| Message ID | 20240910044543.3812642-1-avri.altman@wdc.com |
|---|---|
| State | New |
| Headers | show |
| Series | [v2] scsi: ufs: Use pre-calculated offsets in ufshcd_init_lrb | expand |
On 9/9/24 9:45 PM, Avri Altman wrote: > Replace manual offset calculations for response_upiu and prd_table in > ufshcd_init_lrb() with pre-calculated offsets already stored in the > utp_transfer_req_desc structure. The pre-calculated offsets are set > differently in ufshcd_host_memory_configure() based on the > UFSHCD_QUIRK_PRDT_BYTE_GRAN quirk, ensuring correct alignment and > access. With which host controllers has this patch been tested? Thanks, Bart.
> On 9/9/24 9:45 PM, Avri Altman wrote: > > Replace manual offset calculations for response_upiu and prd_table in > > ufshcd_init_lrb() with pre-calculated offsets already stored in the > > utp_transfer_req_desc structure. The pre-calculated offsets are set > > differently in ufshcd_host_memory_configure() based on the > > UFSHCD_QUIRK_PRDT_BYTE_GRAN quirk, ensuring correct alignment and > > access. > > With which host controllers has this patch been tested? Qualcomm RB5 platform. I guess I'll be needing help with testing it on an exynos platforms? Thanks, Avri > > Thanks, > > Bart.
+Alim > > On 9/9/24 9:45 PM, Avri Altman wrote: > > > Replace manual offset calculations for response_upiu and prd_table > > > in > > > ufshcd_init_lrb() with pre-calculated offsets already stored in the > > > utp_transfer_req_desc structure. The pre-calculated offsets are set > > > differently in ufshcd_host_memory_configure() based on the > > > UFSHCD_QUIRK_PRDT_BYTE_GRAN quirk, ensuring correct alignment and > > > access. > > > > With which host controllers has this patch been tested? > Qualcomm RB5 platform. > I guess I'll be needing help with testing it on an exynos platforms? Would appreciate any help testing it on Exynos. Thanks, Avri > > Thanks, > Avri > > > > > Thanks, > > > > Bart.
On 9/9/24 9:45 PM, Avri Altman wrote: > Replace manual offset calculations for response_upiu and prd_table in > ufshcd_init_lrb() with pre-calculated offsets already stored in the > utp_transfer_req_desc structure. The pre-calculated offsets are set > differently in ufshcd_host_memory_configure() based on the > UFSHCD_QUIRK_PRDT_BYTE_GRAN quirk, ensuring correct alignment and > access. > > Fixes: 26f968d7de82 ("scsi: ufs: Introduce UFSHCD_QUIRK_PRDT_BYTE_GRAN quirk") > Cc: stable@vger.kernel.org > Signed-off-by: Avri Altman <avri.altman@wdc.com> > > --- > Changes in v2: > - add Fixes: and Cc: stable tags > - fix kernel test robot warning about type mismatch by using le16_to_cpu > --- > drivers/ufs/core/ufshcd.c | 5 ++--- > 1 file changed, 2 insertions(+), 3 deletions(-) > > diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c > index 8ea5a82503a9..85251c176ef7 100644 > --- a/drivers/ufs/core/ufshcd.c > +++ b/drivers/ufs/core/ufshcd.c > @@ -2919,9 +2919,8 @@ static void ufshcd_init_lrb(struct ufs_hba *hba, struct ufshcd_lrb *lrb, int i) > struct utp_transfer_req_desc *utrdlp = hba->utrdl_base_addr; > dma_addr_t cmd_desc_element_addr = hba->ucdl_dma_addr + > i * ufshcd_get_ucd_size(hba); > - u16 response_offset = offsetof(struct utp_transfer_cmd_desc, > - response_upiu); > - u16 prdt_offset = offsetof(struct utp_transfer_cmd_desc, prd_table); > + u16 response_offset = le16_to_cpu(utrdlp[i].response_upiu_offset); > + u16 prdt_offset = le16_to_cpu(utrdlp[i].prd_table_offset); > > lrb->utr_descriptor_ptr = utrdlp + i; > lrb->utrd_dma_addr = hba->utrdl_dma_addr + Please always Cc the author of the original patch when posting a candidate fix. Alim, since the upstream kernel code seems to work fine with Exynos UFS host controllers, is the description of UFSHCD_QUIRK_PRDT_BYTE_GRAN perhaps wrong? I'm referring to the following description: /* * This quirk needs to be enabled if the host controller regards * resolution of the values of PRDTO and PRDTL in UTRD as byte. */ UFSHCD_QUIRK_PRDT_BYTE_GRAN = 1 << 9, Thanks, Bart.
+Daejun Thanks, Avri > On 9/9/24 9:45 PM, Avri Altman wrote: > > Replace manual offset calculations for response_upiu and prd_table in > > ufshcd_init_lrb() with pre-calculated offsets already stored in the > > utp_transfer_req_desc structure. The pre-calculated offsets are set > > differently in ufshcd_host_memory_configure() based on the > > UFSHCD_QUIRK_PRDT_BYTE_GRAN quirk, ensuring correct alignment and > > access. > > > > Fixes: 26f968d7de82 ("scsi: ufs: Introduce > UFSHCD_QUIRK_PRDT_BYTE_GRAN > > quirk") > > Cc: stable@vger.kernel.org > > Signed-off-by: Avri Altman <avri.altman@wdc.com> > > > > --- > > Changes in v2: > > - add Fixes: and Cc: stable tags > > - fix kernel test robot warning about type mismatch by using > > le16_to_cpu > > --- > > drivers/ufs/core/ufshcd.c | 5 ++--- > > 1 file changed, 2 insertions(+), 3 deletions(-) > > > > diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c > > index 8ea5a82503a9..85251c176ef7 100644 > > --- a/drivers/ufs/core/ufshcd.c > > +++ b/drivers/ufs/core/ufshcd.c > > @@ -2919,9 +2919,8 @@ static void ufshcd_init_lrb(struct ufs_hba *hba, > struct ufshcd_lrb *lrb, int i) > > struct utp_transfer_req_desc *utrdlp = hba->utrdl_base_addr; > > dma_addr_t cmd_desc_element_addr = hba->ucdl_dma_addr + > > i * ufshcd_get_ucd_size(hba); > > - u16 response_offset = offsetof(struct utp_transfer_cmd_desc, > > - response_upiu); > > - u16 prdt_offset = offsetof(struct utp_transfer_cmd_desc, prd_table); > > + u16 response_offset = le16_to_cpu(utrdlp[i].response_upiu_offset); > > + u16 prdt_offset = le16_to_cpu(utrdlp[i].prd_table_offset); > > > > lrb->utr_descriptor_ptr = utrdlp + i; > > lrb->utrd_dma_addr = hba->utrdl_dma_addr + > > Please always Cc the author of the original patch when posting a candidate fix. > > Alim, since the upstream kernel code seems to work fine with Exynos UFS host > controllers, is the description of UFSHCD_QUIRK_PRDT_BYTE_GRAN perhaps > wrong? I'm referring to the following description: > > /* > * This quirk needs to be enabled if the host controller regards > * resolution of the values of PRDTO and PRDTL in UTRD as byte. > */ > UFSHCD_QUIRK_PRDT_BYTE_GRAN = 1 << 9, > > Thanks, > > Bart.
> +Daejun > > Thanks, > Avri Bart - How do you want to proceed with this fix? Thanks, Avri > > On 9/9/24 9:45 PM, Avri Altman wrote: > > > Replace manual offset calculations for response_upiu and prd_table > > > in > > > ufshcd_init_lrb() with pre-calculated offsets already stored in the > > > utp_transfer_req_desc structure. The pre-calculated offsets are set > > > differently in ufshcd_host_memory_configure() based on the > > > UFSHCD_QUIRK_PRDT_BYTE_GRAN quirk, ensuring correct alignment and > > > access. > > > > > > Fixes: 26f968d7de82 ("scsi: ufs: Introduce > > UFSHCD_QUIRK_PRDT_BYTE_GRAN > > > quirk") > > > Cc: stable@vger.kernel.org > > > Signed-off-by: Avri Altman <avri.altman@wdc.com> > > > > > > --- > > > Changes in v2: > > > - add Fixes: and Cc: stable tags > > > - fix kernel test robot warning about type mismatch by using > > > le16_to_cpu > > > --- > > > drivers/ufs/core/ufshcd.c | 5 ++--- > > > 1 file changed, 2 insertions(+), 3 deletions(-) > > > > > > diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c > > > index 8ea5a82503a9..85251c176ef7 100644 > > > --- a/drivers/ufs/core/ufshcd.c > > > +++ b/drivers/ufs/core/ufshcd.c > > > @@ -2919,9 +2919,8 @@ static void ufshcd_init_lrb(struct ufs_hba > > > *hba, > > struct ufshcd_lrb *lrb, int i) > > > struct utp_transfer_req_desc *utrdlp = hba->utrdl_base_addr; > > > dma_addr_t cmd_desc_element_addr = hba->ucdl_dma_addr + > > > i * ufshcd_get_ucd_size(hba); > > > - u16 response_offset = offsetof(struct utp_transfer_cmd_desc, > > > - response_upiu); > > > - u16 prdt_offset = offsetof(struct utp_transfer_cmd_desc, prd_table); > > > + u16 response_offset = le16_to_cpu(utrdlp[i].response_upiu_offset); > > > + u16 prdt_offset = le16_to_cpu(utrdlp[i].prd_table_offset); > > > > > > lrb->utr_descriptor_ptr = utrdlp + i; > > > lrb->utrd_dma_addr = hba->utrdl_dma_addr + > > > > Please always Cc the author of the original patch when posting a candidate > fix. > > > > Alim, since the upstream kernel code seems to work fine with Exynos > > UFS host controllers, is the description of > > UFSHCD_QUIRK_PRDT_BYTE_GRAN perhaps wrong? I'm referring to the > following description: > > > > /* > > * This quirk needs to be enabled if the host controller regards > > * resolution of the values of PRDTO and PRDTL in UTRD as byte. > > */ > > UFSHCD_QUIRK_PRDT_BYTE_GRAN = 1 << 9, > > > > Thanks, > > > > Bart.
On 10/1/24 12:19 AM, Avri Altman wrote:
> Bart - How do you want to proceed with this fix?
Since this patch affects the Exynos UFSHCI, I think we need feedback
from the maintainers of the Exynos UFSHCI code.
Personally I'm hesitant to proceed with this patch. As you may know the
android-mainline kernel tracks the upstream kernel very closely.
Currently that kernel branch is only one week behind Linus' master
branch. The android-mainline kernel boots fine on Pixel 6 devices. These
devices have an Exynos UFS host controller. Since this patch modifies
the behavior for the Exynos UFS host controller there is a potential
that it will break the support for that controller.
Thanks,
Bart.
On 10/1/24 12:19 AM, Avri Altman wrote:
> Bart - How do you want to proceed with this fix?
As one can see here [1], if this patch is applied on top of the
android-mainline kernel branch (close to v6.12-rc1) then all presubmit
tests pass. This includes booting the kernel on a Pixel 6 device that
has an Exynos UFS host controller. So I'm fine with this patch.
[1] https://android-review.googlesource.com/c/kernel/common/+/3291741
Thanks,
Bart.
On Tue, 10 Sep 2024 07:45:43 +0300, Avri Altman wrote: > Replace manual offset calculations for response_upiu and prd_table in > ufshcd_init_lrb() with pre-calculated offsets already stored in the > utp_transfer_req_desc structure. The pre-calculated offsets are set > differently in ufshcd_host_memory_configure() based on the > UFSHCD_QUIRK_PRDT_BYTE_GRAN quirk, ensuring correct alignment and > access. > > [...] Applied to 6.12/scsi-fixes, thanks! [1/1] scsi: ufs: Use pre-calculated offsets in ufshcd_init_lrb https://git.kernel.org/mkp/scsi/c/d5130c5a0932
diff --git a/drivers/ufs/core/ufshcd.c b/drivers/ufs/core/ufshcd.c index 8ea5a82503a9..85251c176ef7 100644 --- a/drivers/ufs/core/ufshcd.c +++ b/drivers/ufs/core/ufshcd.c @@ -2919,9 +2919,8 @@ static void ufshcd_init_lrb(struct ufs_hba *hba, struct ufshcd_lrb *lrb, int i) struct utp_transfer_req_desc *utrdlp = hba->utrdl_base_addr; dma_addr_t cmd_desc_element_addr = hba->ucdl_dma_addr + i * ufshcd_get_ucd_size(hba); - u16 response_offset = offsetof(struct utp_transfer_cmd_desc, - response_upiu); - u16 prdt_offset = offsetof(struct utp_transfer_cmd_desc, prd_table); + u16 response_offset = le16_to_cpu(utrdlp[i].response_upiu_offset); + u16 prdt_offset = le16_to_cpu(utrdlp[i].prd_table_offset); lrb->utr_descriptor_ptr = utrdlp + i; lrb->utrd_dma_addr = hba->utrdl_dma_addr +
Replace manual offset calculations for response_upiu and prd_table in ufshcd_init_lrb() with pre-calculated offsets already stored in the utp_transfer_req_desc structure. The pre-calculated offsets are set differently in ufshcd_host_memory_configure() based on the UFSHCD_QUIRK_PRDT_BYTE_GRAN quirk, ensuring correct alignment and access. Fixes: 26f968d7de82 ("scsi: ufs: Introduce UFSHCD_QUIRK_PRDT_BYTE_GRAN quirk") Cc: stable@vger.kernel.org Signed-off-by: Avri Altman <avri.altman@wdc.com> --- Changes in v2: - add Fixes: and Cc: stable tags - fix kernel test robot warning about type mismatch by using le16_to_cpu --- drivers/ufs/core/ufshcd.c | 5 ++--- 1 file changed, 2 insertions(+), 3 deletions(-)