| Message ID | 20220428140145.870527-1-a.fatoum@pengutronix.de (mailing list archive) |
|---|---|
| Headers | show |
| Series | KEYS: trusted: Introduce support for NXP CAAM-based trusted keys | expand |
Gave this a go on iMX8QXP with Linux 5.17.5 and I can't quite get it working. I get -ENODEV from add_key() via keyctl. When I traced it in dmesg I couldn't get an as clear picture as I would like but CAAM (and thus possibly JRs?) initialzing after trusted_key. dmesg snips: [ 1.296772] trusted_key: Job Ring Device allocation for transform failed ... [ 1.799768] caam 31400000.crypto: device ID = 0x0a16040000000100 (Era 9) [ 1.807142] caam 31400000.crypto: job rings = 2, qi = 0 [ 1.822667] caam algorithms registered in /proc/crypto [ 1.830541] caam 31400000.crypto: caam pkc algorithms registered in /proc/crypto [ 1.841807] caam 31400000.crypto: registering rng-caam I didn't quite have the time to get a better trace than that. Best regards // John Ernberg
Hello John, On 05.05.22 16:58, John Ernberg wrote: > Gave this a go on iMX8QXP with Linux 5.17.5 and I can't quite get it working. > > I get -ENODEV from add_key() via keyctl. When I traced it in dmesg I couldn't > get an as clear picture as I would like but CAAM (and thus possibly JRs?) > initialzing after trusted_key. > > dmesg snips: > [ 1.296772] trusted_key: Job Ring Device allocation for transform failed > ... > [ 1.799768] caam 31400000.crypto: device ID = 0x0a16040000000100 (Era 9) > [ 1.807142] caam 31400000.crypto: job rings = 2, qi = 0 > [ 1.822667] caam algorithms registered in /proc/crypto > [ 1.830541] caam 31400000.crypto: caam pkc algorithms registered in /proc/crypto > [ 1.841807] caam 31400000.crypto: registering rng-caam > > I didn't quite have the time to get a better trace than that. I don't see a crypto@31400000 node upstream. Where can I see your device tree? Initcall ordering does the right thing, but if CAAM device probe is deferred beyond late_initcall, then it won't help. This is a general limitation with trusted keys at the moment. Anything that's not there by the time of the late_initcall won't be tried again. You can work around it by having trusted keys as a module. We might be able to do something with fw_devlinks in the future and a look into your device tree would help here, but I think that should be separate from this patch series. Please let me know if the module build improves the situation for you. Cheers, Ahmad > > Best regards // John Ernberg
Hi Ahmad, > > > > dmesg snips: > > [ 1.296772] trusted_key: Job Ring Device allocation for transform failed > > ... > > [ 1.799768] caam 31400000.crypto: device ID = 0x0a16040000000100 (Era 9) > > [ 1.807142] caam 31400000.crypto: job rings = 2, qi = 0 > > [ 1.822667] caam algorithms registered in /proc/crypto > > [ 1.830541] caam 31400000.crypto: caam pkc algorithms registered in /proc/crypto > > [ 1.841807] caam 31400000.crypto: registering rng-caam > > > > I didn't quite have the time to get a better trace than that. > > I don't see a crypto@31400000 node upstream. Where can I see your device tree? Apologies for forgetting to mention that, I took it from the NXP tree while removing the SM and SECO bits [1]. I also had to rebase some of their patches onto 5.17 for the CAAM to probe, as the SCU makes some register pages unavailable. > Initcall ordering does the right thing, but if CAAM device probe is deferred beyond > late_initcall, then it won't help. > > This is a general limitation with trusted keys at the moment. Anything that's > not there by the time of the late_initcall won't be tried again. You can work > around it by having trusted keys as a module. We might be able to do something > with fw_devlinks in the future and a look into your device tree would help here, > but I think that should be separate from this patch series. Thank for you the explanation, it makes sense, and I agree that such work would be a different patch set. > > Please let me know if the module build improves the situation for you. > After I changed trusted keys to a module I got it working. Which is good enough for me as QXP CAAM support is not upstream yet. Feel free to add my tested by if you need to make another spin. Tested-by: John Ernberg <john.ernberg@actia.se> # iMX8QXP I didn't test v9 as I would have to patch around the new patch due to the SCU. Best regards // John Ernberg [1]: https://source.codeaurora.org/external/imx/linux-imx/tree/arch/arm64/boot/dts/freescale/imx8-ss-security.dtsi?h=lf-5.10.y
Hello John, On 07.05.22 23:30, John Ernberg wrote: > Hi Ahmad, > >>> >>> dmesg snips: >>> [ 1.296772] trusted_key: Job Ring Device allocation for transform failed >>> ... >>> [ 1.799768] caam 31400000.crypto: device ID = 0x0a16040000000100 (Era 9) >>> [ 1.807142] caam 31400000.crypto: job rings = 2, qi = 0 >>> [ 1.822667] caam algorithms registered in /proc/crypto >>> [ 1.830541] caam 31400000.crypto: caam pkc algorithms registered in /proc/crypto >>> [ 1.841807] caam 31400000.crypto: registering rng-caam >>> >>> I didn't quite have the time to get a better trace than that. >> >> I don't see a crypto@31400000 node upstream. Where can I see your device tree? > > Apologies for forgetting to mention that, I took it from the NXP tree > while removing the SM and SECO bits [1]. > I also had to rebase some of their patches onto 5.17 for the CAAM to > probe, as the SCU makes some register pages unavailable. If the CAAM has a dependency on some SCU-provided resource, this would explain why the driver probes it that late. >> Initcall ordering does the right thing, but if CAAM device probe is deferred beyond >> late_initcall, then it won't help. >> >> This is a general limitation with trusted keys at the moment. Anything that's >> not there by the time of the late_initcall won't be tried again. You can work >> around it by having trusted keys as a module. We might be able to do something >> with fw_devlinks in the future and a look into your device tree would help here, >> but I think that should be separate from this patch series. > > Thank for you the explanation, it makes sense, and I agree that such work > would be a different patch set. > >> >> Please let me know if the module build improves the situation for you. >> > > After I changed trusted keys to a module I got it working. Which is good > enough for me as QXP CAAM support is not upstream yet. Great! > Feel free to add my tested by if you need to make another spin. > Tested-by: John Ernberg <john.ernberg@actia.se> # iMX8QXP > > I didn't test v9 as I would have to patch around the new patch due to > the SCU. Thanks for the test. I will add it to v10 except for - "crypto: caam - determine whether CAAM supports blob encap/decap", which was new in v9 - "doc: trusted-encrypted: describe new CAAM trust source", "MAINTAINERS: add KEYS-TRUSTED-CAAM" as runtime test isn't affected by these. Cheers, Ahmad > > Best regards // John Ernberg > > [1]: https://source.codeaurora.org/external/imx/linux-imx/tree/arch/arm64/boot/dts/freescale/imx8-ss-security.dtsi?h=lf-5.10.y