[v15,0/17] PCI: imx6: refine codes and add the error propagation

Message

Richard Zhu July 14, 2022, 7:30 a.m. UTC

This series patches refine pci-imx6 driver and do the following main changes.
- Encapsulate the clock enable into one standalone function
- Add the error propagation from host_init and resume
- Turn off regulator when the system is in suspend mode
- Let the probe successfully when link never comes up
- Do not hide the phy driver callbacks in core reset and clk_enable.
- Keep symmetric as much as possible between suspend and resume callbacks.
BTW, this series are verified on i.MX8MM EVK board when one NVME is used.

Main changes from v14 to v15 refer to Lucas' comments and Bjorn's suggestions:
Only the five patches listed below have changes.
10/17
- Remove regulator_disable() from imx6_pcie_shutdown() and update the commit
  log accordingly refer to Lucas' comments.
11/17
- Move the regulator enable before the PHY init and core reset assert to
  avoid introducing more failure cleanup paths refer to Lucas' comments.
14/17 has the codes conflictions.
- Rebase the 14/17 patch because of the codes conflictions introduced by
  previous 11/17 new changes.
16/17
- Add the missing the IMX8MQ case and drop the default case in
  imx6_pcie_ltssm_disable() refer to Lucas' comments.
17/17
- Rebase the codes and resolve the codes conflictions introduced by
  previous 11/17 new changes.
- Correct one failure cleanup in imx6_pcie_host_init().

Main changes from v13 to v14 refer to Bjorn's comments:
- To keep suspend/resume symmetric as much as possible. Create
  imx6_pcie_stop_link() and imx6_pcie_host_exit() functions, and invoke
  them in suspend callback.
- Since the imx6_pcie_clk_disable() is invoked by imx6_pcie_host_exit(),
  to be symmetric with imx6_pcie_host_exit(), move imx6_pcie_clk_enable()
  to imx6_pcie_host_init() from imx6_pcie_deassert_core_reset().

Main changes from v12 to v13:
- Call imx6_pcie_host_init() instead of duplicating codes in resume.
- Move the regulator enable out of imx6_pcie_deassert_core_reset().
  Re-format the error handling in imx6_pcie_deassert_core_reset()
  and imx6_pcie_host_init() accordingly.

Main changes from v11 to v12 issued by Bjorn:
- Add four intro patches to move code around to collect similar things
  (PHY management, reset management) together.  This makes the first
  diff to collect clock enables simpler because it's not cluttered with
  unrelated things that didn't actually change.
- Factor out ref clock disables so the disable function structure matches
  the enable structure.
- Drop unused "ret" from "Reduce resume time ..." to avoid bisection
  hole, then add it back in "Do not hide phy driver ..." where we start
  using it again.
- Add patch to make imx6_pcie_clk_disable() symmetric with
  imx6_pcie_clk_enable() in terms of enable/disable ordering.

Main changes from v10 to v11:
No code changes, just do the following operations refer to Bjorn's comments.
  - Split #6 patch into two patches.
  - Rebase to v5.19-rc1 based on for-next branch of Shawn's git.

Main changes from v9 to v10:
- Add the "Reviewed-by: Lucas Stach <[e-mail]>" tag into #3
  and #4 patches.
- Refer to Bjorn's comments:
  - refine the commit of the first patch
  - keep alignment of the message format in the second patch
  - More specific commit and subject of the #5 and #7 patches.
- Move the regualtor_disable into suspend, turn off the regulator when bus
  is powered off and system in suspend mode.
- Let the driver probe successfully, return zero in imx6_pcie_start_link()
  when PCIe link is down. 
  In this link down scenario, only start the PCIe link training in resume
  when the link is up before system suspend to avoid the long latency in
  the link training period.
- Don't hide phy driver callbacks in core reset and clk_enable, and refine
  the error handling accordingly.
- Drop the #8 patch of v9 series, since the clocks and powers are not gated
  off anymore when link is down.

Main changes from v8 to v9:
- Don't change pcie-designware codes, and do the error exit process only in
  pci-imx6 driver internally.
- Move the phy driver callbacks to the proper places

Main changes from v7 to v8:
Regarding Bjorn's review comments.
- Align the format of the dev_info message and refine commit log of
  #6/7/8 patches.
- Rename the err_reset_phy label, since there is no PHY reset in the out

Main changes from v6 to v7:
- Keep the regulator usage counter balance in the #5 patch of v6 series.

Main changes from v5 to v6:
- Refer to the following discussion with Fabio, fix the dump by his patch.
  https://patchwork.kernel.org/project/linux-pci/patch/[e-mail]/
  Refine and rebase this patch-set after Fabio' dump fix patch is merged.
- Add one new #4 patch to disable i.MX6QDL REF clock too when disable clocks
- Split the regulator refine codes into one standalone patch #5 in this version.

Main changes from v4 to v5:
- Since i.MX8MM PCIe support had been merged. Based on Lorenzo's git repos,
  resend the patch-set after rebase.

Main changes from v3 to v4:
- Regarding Mark's comments, delete the regulator_is_enabled() check.
- Squash #3 and #6 of v3 patch into #5 patch of v4 set.

Main changes from v2 to v3:
- Add "Reviewed-by: Lucas Stach <[e-mail]>" tag into
  first two patches.
- Add a Fixes tag into #3 patch.
- Split the #4 of v2 to two patches, one is clock disable codes move,
  the other one is the acutal clock unbalance fix.
- Add a new host_exit() callback into dw_pcie_host_ops, then it could be
  invoked to handle the unbalance issue in the error handling after
  host_init() function when link is down.
- Add a new host_exit() callback for i.MX PCIe driver to handle this case
  in the error handling after host_init.

Main changes from v1 to v2:
Regarding Lucas' comments.
  - Move the placement of the new imx6_pcie_clk_enable() to avoid the
    forward declarition.
  - Seperate the second patch of v1 patch-set to three patches.
  - Use the module_param to replace the kernel command line.
Regarding Bjorn's comments:
  - Use the cover-letter for a multi-patch series.
  - Correct the subject line, and refine the commit logs. For example,
    remove the timestamp of the logs.

Bjorn Helgaas (5):
  PCI: imx6: Move imx6_pcie_grp_offset(), imx6_pcie_configure_type()
    earlier
  PCI: imx6: Move PHY management functions together
  PCI: imx6: Move imx6_pcie_enable_ref_clk() earlier
  PCI: imx6: Factor out ref clock disable to match enable
  PCI: imx6: Disable clocks in reverse order of enable

Richard Zhu (12):
  PCI: imx6: Move imx6_pcie_clk_disable() earlier
  PCI: imx6: Collect clock enables in imx6_pcie_clk_enable()
  PCI: imx6: Propagate .host_init() errors to caller
  PCI: imx6: Disable i.MX6QDL clock when disabling ref clocks
  PCI: imx6: Call host init function directly in resume
  PCI: imx6: Turn off regulator when system is in suspend mode
  PCI: imx6: Move regulator enable out of
    imx6_pcie_deassert_core_reset()
  PCI: imx6: Mark the link down as non-fatal error
  PCI: imx6: Reduce resume time by only starting link if it was up
    before suspend
  PCI: imx6: Do not hide phy driver callbacks and refine the error
    handling
  PCI: imx6: Move the imx6_pcie_ltssm_disable() earlier
  PCI: imx6: Reformat suspend callback to keep symmetric with resume

drivers/pci/controller/dwc/pci-imx6.c | 660 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++------------------------------------------------
1 file changed, 357 insertions(+), 303 deletions(-)

Comments

Bjorn Helgaas July 14, 2022, 5:10 p.m. UTC | #1

On Thu, Jul 14, 2022 at 03:30:52PM +0800, Richard Zhu wrote:
> This series patches refine pci-imx6 driver and do the following main changes.
> - Encapsulate the clock enable into one standalone function
> - Add the error propagation from host_init and resume
> - Turn off regulator when the system is in suspend mode
> - Let the probe successfully when link never comes up
> - Do not hide the phy driver callbacks in core reset and clk_enable.
> - Keep symmetric as much as possible between suspend and resume callbacks.
> BTW, this series are verified on i.MX8MM EVK board when one NVME is used.
> 
> Main changes from v14 to v15 refer to Lucas' comments and Bjorn's suggestions:
> Only the five patches listed below have changes.
> 10/17
> - Remove regulator_disable() from imx6_pcie_shutdown() and update the commit
>   log accordingly refer to Lucas' comments.
> 11/17
> - Move the regulator enable before the PHY init and core reset assert to
>   avoid introducing more failure cleanup paths refer to Lucas' comments.
> 14/17 has the codes conflictions.
> - Rebase the 14/17 patch because of the codes conflictions introduced by
>   previous 11/17 new changes.
> 16/17
> - Add the missing the IMX8MQ case and drop the default case in
>   imx6_pcie_ltssm_disable() refer to Lucas' comments.
> 17/17
> - Rebase the codes and resolve the codes conflictions introduced by
>   previous 11/17 new changes.
> - Correct one failure cleanup in imx6_pcie_host_init().
> 
> Main changes from v13 to v14 refer to Bjorn's comments:
> - To keep suspend/resume symmetric as much as possible. Create
>   imx6_pcie_stop_link() and imx6_pcie_host_exit() functions, and invoke
>   them in suspend callback.
> - Since the imx6_pcie_clk_disable() is invoked by imx6_pcie_host_exit(),
>   to be symmetric with imx6_pcie_host_exit(), move imx6_pcie_clk_enable()
>   to imx6_pcie_host_init() from imx6_pcie_deassert_core_reset().
> 
> Main changes from v12 to v13:
> - Call imx6_pcie_host_init() instead of duplicating codes in resume.
> - Move the regulator enable out of imx6_pcie_deassert_core_reset().
>   Re-format the error handling in imx6_pcie_deassert_core_reset()
>   and imx6_pcie_host_init() accordingly.
> 
> Main changes from v11 to v12 issued by Bjorn:
> - Add four intro patches to move code around to collect similar things
>   (PHY management, reset management) together.  This makes the first
>   diff to collect clock enables simpler because it's not cluttered with
>   unrelated things that didn't actually change.
> - Factor out ref clock disables so the disable function structure matches
>   the enable structure.
> - Drop unused "ret" from "Reduce resume time ..." to avoid bisection
>   hole, then add it back in "Do not hide phy driver ..." where we start
>   using it again.
> - Add patch to make imx6_pcie_clk_disable() symmetric with
>   imx6_pcie_clk_enable() in terms of enable/disable ordering.
> 
> Main changes from v10 to v11:
> No code changes, just do the following operations refer to Bjorn's comments.
>   - Split #6 patch into two patches.
>   - Rebase to v5.19-rc1 based on for-next branch of Shawn's git.
> 
> Main changes from v9 to v10:
> - Add the "Reviewed-by: Lucas Stach <[e-mail]>" tag into #3
>   and #4 patches.
> - Refer to Bjorn's comments:
>   - refine the commit of the first patch
>   - keep alignment of the message format in the second patch
>   - More specific commit and subject of the #5 and #7 patches.
> - Move the regualtor_disable into suspend, turn off the regulator when bus
>   is powered off and system in suspend mode.
> - Let the driver probe successfully, return zero in imx6_pcie_start_link()
>   when PCIe link is down. 
>   In this link down scenario, only start the PCIe link training in resume
>   when the link is up before system suspend to avoid the long latency in
>   the link training period.
> - Don't hide phy driver callbacks in core reset and clk_enable, and refine
>   the error handling accordingly.
> - Drop the #8 patch of v9 series, since the clocks and powers are not gated
>   off anymore when link is down.
> 
> Main changes from v8 to v9:
> - Don't change pcie-designware codes, and do the error exit process only in
>   pci-imx6 driver internally.
> - Move the phy driver callbacks to the proper places
> 
> Main changes from v7 to v8:
> Regarding Bjorn's review comments.
> - Align the format of the dev_info message and refine commit log of
>   #6/7/8 patches.
> - Rename the err_reset_phy label, since there is no PHY reset in the out
> 
> Main changes from v6 to v7:
> - Keep the regulator usage counter balance in the #5 patch of v6 series.
> 
> Main changes from v5 to v6:
> - Refer to the following discussion with Fabio, fix the dump by his patch.
>   https://patchwork.kernel.org/project/linux-pci/patch/[e-mail]/
>   Refine and rebase this patch-set after Fabio' dump fix patch is merged.
> - Add one new #4 patch to disable i.MX6QDL REF clock too when disable clocks
> - Split the regulator refine codes into one standalone patch #5 in this version.
> 
> Main changes from v4 to v5:
> - Since i.MX8MM PCIe support had been merged. Based on Lorenzo's git repos,
>   resend the patch-set after rebase.
> 
> Main changes from v3 to v4:
> - Regarding Mark's comments, delete the regulator_is_enabled() check.
> - Squash #3 and #6 of v3 patch into #5 patch of v4 set.
> 
> Main changes from v2 to v3:
> - Add "Reviewed-by: Lucas Stach <[e-mail]>" tag into
>   first two patches.
> - Add a Fixes tag into #3 patch.
> - Split the #4 of v2 to two patches, one is clock disable codes move,
>   the other one is the acutal clock unbalance fix.
> - Add a new host_exit() callback into dw_pcie_host_ops, then it could be
>   invoked to handle the unbalance issue in the error handling after
>   host_init() function when link is down.
> - Add a new host_exit() callback for i.MX PCIe driver to handle this case
>   in the error handling after host_init.
> 
> Main changes from v1 to v2:
> Regarding Lucas' comments.
>   - Move the placement of the new imx6_pcie_clk_enable() to avoid the
>     forward declarition.
>   - Seperate the second patch of v1 patch-set to three patches.
>   - Use the module_param to replace the kernel command line.
> Regarding Bjorn's comments:
>   - Use the cover-letter for a multi-patch series.
>   - Correct the subject line, and refine the commit logs. For example,
>     remove the timestamp of the logs.
> 
> Bjorn Helgaas (5):
>   PCI: imx6: Move imx6_pcie_grp_offset(), imx6_pcie_configure_type()
>     earlier
>   PCI: imx6: Move PHY management functions together
>   PCI: imx6: Move imx6_pcie_enable_ref_clk() earlier
>   PCI: imx6: Factor out ref clock disable to match enable
>   PCI: imx6: Disable clocks in reverse order of enable
> 
> Richard Zhu (12):
>   PCI: imx6: Move imx6_pcie_clk_disable() earlier
>   PCI: imx6: Collect clock enables in imx6_pcie_clk_enable()
>   PCI: imx6: Propagate .host_init() errors to caller
>   PCI: imx6: Disable i.MX6QDL clock when disabling ref clocks
>   PCI: imx6: Call host init function directly in resume
>   PCI: imx6: Turn off regulator when system is in suspend mode
>   PCI: imx6: Move regulator enable out of
>     imx6_pcie_deassert_core_reset()
>   PCI: imx6: Mark the link down as non-fatal error
>   PCI: imx6: Reduce resume time by only starting link if it was up
>     before suspend
>   PCI: imx6: Do not hide phy driver callbacks and refine the error
>     handling
>   PCI: imx6: Move the imx6_pcie_ltssm_disable() earlier
>   PCI: imx6: Reformat suspend callback to keep symmetric with resume
> 
> drivers/pci/controller/dwc/pci-imx6.c | 660 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++------------------------------------------------
> 1 file changed, 357 insertions(+), 303 deletions(-)

Applied to pci/ctrl/imx6 for v5.20, thanks!