| Message ID | 1605709753-7800-2-git-send-email-richard.gong@linux.intel.com (mailing list archive) |
|---|---|
| State | Superseded, archived |
| Headers | show |
| Series | Extend Intel service layer, FPGA manager and region | expand |
On Wed, Nov 18, 2020 at 08:29:09AM -0600, richard.gong@linux.intel.com wrote: > From: Richard Gong <richard.gong@intel.com> > > Add COMMAND_AUTHENTICATE_BITSTREAM command flag for new added bitstream > authentication feature. Authenticating a bistream is to make sure a signed > bitstream has the valid signatures. > > Except for the actual configuration of the device, the bitstream > authentication works the same way as FPGA configuration does. If the > authentication passes, the signed bitstream will be programmed into QSPI > flash memory and will be expected to boot without issues. > > Clean up COMMAND_RECONFIG_FLAG_PARTIAL flag by resetting it to 0, which > aligns with the firmware settings. Should this be down with the v2: ? > > Signed-off-by: Richard Gong <richard.gong@intel.com> > --- > v2: new added > --- > include/linux/firmware/intel/stratix10-svc-client.h | 11 ++++++++--- > 1 file changed, 8 insertions(+), 3 deletions(-) > > diff --git a/include/linux/firmware/intel/stratix10-svc-client.h b/include/linux/firmware/intel/stratix10-svc-client.h > index a93d859..85463c8 100644 > --- a/include/linux/firmware/intel/stratix10-svc-client.h > +++ b/include/linux/firmware/intel/stratix10-svc-client.h > @@ -51,12 +51,17 @@ > #define SVC_STATUS_NO_SUPPORT 6 > > /** > - * Flag bit for COMMAND_RECONFIG > + * Flag bit for COMMAND_RECONFIG, in bit number > * > * COMMAND_RECONFIG_FLAG_PARTIAL: > - * Set to FPGA configuration type (full or partial). > + * Set for partial FPGA configuration. > + * > + * COMMAND_AUTHENTICATE_BITSTREAM: > + * Set for bitstream authentication, which makes sure a signed bitstream > + * has valid signatures before committing it to QSPI flash memory. > */ > -#define COMMAND_RECONFIG_FLAG_PARTIAL 1 > +#define COMMAND_RECONFIG_FLAG_PARTIAL 0 > +#define COMMAND_AUTHENTICATE_BITSTREAM 1 Can you explain how this commit by itself doesn't break things? Before this change firmware expected BIT(0) to be set for partial reconfiguration, now BIT(0) suddenly means authentication? How doest his work? :) Was there a firmware version change? Did this never work before? If this is version depenedent for firmware, then this might need a different compatible string / id / some form of probing? Entirely possible that I'm missing something, but it doesn't *seem* right. > > /** > * Timeout settings for service clients: > -- > 2.7.4 > Cheers, Moritz
Hi Moritz, On 11/18/20 9:30 AM, Moritz Fischer wrote: > On Wed, Nov 18, 2020 at 08:29:09AM -0600, richard.gong@linux.intel.com wrote: >> From: Richard Gong <richard.gong@intel.com> >> >> Add COMMAND_AUTHENTICATE_BITSTREAM command flag for new added bitstream >> authentication feature. Authenticating a bistream is to make sure a signed >> bitstream has the valid signatures. >> >> Except for the actual configuration of the device, the bitstream >> authentication works the same way as FPGA configuration does. If the >> authentication passes, the signed bitstream will be programmed into QSPI >> flash memory and will be expected to boot without issues. >> >> Clean up COMMAND_RECONFIG_FLAG_PARTIAL flag by resetting it to 0, which >> aligns with the firmware settings. > > Should this be down with the v2: ? I think the commit message should describe all the changes made in the patch, is it? >> >> Signed-off-by: Richard Gong <richard.gong@intel.com> >> --- >> v2: new added >> --- >> include/linux/firmware/intel/stratix10-svc-client.h | 11 ++++++++--- >> 1 file changed, 8 insertions(+), 3 deletions(-) >> >> diff --git a/include/linux/firmware/intel/stratix10-svc-client.h b/include/linux/firmware/intel/stratix10-svc-client.h >> index a93d859..85463c8 100644 >> --- a/include/linux/firmware/intel/stratix10-svc-client.h >> +++ b/include/linux/firmware/intel/stratix10-svc-client.h >> @@ -51,12 +51,17 @@ >> #define SVC_STATUS_NO_SUPPORT 6 >> >> /** >> - * Flag bit for COMMAND_RECONFIG >> + * Flag bit for COMMAND_RECONFIG, in bit number >> * >> * COMMAND_RECONFIG_FLAG_PARTIAL: >> - * Set to FPGA configuration type (full or partial). >> + * Set for partial FPGA configuration. >> + * >> + * COMMAND_AUTHENTICATE_BITSTREAM: >> + * Set for bitstream authentication, which makes sure a signed bitstream >> + * has valid signatures before committing it to QSPI flash memory. >> */ >> -#define COMMAND_RECONFIG_FLAG_PARTIAL 1 >> +#define COMMAND_RECONFIG_FLAG_PARTIAL 0 >> +#define COMMAND_AUTHENTICATE_BITSTREAM 1 > > Can you explain how this commit by itself doesn't break things? > > Before this change firmware expected BIT(0) to be set for partial > reconfiguration, now BIT(0) suddenly means authentication? How doest his > work? :) > > Was there a firmware version change? Did this never work before? > > If this is version depenedent for firmware, then this might need a > different compatible string / id / some form of probing? > > Entirely possible that I'm missing something, but it doesn't *seem* > right. >> It did work before. Before this change, firmware only checks if the received flag value is zero. If the value is zero, it preforms full reconfiguration. Otherwise it does partial reconfiguration. To support bitstream authentication feature, firmware is updated to check the received flag value as below: 0 --- full reconfiguration BIT(0) --- partial reconfiguration BIT(1) --- bitstream authentication Therefore I have updated the command flag setting at Intel service layer driver to align with firmware. Regards, Richard >> /** >> * Timeout settings for service clients: >> -- >> 2.7.4 >> > > Cheers, > Moritz >
Richard, On Wed, Nov 18, 2020 at 12:16:09PM -0600, Richard Gong wrote: > > > -#define COMMAND_RECONFIG_FLAG_PARTIAL 1 > > > +#define COMMAND_RECONFIG_FLAG_PARTIAL 0 > > > +#define COMMAND_AUTHENTICATE_BITSTREAM 1 > > > > Can you explain how this commit by itself doesn't break things? > > > > Before this change firmware expected BIT(0) to be set for partial > > reconfiguration, now BIT(0) suddenly means authentication? How doest his > > work? :) > > > Was there a firmware version change? Did this never work before? > > > > If this is version depenedent for firmware, then this might need a > > different compatible string / id / some form of probing? > > > > Entirely possible that I'm missing something, but it doesn't *seem* > > right. > > It did work before. > > Before this change, firmware only checks if the received flag value is zero. > If the value is zero, it preforms full reconfiguration. Otherwise it does > partial reconfiguration. > > To support bitstream authentication feature, firmware is updated to check > the received flag value as below: > 0 --- full reconfiguration > BIT(0) --- partial reconfiguration > BIT(1) --- bitstream authentication So there are two different versions of firmware involved that behave differently? Old firmware: - ctype.flags = 0x0 -> Full reconfig - ctype.flags != 0 -> Partial reconfig New firmware: - ctype.flags = 0x0 -> Full reconfig - ctype.flags = 0x1 -> Partial reconfig - ctype.flags = 0x2 -> Authenticate Old software: - Send 0x0 for Full - Send 0x1 for Partial New software: - Send 0x0 for Full - Send 0x1 for Partial - Send 0x2 for Auth If I send request for authentication BIT(1) (new software) to old firmware it'd try and attempt a partial reconfiguration with the data I send? Is that safe? Is there a way for software to figure out the firmware version and do the right thing? > Therefore I have updated the command flag setting at Intel service layer > driver to align with firmware. > > Regards, > Richard > > > > /** > > > * Timeout settings for service clients: > > > -- > > > 2.7.4 > > > > > > > Cheers, > > Moritz > > Thanks, Moritz
Hi Moritz, Sorry for late reply, I was out last week. On 11/21/20 7:10 PM, Moritz Fischer wrote: > Richard, > > On Wed, Nov 18, 2020 at 12:16:09PM -0600, Richard Gong wrote: > >>>> -#define COMMAND_RECONFIG_FLAG_PARTIAL 1 >>>> +#define COMMAND_RECONFIG_FLAG_PARTIAL 0 >>>> +#define COMMAND_AUTHENTICATE_BITSTREAM 1 >>> >>> Can you explain how this commit by itself doesn't break things? >>> >>> Before this change firmware expected BIT(0) to be set for partial >>> reconfiguration, now BIT(0) suddenly means authentication? How doest his >>> work? :) >>> > Was there a firmware version change? Did this never work before? >>> >>> If this is version depenedent for firmware, then this might need a >>> different compatible string / id / some form of probing? >>> >>> Entirely possible that I'm missing something, but it doesn't *seem* >>> right. >> >> It did work before. >> >> Before this change, firmware only checks if the received flag value is zero. >> If the value is zero, it preforms full reconfiguration. Otherwise it does >> partial reconfiguration. >> >> To support bitstream authentication feature, firmware is updated to check >> the received flag value as below: >> 0 --- full reconfiguration >> BIT(0) --- partial reconfiguration >> BIT(1) --- bitstream authentication > > So there are two different versions of firmware involved that behave > differently? > > Old firmware: > - ctype.flags = 0x0 -> Full reconfig > - ctype.flags != 0 -> Partial reconfig > > New firmware: > - ctype.flags = 0x0 -> Full reconfig > - ctype.flags = 0x1 -> Partial reconfig > - ctype.flags = 0x2 -> Authenticate > > Old software: > - Send 0x0 for Full > - Send 0x1 for Partial > > New software: > - Send 0x0 for Full > - Send 0x1 for Partial > - Send 0x2 for Auth > > If I send request for authentication BIT(1) (new software) to old > firmware it'd try and attempt a partial reconfiguration with the data I > send? Is that safe? > Yes, it is possible and it is not safe. But we will inform our customers they should update to the latest firmware (SDM firmware and ATF) if they want to have authentication feature. We are migrating boot loader boot flow to the new ATF boot flow, which is SDM firmware -> SPL -> ATF -> U-boot proper -> Linux. The new authentication feature is supported only in the new ATF boot flow. ATF communicates with SDM firmware via mailbox, and SDM firmware performs the actual full/partial reconfiguration and bitstream authentication. ATF sets up EL3 environment and initializes PSCI services. The old boot flow is SDM firmware -> SPL -> U-boot proper -> Linux, which SPL/U-boot handles PSCI services and communicates with SDM firmware via mailbox. SDM firmware performs the actual full/partial reconfiguration. ATF = Arm Trust Firmware, SDM = Secure Device Manager > Is there a way for software to figure out the firmware version and do > the right thing? It is not feasible for kernel driver to get the firmware version per current designs and implementations. I don't think there is other way around this. > >> Therefore I have updated the command flag setting at Intel service layer >> driver to align with firmware. >> >> Regards, >> Richard >> >>>> /** >>>> * Timeout settings for service clients: >>>> -- >>>> 2.7.4 >>>> >>> >>> Cheers, >>> Moritz >>> > > Thanks, > Moritz > Regards, Richard
Hi Richard, On Mon, Nov 30, 2020 at 12:55:44PM -0600, Richard Gong wrote: > > Hi Moritz, > > Sorry for late reply, I was out last week. No worries, usually I'm late with replies ;-) > > On 11/21/20 7:10 PM, Moritz Fischer wrote: > > Richard, > > > > On Wed, Nov 18, 2020 at 12:16:09PM -0600, Richard Gong wrote: > > > > > > > -#define COMMAND_RECONFIG_FLAG_PARTIAL 1 > > > > > +#define COMMAND_RECONFIG_FLAG_PARTIAL 0 > > > > > +#define COMMAND_AUTHENTICATE_BITSTREAM 1 > > > > > > > > Can you explain how this commit by itself doesn't break things? > > > > > > > > Before this change firmware expected BIT(0) to be set for partial > > > > reconfiguration, now BIT(0) suddenly means authentication? How doest his > > > > work? :) > > > > > Was there a firmware version change? Did this never work before? > > > > > > > > If this is version depenedent for firmware, then this might need a > > > > different compatible string / id / some form of probing? > > > > > > > > Entirely possible that I'm missing something, but it doesn't *seem* > > > > right. > > > > > > It did work before. > > > > > > Before this change, firmware only checks if the received flag value is zero. > > > If the value is zero, it preforms full reconfiguration. Otherwise it does > > > partial reconfiguration. > > > > > > To support bitstream authentication feature, firmware is updated to check > > > the received flag value as below: > > > 0 --- full reconfiguration > > > BIT(0) --- partial reconfiguration > > > BIT(1) --- bitstream authentication > > > > So there are two different versions of firmware involved that behave > > differently? > > > > Old firmware: > > - ctype.flags = 0x0 -> Full reconfig > > - ctype.flags != 0 -> Partial reconfig > > > > New firmware: > > - ctype.flags = 0x0 -> Full reconfig > > - ctype.flags = 0x1 -> Partial reconfig > > - ctype.flags = 0x2 -> Authenticate > > > > Old software: > > - Send 0x0 for Full > > - Send 0x1 for Partial > > > > New software: > > - Send 0x0 for Full > > - Send 0x1 for Partial > > - Send 0x2 for Auth > > > > If I send request for authentication BIT(1) (new software) to old > > firmware it'd try and attempt a partial reconfiguration with the data I > > send? Is that safe? > > > > Yes, it is possible and it is not safe. But we will inform our customers > they should update to the latest firmware (SDM firmware and ATF) if they > want to have authentication feature. > > We are migrating boot loader boot flow to the new ATF boot flow, which is > SDM firmware -> SPL -> ATF -> U-boot proper -> Linux. The new authentication > feature is supported only in the new ATF boot flow. ATF communicates with > SDM firmware via mailbox, and SDM firmware performs the actual full/partial > reconfiguration and bitstream authentication. ATF sets up EL3 environment > and initializes PSCI services. Can U-Boot determine whether it's the new or old flow? Can you set a different compatible value in your device-tree, to disambiguate behaviors? > The old boot flow is SDM firmware -> SPL -> U-boot proper -> Linux, which > SPL/U-boot handles PSCI services and communicates with SDM firmware via > mailbox. SDM firmware performs the actual full/partial reconfiguration. > > ATF = Arm Trust Firmware, SDM = Secure Device Manager > > > Is there a way for software to figure out the firmware version and do > > the right thing? > > It is not feasible for kernel driver to get the firmware version per current > designs and implementations. I don't think there is other way around this. > > > > > > Therefore I have updated the command flag setting at Intel service layer > > > driver to align with firmware. > > > > > > Regards, > > > Richard > > > > > > > > /** > > > > > * Timeout settings for service clients: > > > > > -- > > > > > 2.7.4 > > > > > > > > > > > > > Cheers, > > > > Moritz > > > > > > > > Thanks, > > Moritz > > > Regards, > Richard Thanks, Moritz
Hi Richard, On Tue, Dec 01, 2020 at 01:30:16PM -0600, Richard Gong wrote: > > Can U-Boot determine whether it's the new or old flow? Can you set a > > different compatible value in your device-tree, to disambiguate > > behaviors? > > > > The boot flow is determined by defconfig during compilation, which means > each boot flow will have its own defconfig. > > SDM firmware loads SPL into OCRAM, then SPL will load the apporiate ATF or > U-boot into the DRAM according to the setting of CONFIG_SPL_ATF. If > CONFIG_SPL_ATF=y, SPL loads ATF and then jumps to ATF. ATF setups EL3 > environment and initialize the PSCI services. > > CONFIG_SPL_ATF is not set for the old boot flow. So you know at (U-Boot) build time? Can you just pass a different DT to the kernel in that case? - Moritz
Hi Moritz, On 11/30/20 10:31 PM, Moritz Fischer wrote: > Hi Richard, > > On Mon, Nov 30, 2020 at 12:55:44PM -0600, Richard Gong wrote: >> >> Hi Moritz, >> >> Sorry for late reply, I was out last week. > > No worries, usually I'm late with replies ;-) >> >> On 11/21/20 7:10 PM, Moritz Fischer wrote: >>> Richard, >>> >>> On Wed, Nov 18, 2020 at 12:16:09PM -0600, Richard Gong wrote: >>> >>>>>> -#define COMMAND_RECONFIG_FLAG_PARTIAL 1 >>>>>> +#define COMMAND_RECONFIG_FLAG_PARTIAL 0 >>>>>> +#define COMMAND_AUTHENTICATE_BITSTREAM 1 >>>>> >>>>> Can you explain how this commit by itself doesn't break things? >>>>> >>>>> Before this change firmware expected BIT(0) to be set for partial >>>>> reconfiguration, now BIT(0) suddenly means authentication? How doest his >>>>> work? :) >>>>> > Was there a firmware version change? Did this never work before? >>>>> >>>>> If this is version depenedent for firmware, then this might need a >>>>> different compatible string / id / some form of probing? >>>>> >>>>> Entirely possible that I'm missing something, but it doesn't *seem* >>>>> right. >>>> >>>> It did work before. >>>> >>>> Before this change, firmware only checks if the received flag value is zero. >>>> If the value is zero, it preforms full reconfiguration. Otherwise it does >>>> partial reconfiguration. >>>> >>>> To support bitstream authentication feature, firmware is updated to check >>>> the received flag value as below: >>>> 0 --- full reconfiguration >>>> BIT(0) --- partial reconfiguration >>>> BIT(1) --- bitstream authentication >>> >>> So there are two different versions of firmware involved that behave >>> differently? >>> >>> Old firmware: >>> - ctype.flags = 0x0 -> Full reconfig >>> - ctype.flags != 0 -> Partial reconfig >>> >>> New firmware: >>> - ctype.flags = 0x0 -> Full reconfig >>> - ctype.flags = 0x1 -> Partial reconfig >>> - ctype.flags = 0x2 -> Authenticate >>> >>> Old software: >>> - Send 0x0 for Full >>> - Send 0x1 for Partial >>> >>> New software: >>> - Send 0x0 for Full >>> - Send 0x1 for Partial >>> - Send 0x2 for Auth >>> >>> If I send request for authentication BIT(1) (new software) to old >>> firmware it'd try and attempt a partial reconfiguration with the data I >>> send? Is that safe? >>> >> >> Yes, it is possible and it is not safe. But we will inform our customers >> they should update to the latest firmware (SDM firmware and ATF) if they >> want to have authentication feature. >> >> We are migrating boot loader boot flow to the new ATF boot flow, which is >> SDM firmware -> SPL -> ATF -> U-boot proper -> Linux. The new authentication >> feature is supported only in the new ATF boot flow. ATF communicates with >> SDM firmware via mailbox, and SDM firmware performs the actual full/partial >> reconfiguration and bitstream authentication. ATF sets up EL3 environment >> and initializes PSCI services. > > Can U-Boot determine whether it's the new or old flow? Can you set a > different compatible value in your device-tree, to disambiguate > behaviors? > The boot flow is determined by defconfig during compilation, which means each boot flow will have its own defconfig. SDM firmware loads SPL into OCRAM, then SPL will load the apporiate ATF or U-boot into the DRAM according to the setting of CONFIG_SPL_ATF. If CONFIG_SPL_ATF=y, SPL loads ATF and then jumps to ATF. ATF setups EL3 environment and initialize the PSCI services. CONFIG_SPL_ATF is not set for the old boot flow. >> The old boot flow is SDM firmware -> SPL -> U-boot proper -> Linux, which >> SPL/U-boot handles PSCI services and communicates with SDM firmware via >> mailbox. SDM firmware performs the actual full/partial reconfiguration. >> >> ATF = Arm Trust Firmware, SDM = Secure Device Manager >> >>> Is there a way for software to figure out the firmware version and do >>> the right thing? >> >> It is not feasible for kernel driver to get the firmware version per current >> designs and implementations. I don't think there is other way around this. >> >>> >>>> Therefore I have updated the command flag setting at Intel service layer >>>> driver to align with firmware. >>>> >>>> Regards, >>>> Richard >>>> >>>>>> /** >>>>>> * Timeout settings for service clients: >>>>>> -- >>>>>> 2.7.4 >>>>>> >>>>> >>>>> Cheers, >>>>> Moritz >>>>> >>> >>> Thanks, >>> Moritz >>> >> Regards, >> Richard > > Thanks, > Moritz > Regards, Richard
Hi Moritz, On 12/1/20 1:19 PM, Moritz Fischer wrote: > Hi Richard, > > On Tue, Dec 01, 2020 at 01:30:16PM -0600, Richard Gong wrote: > >>> Can U-Boot determine whether it's the new or old flow? Can you set a >>> different compatible value in your device-tree, to disambiguate >>> behaviors? >>> >> >> The boot flow is determined by defconfig during compilation, which means >> each boot flow will have its own defconfig. >> >> SDM firmware loads SPL into OCRAM, then SPL will load the apporiate ATF or >> U-boot into the DRAM according to the setting of CONFIG_SPL_ATF. If >> CONFIG_SPL_ATF=y, SPL loads ATF and then jumps to ATF. ATF setups EL3 >> environment and initialize the PSCI services. >> >> CONFIG_SPL_ATF is not set for the old boot flow. > > So you know at (U-Boot) build time? Can you just pass a different DT to > the kernel in that case? > Yes, we have decided the boot flow at build time. Starting from the next release, our U-boot will use the ATF boot flow. Per my limited knowledge in U-boot, I don't think we can follow your suggestion. Or it will take a lot of efforts to achieve. I think that back compatibility is your main concern, correct? the issue does exist with the old boot flow and old firmware, whenever the customers try to use authentication. Unfortunately we can't update U-boot or firmware that has been released. The authentication feature is supported only at the ATF boot flow, updated kernel and firmware. We will have a well-documented document to inform our customers that if they want to have authentication feature, they need to upgrade the latest U-boot, kernel and firmware. We always encourage our customers to take the latest U-boot, kernel and firmware releases in their developments. > - Moritz > Regards, Richard
diff --git a/include/linux/firmware/intel/stratix10-svc-client.h b/include/linux/firmware/intel/stratix10-svc-client.h index a93d859..85463c8 100644 --- a/include/linux/firmware/intel/stratix10-svc-client.h +++ b/include/linux/firmware/intel/stratix10-svc-client.h @@ -51,12 +51,17 @@ #define SVC_STATUS_NO_SUPPORT 6 /** - * Flag bit for COMMAND_RECONFIG + * Flag bit for COMMAND_RECONFIG, in bit number * * COMMAND_RECONFIG_FLAG_PARTIAL: - * Set to FPGA configuration type (full or partial). + * Set for partial FPGA configuration. + * + * COMMAND_AUTHENTICATE_BITSTREAM: + * Set for bitstream authentication, which makes sure a signed bitstream + * has valid signatures before committing it to QSPI flash memory. */ -#define COMMAND_RECONFIG_FLAG_PARTIAL 1 +#define COMMAND_RECONFIG_FLAG_PARTIAL 0 +#define COMMAND_AUTHENTICATE_BITSTREAM 1 /** * Timeout settings for service clients: