begin with :-/ > > It is in patchwork just fine: > > https://patchwork.kernel.org/patch/9438323/ > > git am is unfortunately much more limited than patchwork, but you can > apply right out of patchwork. > > You forgot to add your signed-off-by though. It's not forgotten. I'm yet to test it properly. I just quickly wanted to make it available... /Jarkko > > And this one fixes the sysfs problem > > https://patchwork.kernel.org/patch/9438331/ > > Here is a commit message: > > tpm: Do not call tpm_chip_unregister if register has not succeeded > > This is no longer allowed, vtpm was relying on the check inside the > tpm core which was removed. Fixes an oops during error unwind. > > Fixes: a8ed92219a04 ("tpm: Get rid of TPM_CHIP_FLAG_REGISTERED") > Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > Tested-by: Stefan Berger <stefanb@linux.vnet.ibm.com> > > That just leaves the ACPI thing.. > > Jason ------------------------------------------------------------------------------
On Mon, Nov 21, 2016 at 10:55:20AM -0700, Jason Gunthorpe wrote: > On Sun, Nov 20, 2016 at 12:14:51PM +0000, Jarkko Sakkinen wrote: > > I was able to apply the patch after manually editing the mbox file. > > Not nice. It'd be better to send these things with git-send-email > > to simplify the mgmt (like getting them into patchwork for starters) > > especially in a situation that is bit chaotic to begin with :-/ > > It is in patchwork just fine: > > https://patchwork.kernel.org/patch/9438323/ > > git am is unfortunately much more limited than patchwork, but you can > apply right out of patchwork. > > You forgot to add your signed-off-by though. > > And this one fixes the sysfs problem > > https://patchwork.kernel.org/patch/9438331/ > > Here is a commit message: > > tpm: Do not call tpm_chip_unregister if register has not succeeded > > This is no longer allowed, vtpm was relying on the check inside the > tpm core which was removed. Fixes an oops during error unwind. > > Fixes: a8ed92219a04 ("tpm: Get rid of TPM_CHIP_FLAG_REGISTERED") > Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > Tested-by: Stefan Berger <stefanb@linux.vnet.ibm.com> > > That just leaves the ACPI thing.. I like the original commit message more. It documents the change. /Jarkko > > Jason ------------------------------------------------------------------------------
On Mon, Nov 21, 2016 at 10:32:44PM +0200, Jarkko Sakkinen wrote: > On Mon, Nov 21, 2016 at 10:55:20AM -0700, Jason Gunthorpe wrote: > > On Sun, Nov 20, 2016 at 12:14:51PM +0000, Jarkko Sakkinen wrote: > > > I was able to apply the patch after manually editing the mbox file. > > > Not nice. It'd be better to send these things with git-send-email > > > to simplify the mgmt (like getting them into patchwork for starters) > > > especially in a situation that is bit chaotic to begin with :-/ > > > > It is in patchwork just fine: > > > > https://patchwork.kernel.org/patch/9438323/ > > > > git am is unfortunately much more limited than patchwork, but you can > > apply right out of patchwork. > > > > You forgot to add your signed-off-by though. > > > > And this one fixes the sysfs problem > > > > https://patchwork.kernel.org/patch/9438331/ > > > > Here is a commit message: > > > > tpm: Do not call tpm_chip_unregister if register has not succeeded > > > > This is no longer allowed, vtpm was relying on the check inside the > > tpm core which was removed. Fixes an oops during error unwind. > > > > Fixes: a8ed92219a04 ("tpm: Get rid of TPM_CHIP_FLAG_REGISTERED") > > Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > > Tested-by: Stefan Berger <stefanb@linux.vnet.ibm.com> > > > > That just leaves the ACPI thing.. > > I like the original commit message more. It documents the change. I added to the commits reviewed-by's and signed-off-by's as I agree on the correctness of the changes. I'll do more formal testing before adding testd-by... /Jarkko ------------------------------------------------------------------------------
On Mon, Nov 21, 2016 at 10:32:44PM +0200, Jarkko Sakkinen wrote: > > And this one fixes the sysfs problem > > > > https://patchwork.kernel.org/patch/9438331/ > > > > Here is a commit message: > > > > tpm: Do not call tpm_chip_unregister if register has not succeeded > > > > This is no longer allowed, vtpm was relying on the check inside the > > tpm core which was removed. Fixes an oops during error unwind. > > > > Fixes: a8ed92219a04 ("tpm: Get rid of TPM_CHIP_FLAG_REGISTERED") > > Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > > Tested-by: Stefan Berger <stefanb@linux.vnet.ibm.com> > > > > That just leaves the ACPI thing.. > > I like the original commit message more. It documents the change. ?? This is for https://patchwork.kernel.org/patch/9438331/ which didn't have a message yet. Jason ------------------------------------------------------------------------------
On Mon, Nov 21, 2016 at 01:37:08PM -0700, Jason Gunthorpe wrote: > On Mon, Nov 21, 2016 at 10:32:44PM +0200, Jarkko Sakkinen wrote: > > > > And this one fixes the sysfs problem > > > > > > https://patchwork.kernel.org/patch/9438331/ > > > > > > Here is a commit message: > > > > > > tpm: Do not call tpm_chip_unregister if register has not succeeded > > > > > > This is no longer allowed, vtpm was relying on the check inside the > > > tpm core which was removed. Fixes an oops during error unwind. > > > > > > Fixes: a8ed92219a04 ("tpm: Get rid of TPM_CHIP_FLAG_REGISTERED") > > > Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > > > Tested-by: Stefan Berger <stefanb@linux.vnet.ibm.com> > > > > > > That just leaves the ACPI thing.. > > > > I like the original commit message more. It documents the change. > > ?? > > This is for https://patchwork.kernel.org/patch/9438331/ which didn't > have a message yet. This was the commit message when I did git am: tpm: vtpm_proxy: Do not access host's event log On Thu, Nov 17, 2016 at 06:15:20PM -0500, Stefan Berger wrote: > >>Further, I had the impression that the error unwinding following -ENODEV has > >>an issue related to sysfs. > >I don't follow this comment.. > > I have encountered this error here, which gets masked when applying the > previously shown patch. If tpm_chip_register fails vtpm must not call tpm_chip_unregister: > [ 58.271017] [<ffffffff8155bd32>] dpm_sysfs_remove+0x22/0x60 > [ 58.271017] [<ffffffff8154e438>] device_del+0x58/0x280 > [ 58.271017] [<ffffffffa024c020>] tpm_chip_unregister+0x40/0xb0 [tpm] > [ 58.271017] [<ffffffffa0292360>] vtpm_proxy_fops_release+0x40/0x60 [tpm_vtpm_proxy] So, this is a vtpm thing I missed for 'tpm: Get rid of TPM_CHIP_FLAG_REGISTERED' Does this do the trick? /Jarkko ------------------------------------------------------------------------------
On Mon, Nov 21, 2016 at 10:46:10PM +0200, Jarkko Sakkinen wrote: > On Mon, Nov 21, 2016 at 01:37:08PM -0700, Jason Gunthorpe wrote: > > On Mon, Nov 21, 2016 at 10:32:44PM +0200, Jarkko Sakkinen wrote: > > > > > > And this one fixes the sysfs problem > > > > > > > > https://patchwork.kernel.org/patch/9438331/ > > > > > > > > Here is a commit message: > > > > > > > > tpm: Do not call tpm_chip_unregister if register has not succeeded > > > > > > > > This is no longer allowed, vtpm was relying on the check inside the > > > > tpm core which was removed. Fixes an oops during error unwind. > > > > > > > > Fixes: a8ed92219a04 ("tpm: Get rid of TPM_CHIP_FLAG_REGISTERED") > > > > Signed-off-by: Jason Gunthorpe <jgunthorpe@obsidianresearch.com> > > > > Tested-by: Stefan Berger <stefanb@linux.vnet.ibm.com> > > > > > > > > That just leaves the ACPI thing.. > > > > > > I like the original commit message more. It documents the change. > > > > ?? > > > > This is for https://patchwork.kernel.org/patch/9438331/ which didn't > > have a message yet. > > This was the commit message when I did git am: Well, that is just a conversation from email, not a commit message, you need to edit it. The above I gave you is much better. Jason ------------------------------------------------------------------------------
On 11/21/2016 10:45 PM, Jason Gunthorpe wrote: > On Mon, Nov 21, 2016 at 12:06:20AM +0530, Nayna wrote: >>> rc = tpm_bios_log_setup(chip); >>> - if (rc == -ENODEV) >>> + if (rc != 0 && rc != -ENODEV) >>> return rc; >> >> This will return in case of -EFAULT as well, where the check is that log is >> already initialized. Do we want to fail the probe here as well ? >> >> -EFAULT is returned from tpm_read_log() as below: > > That is fine, we should never read the log twice. > >>> index fb603a74cbd29e..2a15b866ac257a 100644 >>> +++ b/drivers/char/tpm/tpm_eventlog.c >>> @@ -377,14 +377,21 @@ static int tpm_read_log(struct tpm_chip *chip) >>> } >>> >>> rc = tpm_read_log_acpi(chip); >> >> This is to understand.. >> It can return -ENOMEM error here, contd below... >> >>> - if ((rc == 0) || (rc == -ENOMEM)) >>> + if (rc != -ENODEV) >>> return rc; >>> >>> - rc = tpm_read_log_of(chip); >>> - >>> - return rc; >>> + return tpm_read_log_of(chip); >> >> So, in ACPI if -ENOMEM error is returned, it will continue to >> tpm_read_log_of(chip), which will return -ENODEV. So, -ENOMEM error is now >> masked with -ENODEV error. > > No, if acpi is -ENOMEM then 'if (rc != -ENODEV)' is true and it > returns -ENOMEM. Yeah, I missed it. Sorry. > >>> sizep = of_get_property(np, "linux,sml-size", NULL); >>> - if (sizep == NULL) >>> + basep = of_get_property(np, "linux,sml-base", NULL); >>> + if (sizep == NULL && basep == NULL) >>> + return -ENODEV; >>> + if (sizep == NULL || basep == NULL) >>> return -EIO; >> >> To confirm my understanding, >> >> For -ENODEV, it means that both properties are not supported, so event log >> is not supported. > > Yes > >> For -EIO , it means that event log is supported but there is some failure in >> getting one of them, so should fail the probe. >> Is my understanding right ? > > Yes Thanks for explaining. Looks good now. Thanks & Regards, - Nayna > > Jason > ------------------------------------------------------------------------------
On Mon, Nov 21, 2016 at 10:29:15PM +0200, Jarkko Sakkinen wrote: > I just went through tpm_bios_log_setup(), tpm_read_log_of() and > tpm_read_log_acpi(). The error handling is sound now but the condition > should be in tpm_bios_log_setup(). Not in tpm_chip_register(). > > It is applied but if you don't mind I would like make a small commit > that moves the condition to that function. I'm not particularly concerned where the check lives.. Jason ------------------------------------------------------------------------------
On Tue, Nov 22, 2016 at 09:37:20AM -0700, Jason Gunthorpe wrote: > On Mon, Nov 21, 2016 at 10:29:15PM +0200, Jarkko Sakkinen wrote: > > > I just went through tpm_bios_log_setup(), tpm_read_log_of() and > > tpm_read_log_acpi(). The error handling is sound now but the condition > > should be in tpm_bios_log_setup(). Not in tpm_chip_register(). > > > > It is applied but if you don't mind I would like make a small commit > > that moves the condition to that function. > > I'm not particularly concerned where the check lives.. It's just a minor glitch but still an obvious inconsistency. Maybe I'll ignore for th moment as everything is well tested. There is still two commits in the series that are untested and unreviewed. /Jarkko ------------------------------------------------------------------------------
diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c index 3f27753d96aab5..7a4869151d3b90 100644 --- a/drivers/char/tpm/tpm-chip.c +++ b/drivers/char/tpm/tpm-chip.c @@ -346,7 +346,7 @@ int tpm_chip_register(struct tpm_chip *chip) tpm_sysfs_add_device(chip); rc = tpm_bios_log_setup(chip); - if (rc == -ENODEV) + if (rc != 0 && rc != -ENODEV) return rc; tpm_add_ppi(chip); diff --git a/drivers/char/tpm/tpm_acpi.c b/drivers/char/tpm/tpm_acpi.c index 0cb43ef5f79a6e..99366bf64f3359 100644 --- a/drivers/char/tpm/tpm_acpi.c +++ b/drivers/char/tpm/tpm_acpi.c @@ -56,12 +56,18 @@ int tpm_read_log_acpi(struct tpm_chip *chip) log = &chip->log; + /* Unfortuntely ACPI does not associate the event log with a specific + * TPM, like PPI. Thus all ACPI TPMs will read the same log. + */ + if (!chip->acpi_dev_handle) + return -ENODEV; + /* Find TCPA entry in RSDT (ACPI_LOGICAL_ADDRESSING) */ status = acpi_get_table(ACPI_SIG_TCPA, 1, (struct acpi_table_header **)&buff); if (ACPI_FAILURE(status)) - return -EIO; + return -ENODEV; switch(buff->platform_class) { case BIOS_SERVER: diff --git a/drivers/char/tpm/tpm_eventlog.c b/drivers/char/tpm/tpm_eventlog.c index fb603a74cbd29e..2a15b866ac257a 100644 --- a/drivers/char/tpm/tpm_eventlog.c +++ b/drivers/char/tpm/tpm_eventlog.c @@ -377,14 +377,21 @@ static int tpm_read_log(struct tpm_chip *chip) } rc = tpm_read_log_acpi(chip); - if ((rc == 0) || (rc == -ENOMEM)) + if (rc != -ENODEV) return rc; - rc = tpm_read_log_of(chip); - - return rc; + return tpm_read_log_of(chip); } +/* + * tpm_bios_log_setup() - Read the event log from the firmware + * @chip: TPM chip to use. + * + * If an event log is found then the securityfs files are setup to + * export it to userspace, otherwise nothing is done. + * + * Returns -ENODEV if the firmware has no event log. + */ int tpm_bios_log_setup(struct tpm_chip *chip) { const char *name = dev_name(&chip->dev); @@ -395,15 +402,8 @@ int tpm_bios_log_setup(struct tpm_chip *chip) return 0; rc = tpm_read_log(chip); - /* - * read_log failure means event log is not supported except for ENOMEM. - */ - if (rc < 0) { - if (rc == -ENOMEM) - return -ENODEV; - else - return rc; - } + if (rc) + return rc; cnt = 0; chip->bios_dir[cnt] = securityfs_create_dir(name, NULL); diff --git a/drivers/char/tpm/tpm_of.c b/drivers/char/tpm/tpm_of.c index 36df9df4c472b9..7dee42d7b5e05c 100644 --- a/drivers/char/tpm/tpm_of.c +++ b/drivers/char/tpm/tpm_of.c @@ -29,13 +29,16 @@ int tpm_read_log_of(struct tpm_chip *chip) struct tpm_bios_log *log; log = &chip->log; - if (chip->dev.parent->of_node) + if (chip->dev.parent && chip->dev.parent->of_node) np = chip->dev.parent->of_node; else return -ENODEV; sizep = of_get_property(np, "linux,sml-size", NULL); - if (sizep == NULL) + basep = of_get_property(np, "linux,sml-base", NULL); + if (sizep == NULL && basep == NULL) + return -ENODEV; + if (sizep == NULL || basep == NULL) return -EIO; if (*sizep == 0) { @@ -43,10 +46,6 @@ int tpm_read_log_of(struct tpm_chip *chip) return -EIO; } - basep = of_get_property(np, "linux,sml-base", NULL); - if (basep == NULL) - return -EIO; - log->bios_event_log = kmalloc(*sizep, GFP_KERNEL); if (!log->bios_event_log) return -ENOMEM;